<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Whitespectre on Medium]]></title>
        <description><![CDATA[Stories by Whitespectre on Medium]]></description>
        <link>https://medium.com/@whitespectre?source=rss-a894252ac1d1------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Ple8HWsNye1jr3MMXzKRAQ.png</url>
            <title>Stories by Whitespectre on Medium</title>
            <link>https://medium.com/@whitespectre?source=rss-a894252ac1d1------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 05:26:41 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@whitespectre/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Working Demo, So What? The Reality Gap Between AI Prototypes and Production]]></title>
            <link>https://medium.com/whitespectre/working-demo-so-what-the-reality-gap-between-ai-prototypes-and-production-ea5e2c181d79?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/ea5e2c181d79</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[task-management]]></category>
            <category><![CDATA[collaboration]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[jira]]></category>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Tue, 08 Apr 2025 14:46:19 GMT</pubDate>
            <atom:updated>2025-04-08T14:46:19.347Z</atom:updated>
            <content:encoded><![CDATA[<p>How rapid prototyping enabled by AI tools is transforming product development — and creating new challenges in the journey from proof-of-concept to production.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3S0v6QIyJWXR-n_yTkGygw.png" /></figure><p>Four hours versus three weeks.</p><p>That’s the stark difference we experienced at Whitespectre when comparing an AI-assisted prototyping approach to traditional development processes for experimental ideas. The key insight: projects that would have been impossible to prioritize due to resource constraints can now be explored through rapid prototyping.</p><p>Before AI-assisted tools, many innovative ideas remained unexplored simply because allocating developers, designers, and product managers to a speculative internal tool couldn’t be justified. Now, product managers armed with tools like Cursor and ChatGPT can independently create functional prototypes to validate concepts without consuming significant team resources.</p><p>But a critical question emerges: what happens after you have that working demo? Does functional mean production-ready? We’ve learned there’s a significant — and often invisible — gap between prototype and production-grade code.</p><p>This article explores our experience with the Jira Evaluator tool and reveals practical insights about the changing nature of product development in an era where anyone can build a working prototype.</p><h3>The Promise of AI-Powered Product Development</h3><p>Our Jira Evaluator project began with a clear challenge: to extract meaningful metrics and analysis from our Jira data, we needed to standardize our Jira cards. But this presented a classic internal tooling dilemma — how to justify dedicating resources to an experimental project with uncertain ROI?</p><p>While we consistently foster experiments like this as part of our Whitespectre Labs internal product incubator, this kind of speculative tool would likely never get prioritized in the traditional product development model. The resource investment would be too substantial: assessing technical feasibility, creating wireframes, conducting stakeholder discussions, developing clickable prototypes, and finally allocating precious engineering resources. For resource-stretched teams, these kinds of experimental tools often remain just ideas.</p><p>AI-assisted development tools changed this equation completely. Using v0, Cursor, and ChatGPT, a single product manager built a working prototype in approximately four hours — unlocking a project that might not have been explored otherwise.</p><p>This represents a fundamental shift in what’s possible for internal innovation. Rather than having to make difficult tradeoffs about resource allocation, we could validate the concept with minimal investment. If the experiment proved valuable, we could then make an informed decision about further development — and if not, we’d only invested a few hours rather than weeks of cross-functional time.</p><h3>The Challenges</h3><p>Despite the incredible speed and autonomy, our product manager encountered several challenges (some anticipated) in the prototype development process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_VACDY9ZuJwMTPofM0ifFA.png" /><figcaption>Effective communication: Our product manager’s Cursor prompt as part of demo creation.</figcaption></figure><h4>Constraint-Setting, Not Just Goal-Setting</h4><p>Effective communication with AI tools requires not just articulating what you want, but explicitly describing what you don’t want to change. This mirrors an essential prompting principle: constraint-setting is as crucial as goal-setting.</p><p>Without clear boundaries, AI will make assumptions and introduce unexpected logic or design choices. The prototype development process taught us to be methodical about establishing constraints — a skill that product managers already possess from writing detailed requirements, but which takes on new importance when working with AI tools.</p><h4>The Troubleshooting Wall</h4><p>Perhaps the most frustrating limitation was hitting the wall of technical knowledge when things broke. While AI tools could generate code quickly, they weren’t effective at helping troubleshoot complex technical issues.</p><p>Particularly challenging were third-party integrations like API connections, where the AI lacked sufficient context or up-to-date knowledge to resolve problems. This points to a critical limitation: even with powerful AI tools, technical expertise remains necessary, especially for integration work and debugging complex issues.</p><p>Dealing with these issues meant getting a developer involved.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/687/1*MUilDNaO7E5JJ9vJWFqn7Q.png" /><figcaption>Endless tweaking: One of several different UIs tried during the prototyping process.</figcaption></figure><h4>The Rabbit Hole of Endless Tweaking</h4><p>Having complete control over every detail created another unexpected challenge — the rabbit hole of endless tweaking. With the power to change anything at a moment’s notice, it became tempting to continuously refine minor details simply because it was possible.</p><p>The prototype development process revealed how easy it is to lose sight of the bigger product vision while obsessing over refinements that don’t matter at the prototype stage. The flexibility that makes AI tools powerful also makes them dangerous time sinks without proper discipline.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RBWF8fNYOWuqKC_44aPQHw.png" /><figcaption>Challenging assumptions: An early version featured scoring removed after team feedback.</figcaption></figure><h4>Missing the Valuable Friction</h4><p>Perhaps most interestingly, we noticed the absence of valuable team pushback during the solo development process. The traditional product development cycle incorporates multiple perspectives, challenges assumptions, and catches edge cases through collaborative discussion.</p><p>When building with AI, this friction disappears. There’s no one to question approaches, suggest alternatives, or identify blind spots. This highlights a fascinating paradox: while AI tools accelerate initial development, they temporarily remove the cross-functional collaboration that ultimately strengthens the product.</p><p>The prototype did, however, provide a concrete artifact for subsequent discussions, which proved more effective than attempting to align teams on abstract concepts.</p><h3>The Technical Reality Check</h3><p>Moving from prototype to production revealed several technical realities about AI-generated code. When our development team reviewed the Jira evaluator prototype, three issues immediately stood out: accumulated dead code from iterative prompting, security vulnerabilities like hardcoded API keys, and significant code duplication.</p><p>These issues reflect how AI tools approach development — optimizing for making things work quickly rather than architectural elegance. Each time the prototype was modified through new prompts, the AI would add functionality without properly cleaning up previous iterations. The result was a codebase that functioned but contained numerous remnants of its development journey.</p><p>Despite these challenges, approximately 30% of the prototype code proved salvageable. The core functionality and interface design translated well to production, while the implementation details needed significant refinement. This ratio seems typical for AI-generated prototypes — the conceptual work and basic structure provide value, but the technical implementation requires professional refinement.</p><p>For browser-based tools like our Jira evaluator, production-readiness concerns centered on three key areas:</p><p><strong>1. Code simplification: </strong>Removing duplicated functionality and dead code paths that accumulated during iterative development</p><p><strong>2. Security improvements:</strong> Eliminating hardcoded credentials and implementing proper configuration management</p><p><strong>3. Privacy considerations:</strong> Auditing logs and data handling to ensure sensitive information wasn’t being captured inappropriately</p><p>Interestingly, we found that using more structured prompting techniques with tools like Cursor significantly improved code quality. Rules and constraints specifically tailored to programming languages helped guide the AI toward better practices, though keeping an expert in the loop is still a must to reach a scalable production standard.</p><p>For straightforward applications like browser extensions, the architectural limitations of AI-generated code were less problematic. However, for more complex systems, the architectural decisions made during rapid prototyping would likely require more substantial revision before production deployment.</p><h3>Unexpected Benefits and Insights</h3><p>Despite the challenges, the AI-accelerated prototype development process yielded several unexpected benefits and insights.</p><h4>Democratized Experimentation</h4><p>The most significant organizational impact has been the democratization of technical experimentation. Team members with product knowledge but limited development skills can now create functional prototypes that would have been impossible before.</p><p>This has fundamentally changed expectations around idea validation. Previously, proposing a new concept might lead to a series of discussions about whether it merited investment. Now, the appropriate response is “spend a few hours on it and come back with something we can play with.”</p><p>This shift is particularly transformative for product managers, who gain a powerful new tool for ideation and stakeholder discussions. The ability to quickly materialize concepts removes significant barriers to innovation, especially for internal tools that might otherwise never receive development resources.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8kQkUoyN9JzGXS7RXOn4bw.jpeg" /><figcaption>Intuitive interface: The UX while the AI is re-analizing the card following some tweaks</figcaption></figure><h4>User-Centric Technical Approaches</h4><p>An interesting benefit emerged from building without a deep technical background: approaching problems from a user perspective rather than a developer perspective.</p><p>When creating interfaces or solving technical challenges, product managers naturally gravitate toward intuitive user-focused solutions rather than conventional technical implementations. This leads to designs that can be more aligned with user needs, highlighting how diverse perspectives enhance product development.</p><p>One example from our JIRA Evaluator tool was designing the settings page. Rather than using environment variables or configuration files (the typical developer approach), the prototype featured a simple settings page with a form field — immediately more intuitive for non-technical users. This approach gave users the flexibility to use their own OpenAI accounts, addressing data privacy concerns and putting them in control of their API processing and usage.</p><h4>The Rising Value of Requirement Definition Skills</h4><p>An unexpected insight from our experience was how product management skills in requirement definition translate directly to effective AI interaction. The ability to articulate clear, structured requirements — long valued in traditional development — has essentially become a new form of programming.</p><p>Working with AI tools requires the same precision in describing desired outcomes and constraints that makes for good Jira tickets or user stories. This represents a significant advantage for product managers and others skilled in breaking down complex needs into clear, actionable specifications.</p><p>“Requirement writing as a new kind of programming language” became our shorthand for this phenomenon. Those with experience in structured thinking and clear problem framing found themselves naturally equipped to direct AI tools effectively. This suggests that as AI democratizes technical creation, the differentiating factor shifts from coding ability to the clarity of thought and communication — a fascinating inversion of traditional technical hierarchies.</p><h4>Enhanced Communication</h4><p>The ability to quickly produce a functional prototype dramatically improved communication effectiveness. Abstract ideas transformed into tangible interfaces that stakeholders could interact with and provide feedback on.</p><p>This improved communication extends to all parties — from technical teams to business leaders — allowing everyone to engage with a concrete implementation rather than abstract descriptions. For internal tools especially, this accelerates the feedback loop and increases stakeholder buy-in.</p><h3>The New Collaborative Process</h3><p>The AI-accelerated development process doesn’t eliminate traditional roles but transforms how they interact and when they engage in the process.</p><h4>Shifting Collaboration Patterns</h4><p>The AI-accelerated prototype transforms traditional collaboration by providing a more tangible starting point. While our established planning process has always been effective for standard development, these rapid prototypes offer a better approach for experimental concepts. Team discussions benefit from having a working artifact everyone can interact with, making abstract ideas immediately concrete.</p><p>This represents an additional tool in our collaboration model. For certain types of projects — particularly exploratory ones — teams can evaluate a functioning prototype before committing significant resources. This approach doesn’t replace thoughtful planning but adds another option to our toolkit and helps give us better estimates when it comes to building the real thing.</p><p>Importantly, these prototypes serve primarily as proof-of-concept demonstrations. Rather than trying to refactor prototype code into production systems, the team often rebuilds from scratch using the validated concept as a guide. The prototype’s value lies in validating the idea and creating a shared reference point that accelerates decision-making, not in providing production-ready code.</p><h4>When to Bring in Specialists</h4><p>A critical skill in this new paradigm is knowing when to involve specialists. The autonomy of AI-assisted development can create a false sense of self-sufficiency. Working in isolation without pushback can make developers or product managers feel like they’re always right — a dangerous trap.</p><p>The prototype development experience taught us to establish clear stopping points and recognize when to seek feedback. Self-discipline becomes crucial: knowing when a prototype has reached sufficient maturity to benefit from broader input, and when continuing solo development risks breaking functional elements or going down unproductive paths.</p><p>This highlights a crucial insight: the prototype should be a starting point for collaboration, not a replacement for it.</p><h4>Impact on Project Timelines</h4><p>The acceleration of the initial prototyping phase fundamentally changes project economics. Ideas that would never have been explored due to resource constraints can now be quickly validated or rejected.</p><p>Previously, the decision to create a prototype would involve careful planning: would we learn enough to justify the development cost? Now, with prototype development taking hours instead of weeks, the calculus has shifted. A few hours of exploration is almost always worth the potential insight gained.</p><p>This dramatically lowers the barrier to experimentation and allows teams to explore more ideas with less investment — particularly valuable for internal tools, proofs of concept, and speculative features.</p><h3>Practical Takeaways for Teams</h3><p>Based on our experience, here are key takeaways for different team members exploring AI-accelerated development:</p><h4>For Product Managers:</h4><ul><li><strong>Start with the right problem: </strong>AI tools can accelerate validation but won’t fix a misidentified problem. Before diving into prototyping, ensure you’re solving a genuine user or business need. The most elegant solution to the wrong problem remains worthless.</li><li><strong>Set clear boundaries for your prototype: </strong>Define the scope and establish clear stopping points to avoid the endless tweaking trap. Know when your prototype has achieved its goal of demonstrating the concept.</li><li><strong>Recognize the limits of technical knowledge: </strong>Be prepared to hit troubleshooting walls and know when to involve technical experts, particularly for third-party integrations and complex bugs.</li><li><strong>Practice constraint-setting in your prompts:</strong> Clearly define not just what you want the AI to do, but what you don’t want it to change. This reduces unexpected behavior and improves output quality.</li><li><strong>Balance self-sufficiency with collaboration:</strong> While AI tools enable independent exploration, they can create a false sense of completeness. Prioritize diverse perspectives early in the process to strengthen concepts and identify blind spots before they become embedded in the solution.</li></ul><h4>For Developers:</h4><ul><li><strong>Re-evaluate the feature set: </strong>When transitioning from a prototype to production, start by simplifying. AI tools make it easy to implement numerous features quickly, but this can lead to an unstable foundation. Strip back to core functionality before building a more robust version.</li><li><strong>Assess what can be salvaged:</strong> Not all prototype code needs to be discarded. Evaluate which components can serve as a foundation for the production version. The quality of AI-generated code varies significantly based on the prompts and tools used — prototypes created with tools like Cursor from the start tend to produce more usable code when combined with right models and prompts.</li><li><strong>Implement proper version control: </strong>One of the biggest risks with AI-assisted development is the temptation to continuously iterate without structure. Establish version control using Git early and adhere to commit discipline rigorously. For features beyond the initial prototype, follow the standard development workflow of planning, validation, estimation, and implementation rather than continuing to build directly on the prototype.</li><li><strong>Conduct security audits:</strong> AI-generated code requires thorough security review. While functional, it may contain vulnerabilities or implement outdated security practices. Always perform a comprehensive security assessment before moving prototype code toward production.</li><li><strong>Align with team standards:</strong> When receiving a prototype from a product manager, assess how it aligns with your team’s technical standards. Converting a working but non-standard prototype into production code that fits your architecture and coding standards should be approached methodically rather than piecemeal.</li></ul><h4>For Organizations:</h4><ul><li><strong>Lower the barrier to experimentation:</strong> Embrace the reduced cost of idea validation to explore more potential innovations, particularly for internal tools that might never otherwise get prioritized.</li><li><strong>Maintain cross-functional collaboration: </strong>Don’t let the ability to create prototypes individually reduce valuable team interaction. The feedback process remains essential for quality outcomes.</li><li><strong>Adjust evaluation thresholds:</strong> When someone proposes an idea, the appropriate response is now “spend a few hours on it and come back with something we can play with.” The cost/benefit analysis for prototyping has fundamentally changed.</li><li><strong>Create clear prototype-to-production pathways:</strong> Establish structured processes for transitioning validated prototypes into production projects. This includes criteria for assessment, required documentation, and guidelines for when to rebuild versus refactor prototype code.</li><li><strong>Recognize when prototypes are sufficient: </strong>Some internal tools and experiments may be valuable even without production-level polish. Identify when “good enough” is actually good enough for the intended purpose.</li></ul><h3>Conclusion</h3><p>AI-powered development tools have fundamentally changed how teams can approach product development — enabling rapid experimentation, lowering barriers to technical creation, and transforming collaboration patterns.</p><p>However, our experience with the Jira Evaluator reinforces that human expertise remains essential. The gap between a working prototype and production-ready code is real, even if it’s sometimes invisible to non-technical eyes.</p><p>The most effective approach blends the speed and autonomy of AI-assisted prototyping with the expertise and collaboration of cross-functional teams. Neither approach alone delivers optimal results — it’s the thoughtful integration of both that leads to successful products.</p><p>As we continue to explore this new territory, one thing is clear: the question isn’t whether to use AI tools or follow traditional development processes. It’s how to combine the strengths of both to create better products, faster, without sacrificing quality.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ea5e2c181d79" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/working-demo-so-what-the-reality-gap-between-ai-prototypes-and-production-ea5e2c181d79">Working Demo, So What? The Reality Gap Between AI Prototypes and Production</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Future-Proof Your Codebase: Early Technical Decisions That Scale]]></title>
            <link>https://medium.com/whitespectre/future-proof-your-codebase-early-technical-decisions-that-scale-273d0a023624?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/273d0a023624</guid>
            <category><![CDATA[scalability]]></category>
            <category><![CDATA[code-quality]]></category>
            <category><![CDATA[clean-code]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[automation]]></category>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Mon, 24 Mar 2025 16:48:01 GMT</pubDate>
            <atom:updated>2025-03-24T16:58:55.071Z</atom:updated>
            <content:encoded><![CDATA[<p>The wrong technical foundation becomes a bottleneck when you need to scale. From code styling to architecture choices, learn actionable strategies to give your project the best start.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4kb_YO9SBLSISKYpxo9Jhw.jpeg" /></figure><p>Any software project that’s easy to maintain and extend is built on the right code, written in the right way, at the right time. Poor (or a lack of) technical decisions early in a project’s lifecycle can and will snowball into bottlenecks, slowing development speed, and technical debt.</p><p>As “Broken Windows Theory” states, when minor signs of neglect appear, they invite further degradation. The same applies to inconsistent code and rushed architecture which gradually erode a project’s health over time.</p><p>So how do you avoid that degradation? Should you enforce a rigid structure from day one? How do you make the “correct” decisions when every project has unique constraints and requirements?</p><p>The truth is, there’s no silver bullet. Over-engineering upfront can slow progress, while not planning at all sets the stage for chaos.</p><p>In this article, we’ll explore the technical decisions that matter most when launching a project — and why they matter. Drawing from our experience building and rescuing software products, we’ll break down actionable strategies to help you start on the right foot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*YMimfzz60VEeM11i" /></figure><h3>Code Styling and Consistency — Maintaining a Clean Code</h3><h4>Why It Matters</h4><p>Code quality isn’t just about nitpicking syntax or following a style guide. It’s about reducing the chance of bugs, improving maintainability, and establishing a shared language across the team.</p><p>When code is consistent, onboarding new team members is smoother, pull request reviews are more effective, and errors become easier to spot and fix.</p><p>Over time, maintaining a consistent codebase will boost your team’s speed, allowing them to focus on delivering features, not arguing over code style in PRs, or trying to figure out messy code.</p><h4>How to Get Started</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*XxDCBe90xmcWEmdm" /><figcaption>Messy code (poorly formatted, inconsistent, hard to read)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*ySl57wZK-4KrJW9n" /><figcaption>Clean code (formatted with ESLint and Prettier, readable and maintainable)</figcaption></figure><p><strong>Linting and Formatting</strong></p><p>Use tools like ESLint and Prettier for JavaScript/TypeScript code to automatically flag potential issues and enforce style rules.</p><p>For consistency, consider adopting or referencing a recognized style guide, such as Airbnb’s JavaScript Style Guide. And make sure to configure them accordingly, for example by using <em>.prettierrc</em> or <em>.eslintrc</em> files that reference the chosen style guide.</p><p><strong>CommitLint</strong></p><p>Ensures your commit messages follow a standard format (e.g., Conventional Commits). This helps with automated release notes and code history, making it easier to search for when something was implemented, fixed, or introduced.</p><p><strong>Take Advantage of AI Tools</strong></p><p>AI-powered tools can significantly boost productivity and help maintain a cleaner, more consistent codebase. Tools like Cursor and GitHub Copilot can provide suggestions and approaches, helping your team to write maintainable code from the start.</p><p>Other solutions like CodeScene and DeepCode can scan your code for potential errors, code smells, and security vulnerabilities when opening a PR.</p><h3>Define Your Architecture</h3><h4>Why It Matters</h4><p>As a codebase grows, files and modules can easily become disorganized, increasing cognitive load and extending the time needed to implement new fixes or features. That’s why it’s important to define an architecture from the start.</p><p>It doesn’t have to be rigid from the start, things will inevitably change, but a well-structured approach ensures you can modify or extend specific features without affecting unrelated parts of the system.</p><p>Your goal here is to ensure changes can be made easily when needed (for example, integrating or swapping a third-party service). This helps reduce cognitive load, making it easier for developers to add or modify functionality without unintended side effects.</p><h4>How to Get Started</h4><p><strong>Monorepo vs. Multiple Repositories</strong></p><p>Decide early whether you’ll keep your entire codebase in a single monorepo (using a workspace manager like Yarn or PNPM) or split it into separate repositories for each service or application.</p><p>A monorepo can simplify dependency management, code sharing, and the use of consistent tooling, but it may grow large and require more sophisticated partial build workflows. From our own experience, if your applications share a significant amount of logic and your developers are proficient in one language, a monorepo approach (like using T3 stack) can be a good choice.</p><p>On the other hand, multiple repositories allow each project to have its own lifecycle and technology stack, but shared libraries may need to be published or referenced in a separate way. Choose the approach that best fits your team’s workflow and strengths.</p><p><strong>Define a Clear Structure</strong></p><p>It’s essential to define a clear separation of concerns within your codebase, there is no silver bullet here. The important part is having a well-defined structure (feature-based for example) that suits your current context and can evolve as the product or team grows.</p><p><strong>Keep It Doing Just One Thing</strong></p><p>Resist the urge to plan for every possible future scenario. Focus on immediate needs, but structure code so it can be extended safely.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wPpdr_jx9etk2IFOPQom5w.png" /></figure><h3>Internationalization Early On</h3><h4>Why It Matters</h4><p>Internationalization from the start is significantly easier than implementing it later. Adding it to i18n later can involve substantial refactoring and relabeling of strings. By planning for multiple locales early, you ensure consistent and scalable solutions for text and other localized content, saving both time and effort in the long run.</p><h4>How to Get Started</h4><p>Start simple by using established libraries like i18next for managing translations. VSCode plugins like i18n Ally can help you quickly search, translate, and manage localized strings efficiently.</p><h3>Theming &amp; Typography Structure</h3><h4>Why It Matters</h4><p>Consistent styling across an application not only creates a cohesive user experience but also reduces “change amplification.”</p><p>If you need to update a color or font (and probably that might happen in the beginning of your project), or even implement light and dark modes, you can do so in one place rather than searching and implementing new logic and conditions on the entire codebase.</p><h4>How to Get Started</h4><p><strong>Basic Theme/Typography Structure</strong></p><p>You don’t need to start with a fancy or complex setup. Begin by defining some key aspects of your theme in a centralized configuration file. This file might include:</p><ul><li>Colors: Primary, secondary, accent colors, and shades for backgrounds, texts, and borders.</li><li>Typography: Font families, sizes, line heights, and weights for headings, body text, and other elements.</li><li>Spacing: Standardized margins and paddings to ensure consistent layout across components.</li></ul><p><strong>Use Popular Libraries</strong></p><p>There are several libraries that can help you structure and implement theme and typography systems. For React/Next.js, you can use Shadcn, Chakra UI, or Tailwind CSS. For React Native, options like Tamagui, Restyle, and Unistyles also are an excellent choice depending on your scenario.</p><p>By organizing these elements in one place, you create a single source of truth for design decisions and appearance.</p><p>Restyle: A TypeScript library for building themable UI components in React Native, promoting consistency with minimal custom styles.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*_-GgBMr6_7oVcCOTB92ZkA.jpeg" /><figcaption>Restyle: A TypeScript library for building themable UI components in React Native, promoting consistency with minimal custom styles.</figcaption></figure><h3>Reusable Components</h3><h4>Why It Matters</h4><p>Think of it as theming and typography, but for components. By centralizing key UI elements, you can maximize code reuse and encapsulate recurring patterns (such as buttons, modals, and cards).</p><p>This ensures consistency across the project, reduces duplication, and makes changes easier to implement, accelerating both development and shipping.</p><h4>How to Get Started</h4><p><strong>Component Library</strong></p><p>Begin with a small library of the most frequently used UI elements — cards, buttons, inputs, etc. Expand as needed, and align these components with your theme and typography.</p><h3>Automation and Continuous Integration</h3><h4>Why It Matters</h4><p>Automating builds, tests, and deployments, reduces manual effort and minimizes human error, so the developers can focus on delivering value.</p><h4>How to Get Started</h4><p><strong>Choose a Simple CI Platform</strong><br>Tools like GitHub Actions, Bitrise, or EAS (for React Native) offer ready-to-use workflows.</p><h3>Test the Right Things</h3><h4>Why It Matters</h4><p>A good testing strategy builds confidence in your codebase and allows your team to move faster without constantly worrying about regressions.</p><h4>How to Get Started</h4><p>Identify Critical and Happy Paths: Pinpoint the flows that directly impact your business or user experience, like authentication, payment, the core values or your system, while also ensuring that the most common and simplest usage paths, the “happy paths”, are covered, and monitored for performance.</p><p>Don’t chase 100% coverage if it doesn’t provide real value, as the project matures you and your team can gradually expand test coverage.</p><h3>Team and Process</h3><h4>Why It Matters</h4><p>A well-structured team keeps development organized and transparent, helping teams move quickly without confusion. A defined process ensures that each phase of development from planning to deployment is both consistent and predictable.</p><h4>How to Get Started</h4><p><strong>Branching Strategy</strong><br>Pick a branching model that suits your team’s size and release schedule. Gitflow is popular for larger teams with distinct release phases, while Trunk-Based Development can be more agile for smaller teams.</p><p><strong>Clear Environments</strong><br>Set up distinct environments (e.g., dev, staging, production) so changes can be tested in isolation before going live.</p><p><strong>Regular Release Cycle</strong><br>Decide how often you want to release (weekly, bi-weekly, continuous) and stick to that cadence. This predictability helps QA, product managers, and stakeholders plan accordingly.</p><p><strong>Define Roles &amp; Responsibilities</strong><br>Make it clear who merges pull requests, who is responsible for final approval, and how hotfixes or urgent changes are handled.</p><p><strong>Code Reviews</strong><br>Set a process for peer reviews, even if your team is small. Since code styling is already handled by ESLint, the peer review can focus on readability, adherence to architectural principles, and overall code quality.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KLadccwGPOjncjnUkR-Ipw.jpeg" /></figure><h3>Conclusion</h3><p>The choices we make at the beginning of a project will have a significant impact on its success. Prioritizing a strong foundation allows your project to adapt and grow predictably.</p><p>Remember, you don’t need to do it all at once, and it doesn’t need to be rigid. Choose one or two improvements that resonate most with your project’s immediate needs. Over time, refine, iterate, and evolve. A well-structured, thoughtfully planned codebase not only leads to happier engineers but also creates a path for your product’s long-term success.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=273d0a023624" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/future-proof-your-codebase-early-technical-decisions-that-scale-273d0a023624">Future-Proof Your Codebase: Early Technical Decisions That Scale</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[5 Steps for Successful AI Development Tool Adoption]]></title>
            <link>https://medium.com/whitespectre/5-steps-for-successful-ai-development-tool-adoption-1db4006cd5d6?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/1db4006cd5d6</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Mon, 24 Mar 2025 16:40:02 GMT</pubDate>
            <atom:updated>2025-03-24T16:40:02.429Z</atom:updated>
            <content:encoded><![CDATA[<p>Getting the best from AI dev tools requires a systematic, methodical approach to adoption rather than rushing into widespread implementation. We share how to win buy-in from developers with our framework, based on findings from our internal AI Development Tools initiative.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*irNt6lvY5wjC2sRH2QA84A.jpeg" /></figure><p>While an estimated 97% of developers now use AI tools individually, organizations struggle with systematic adoption that enables knowledge sharing, security compliance, and measurable improvement across teams.</p><p>Since starting our structured AI adoption journey in early 2023, we’ve developed and refined a framework for AI dev tools adoption that addresses the common pitfalls organizations face through collaboration and open communication.</p><p>Our AI Developer Tools (AIDT) initiative brings together engineers from different teams, and at different stages of adoption, to share tips, challenges, and time-savers, as part of our wider R&amp;D efforts with <a href="https://www.whitespectre.com/labs/">Whitespectre Labs</a>.</p><p>By encouraging collaboration and experimentation, and by tracking and assessing results, we’re able to identify which tools we want to roll-out across the entire team — and then roll them out successfully.</p><p>In this article, you’ll learn how we identified our favourite AI development tool — Cursor — and rolled it out across our whole team.</p><h3>Common pitfalls for successful tool adoption</h3><p>Before we get into the steps for successful adoption, it’s worth outlining some of the common pitfalls we’ve seen when trying to identify, validate, and roll-out new tools for development teams:</p><p><strong>Ad-hoc individual adoption without systematic approach to knowledge sharing:</strong> AI adoption doesn’t stop at getting individuals to experiment with the tools; it’s about testing a variety of tools in real-life scenarios and sharing what works (and what doesn’t). That way you can create processes to identify the best tools and roll them out across your whole team, rather than people working ad-hoc and in siloes.</p><p><strong>Security oversights:</strong> You need to be cautious when it comes to your team using AI without a systematic approach or any oversight because of the very real concerns around security. You need to make sure the tools everybody’s using to save time aren’t coming at a cost to you or your clients’ data privacy. This includes being vigilant about potential security vulnerabilities that might be introduced by AI-generated code.</p><p><strong>Shallow adoption vs true mastery:</strong> Without a way to share best practices, most engineers end up not knowing how to use AI development tools to their fullest extent, only using auto completion features without taking full advantage of more complex tools in their workflows.</p><p><strong>Blind trust in AI-generated solutions:</strong> Teams can fall into the trap of implementing AI suggestions without proper verification, leading to problems like non-existent API calls, deprecated methods, or security vulnerabilities. It’s crucial to use AI tools that can index official documentation and still double-check the results afterward.</p><p><strong>Over-engineering simple solutions: </strong>AI can sometimes introduce unnecessary complexity into straightforward tasks, resulting in:</p><ul><li>Overcomplication of simple requirements</li><li>Code entanglement and hard-to-track dependencies</li><li>Obfuscated logic that becomes difficult to maintain</li></ul><p>The key is to keep AI contributions simple, small, and aligned with project requirements. Complexity should be intentional, not accidental.</p><p><strong>Debugging challenges from excessive iteration:</strong> When teams rely too heavily on AI for debugging, they can encounter:</p><ul><li>False positives/negatives that waste time on non-existent issues</li><li>Misleading suggestions based on limited context</li><li>Code that becomes increasingly difficult to debug due to multiple AI-driven changes</li></ul><p>With those pitfalls in mind, let’s look at how to make AI dev tool adoption a success.</p><h3>A 5-step framework for enterprise-wide AI tool adoption</h3><p>To successfully roll out a new AI development tool you need to follow these five steps:</p><ul><li>Make sure it’s the right change (and prove it)</li><li>Get a core group of impacted people on board</li><li>Bring in the wider group, explain the why, the how and have others who have experienced it demonstrate benefits</li><li>Provide channels for peer-to-peer feedback, support and coaching</li><li>Track adoption and impact</li></ul><p>Here’s how:</p><h3>1. Make sure it’s the right change (and prove it)</h3><p>The first step to successful AI tool adoption is to validate that your chosen tools are the right ones.</p><p>Developers and engineers can be justifiably sceptical about changes to their established workflows. The last thing you want to do is roll out the wrong tool and then replace it with something else a few months later — that’s a great way to squander good will and erode buy in.</p><p>Starting with a few experienced personnel, the core members of our AIDT team validated the usefulness of Cursor alongside other tools over many months and measured the quantifiable impact on quick coding tasks, as well as getting up to speed with new projects and coding languages. By the end, we had the data to back up our choice.</p><p>For our pilot group, we started with a daily survey sent over Slack which asked team members:</p><ul><li>What AI tools did you use today?</li><li>On a scale of -3 to +3, how much time did it save you?</li><li>And can you tell us about your experience?</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*h7-WfvAFJyN-EZ1y" /></figure><p>These quick check-ins were key to understanding how AI worked in real workflows and day-to-day activities, to help the team work out where they were losing time on the tools and where it was being gained.</p><p>To support adoption, we had the group meet once a week to share tips, tricks, and observations. This surfaced both benefits, like quickly identifying code areas impacted by a feature, and drawbacks, like the need for precise prompts to avoid errant results.</p><p>The survey showed that even skeptics progressed along the adoption curve and came to see the value of the tools for certain tasks and, just as importantly, discovered where using them would be a potential timesink (where fixing the AI code would take longer than writing the code from scratch).</p><p>Crucially, this first phase also ensured there wasn’t any data security risk for our clients with Cursor, as it only saves telemetry and data usage, and doesn’t store code.</p><h3>2. Get a core group of impacted people on board</h3><p>Once we’d established the usefulness of Cursor with a few key AIDT project members, the next phase involved bringing together a larger group of senior engineers to test the tool more widely in real-life scenarios and further validate it before a complete roll out.</p><p>The people selected came with varying levels of buy-in and experience with AI, and were well-placed to share best practices — what worked and what didn’t. Through knowledge sharing via regular meetings and a dedicated Slack channel, we reached a point where everybody on that team was an internal evangelist for Cursor.</p><p>We continued the regular check ins and daily surveys that were part of the previous step, only now with a larger group. Across weekly meetings, they quickly uncovered the strengths and weaknesses of specific tools. This helped us understand where they might save keystrokes and create new code faster — as well as which areas showed little to no time gained (or additional frustration!).</p><p>90% of our engineers saw a significant increase in productivity within 2 weeks of using AI-driven developer tools. For the last two years, we’ve been doing our own experimentation, piloting various AI tools to keep raising the bar on our work.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*GQ63HbcQRbGOKsdU" /></figure><h3>3. Bring in the wider group, explain the why, the how and have others who have experienced it demonstrate benefits</h3><p>Now that we had the data to back up our choice and the confidence we’d selected the right tool, we could begin rolling Cursor out for the whole dev team. To ease the transition, we positioned Cursor as a familiar tool — like VS Code, used by the entire team — but enhanced with AI features.</p><p>It’s important any change like this isn’t just top down. Rolling out Cursor, the first thing we did was have a call with the whole dev team where a senior dev from the AIDT project demoed the tool and used real-life examples (from client-facing work) to demonstrate value. It’s important to get somebody credible — experienced and senior — to lead this call.</p><p>They were able to show how they use the Cursor day-to-day and what tangible benefits the rest of the team could expect, while also setting realistic expectations for the thought and skill needed to extract the most value from the tool.</p><p>This involved a live example where the senior dev showed how Cursor could be deployed to make changes to a legacy codebase, integrate those changes within a system you don’t have knowledge of, and then work with technology that you’re not proficient in. This demonstration proved powerful — based firmly in a real-life, day-to-day use case.</p><h3>4. Provide channels for peer-to-peer feedback, support and coaching — and track adoption and impact</h3><p>After the roll-out call/demo, we established a dedicated Slack channel to foster ongoing dialogue and so our AIDT veterans can help the wider team on their journey to proficiency.</p><p>It’s important for everyone to share experience in an ongoing forum, not just your most experienced people. It’s also best practice to keep momentum on regular knowledge sharing sessions.</p><p>You can’t just throw something out there and assume you’re “done” — you have to measure and manage it. So we’re tracking usage of the tools, identifying patterns against projects and use cases — and relaying those back into the organization to make everyone successful.</p><h3>5. Help adopters take usage to the next level</h3><p>It’s a common trap for companies to think they’ve successfully adopted a new tool when 90% of their team is using it, but that’s actually when the real work begins.</p><p>When it comes to adopting something like Cursor, there are 2 levels of proficiency:</p><p><strong>Level 1: The Quick Wins</strong></p><ul><li>Developers begin with basic autocomplete functionality</li><li>Tab completion for classes, methods, and common patterns</li><li>Immediate productivity boost through faster coding, but only single file changes</li></ul><p><strong>Level 2: Advanced Techniques</strong></p><ul><li>Multi file changes</li><li>Using “compose” i.e. engaging with the AI as a collaborative “junior” developer for simple tasks or bootstrapping new projects</li><li>Using chat functionality to get insights on the codebase or as a thought partner for new features and enhancements</li><li>The key challenge for engineering leaders: How do you encourage teams to progress from Level 1 to Level 2?</li></ul><p>The key challenge for engineering leaders: How do you encourage teams to progress from Level 1 to Level 2?</p><p><strong>What’s working for us:</strong></p><ul><li>Tracking usage patterns to identify where developers are in their adoption journey</li><li>Organizing working groups where power users share their experiences</li><li>Creating dedicated sessions for knowledge sharing and capability building</li></ul><p>But it doesn’t stop there, especially with AI tools advancing so quickly. You need to keep an eye out for new tools and developments to ensure your team has the best support and your company isn’t left behind.</p><p>Right now Cursor is working for us but we won’t be resting on our laurels. This is an ongoing process.</p><h3>Conclusion</h3><p>Successfully adopting AI development tools requires a systematic, measured approach rather than rushing into widespread implementation. Through our careful testing and phased rollout of Cursor across the development team, we demonstrated that methodical validation, peer-led demonstration, and ongoing support are crucial elements for gaining genuine buy-in from developers.</p><p>The five-step process we’ve outlined — validating the tool’s effectiveness, building a core group of advocates, explaining benefits through peer demonstration, providing continuous support channels, and tracking adoption metrics — provides a framework that other organizations can follow. This approach not only ensures the selected tools truly enhance productivity but also addresses common concerns around security and workflow disruption while building lasting adoption among traditionally skeptical development teams.</p><p>You need to think carefully and introduce tools that don’t cause friction or are easy to migrate to. We were already familiar with VS code which is very similar to Cursor, so it wasn’t too hard for the team to adopt. If we’d picked a completely different tool it would have been way harder to make the change and win buy-in from seasoned devs.</p><p>Most importantly, by bringing team members into the evaluation and decision-making processes, you ensure the team wants and owns the change.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1db4006cd5d6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/5-steps-for-successful-ai-development-tool-adoption-1db4006cd5d6">5 Steps for Successful AI Development Tool Adoption</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why we Continue to Invest in React Native in 2025]]></title>
            <link>https://medium.com/whitespectre/why-we-continue-to-invest-in-react-native-in-2025-a0e1bccc1eea?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0e1bccc1eea</guid>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[react-native]]></category>
            <category><![CDATA[cross-platform]]></category>
            <category><![CDATA[typescript]]></category>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Thu, 06 Mar 2025 11:37:42 GMT</pubDate>
            <atom:updated>2025-03-06T11:37:42.497Z</atom:updated>
            <content:encoded><![CDATA[<p>Find out how React Native matured from a Facebook experiment to become the industry standard — powering 30% of top iOS apps with native-like performance — and why it’s the strategic choice.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LerXGz-0tuq6tAJOvUAbRA.jpeg" /></figure><p>Is it worth investing in React Native in 2025? At Whitespectre, our experience tells us that the answer is yes. Today, we’re not just choosing React Native because it saves cost, we’re investing in it as a strategic technology that accelerates development, delivering tangible value.</p><p>In this article, we explore the evolution and maturity process of React Native, highlight its recent updates, and share success stories outlining why React Native is the framework of choice for building robust, high-performance mobile apps in 2025.</p><h3>From Promising Prototype to Mature Platform</h3><p>To understand the history of React Native, let’s go back a decade, when Facebook open-sourced it, promising a unified codebase capable of powering both iOS and Android apps. Here’s a brief timeline:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PKppN4DWMWQSS6TRw6fv1Q.jpeg" /></figure><h3>The Why</h3><p>Now that we’ve explained how React Native has evolved into a truly mature platform as of 2025, let’s explore its benefits in more detail:</p><h4>Talent Portability &amp; Unified Skill Sets</h4><p>React Native utilizes JavaScript and TypeScript languages already familiar to a vast pool of web and mobile developers. This makes it simpler to transition talent across projects, reduce onboarding time, and promote a culture of shared knowledge. Teams adept in web technologies can quickly adapt to building native-like mobile apps.</p><h4>Code Reusability &amp; Faster Iterations</h4><p>By maintaining a single codebase for both iOS and Android, teams eliminate the need to duplicate efforts. They can quickly tailor incremental changes for specific platforms (iOS or Android) when needed, accelerating feature development and simplifying maintenance.</p><h4>Active Community</h4><p>Besides receiving strong support from Meta, React Native boasts a thriving, global community of developers that consistently contributes open-source libraries, plugins, and core enhancements to React Native continuously improving its performance and capabilities. This ecosystem promotes innovation and ensures quick resolution of common issues.</p><h4>Performance</h4><p>React Native apps can now achieve native- like performance, with libraries like <em>reanimated</em>, <em>react-native-mmkv</em> and <em>react-native-flashlist</em>, all while preserving the advantages of a popular language and an active community.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w_HomVJxqvVEFqKGvwH8Rw.png" /><figcaption>Example Code Snippet: React Native + Expo (2025 Best Practices)</figcaption></figure><h4>Real Stories</h4><p>A <a href="https://x.com/Baconbrix/status/1888633966938276267">Recent 2025 App Store research </a>positions React Native as the top cross-platform framework for building shopping apps. Impressively, 30 of the top 100 iOS apps (including the number-one-ranked app) are built with React Native. This widespread adoption shows its strong performance and ability to deliver exceptional user experiences in production.</p><p>If you want to explore more major apps using React Native, including Amazon, Facebook, Discord, and Shopify, check out <a href="https://reactnative.dev/showcase">Meta’s showcase</a>.</p><h3>Conclusion &amp; the Future</h3><p>Over the past decade, React Native has evolved from a promising prototype into a fully mature cross-platform framework. Its native-like feel, unified codebase, and robust ecosystem have empowered us to build scalable, efficient apps that meet the demands of today’s market for the likes of <a href="https://www.whitespectre.com/work/purple-carrot-growth-to-acquisition/">Purple Carrot</a> and <a href="https://www.whitespectre.com/work/spoton-case-study-gps-fence/">SpotOn</a>.</p><p>While other frameworks like Flutter continue to offer innovative approaches, our experience has shown that React Native’s deep integration with native modules, Expo ecosystem and supportive community make it the optimal choice for our strategic initiatives at Whitespectre.</p><p>Looking ahead, React 19 is already bringing even greater performance gains by automating many of the optimizations that previously depended on manual hooks like useMemo and useCallback. This evolution promises cleaner codebases and faster, more responsive apps.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8RADIOJkaURNrKNiSPFAiQ.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0e1bccc1eea" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/why-we-continue-to-invest-in-react-native-in-2025-a0e1bccc1eea">Why we Continue to Invest in React Native in 2025</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Five IoT Innovations Powered by Software Alone]]></title>
            <link>https://whitespectre.medium.com/five-iot-innovations-powered-by-software-alone-b0aa2376c503?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0aa2376c503</guid>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[product-innovation]]></category>
            <category><![CDATA[internet-of-things]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Fri, 21 Feb 2025 14:42:45 GMT</pubDate>
            <atom:updated>2025-02-21T14:42:45.209Z</atom:updated>
            <content:encoded><![CDATA[<h3>From AirPods becoming hearing aids to smart bulbs detecting motion, software innovations are revolutionizing IoT capabilities while reducing costs. Learn how companies are maximizing hardware potential through smart algorithms.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nwEGHg0Co_xpYBWvzDDTGQ.jpeg" /></figure><p>When device performance needs improvement, or users are requesting new features, many teams default to hardware upgrades.</p><p>But developers for some of the world’s biggest IoT products — like Apple, Tesla, and Amazon — show that strategic software updates often deliver better results at a fraction of the cost and at scale.</p><p>Finding new ways to use existing sensors, for example, not only negates the need for expensive new sensors: it also means faster product updates, accessible across existing devices, which means happier users and more value from existing inventory.</p><p>For example, Apple turned its AirPods Pro 2 into FDA-approved hearing aids with a software update which applied hearing assist features once limited to specialized medical devices.</p><p>Meanwhile, Tesla replaced expensive light and range detection hardware (LiDAR) with an AI-powered vision system that only requires existing cameras, driving down hardware costs.</p><p>Check out our infographic for 5 examples of this trend we love (<a href="https://www.whitespectre.com/work/spoton-case-study-gps-fence/">including our own work with SpotOn</a>), along with 4 top tips for how to maximize your IoT hardware:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UAOS2h06e0Yf5nEFtYN1lA.jpeg" /></figure><p>For more IoT insights, find out how we developed an award-winning GPS dog collar and virtual fence with <a href="https://www.whitespectre.com/work/spoton-case-study-gps-fence/">SpotOn</a> and our work on a <a href="https://www.whitespectre.com/work/iot-wearable-healthcare-technology-case-study/">healthtech wearable</a> that drives better patient outcomes with revolutionary ID-monitoring software.</p><p>You can also read <a href="https://www.whitespectre.com/ideas/engineering-offline-iot-solutions/">our guide to engineering offline-resilient IoT applications</a>, turning the challenge of disconnection into an opportunity for enhanced functionality, along with our <a href="https://www.whitespectre.com/ideas/ux-principles-for-iot-apps/">top 6 UX principles for creating great IoT apps</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0aa2376c503" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Debunking React Native Myths: Why Expo is Now Our Go-To (Even for Complex Projects)]]></title>
            <link>https://medium.com/whitespectre/debunking-react-native-myths-why-expo-is-now-our-go-to-even-for-complex-projects-40513f902d8a?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/40513f902d8a</guid>
            <category><![CDATA[mobile-development]]></category>
            <category><![CDATA[react-native]]></category>
            <category><![CDATA[expo]]></category>
            <category><![CDATA[tech-insights]]></category>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Fri, 31 Jan 2025 17:55:58 GMT</pubDate>
            <atom:updated>2025-01-31T17:55:58.654Z</atom:updated>
            <content:encoded><![CDATA[<h4>Old perceptions die hard in tech, and Expo’s reputation as a ‘beginners-only’ React Native framework persists despite its evolution. Learn why we now choose it even for our most demanding, high-performance mobile projects, and why you should too.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iEwpl6GuuR7_p1-2xQMYoQ.jpeg" /></figure><p>For many years, a common belief in the React Native community was that Expo was mainly suited for beginners and simple projects, while Bare React Native was the only viable option for complex, production-ready applications.</p><p>This wasn’t just perception — in Expo’s early days, essential tools like Firebase required ejecting, and native module integration was limited. But Expo has undergone a remarkable transformation. Today, those old limitations have become myths, persisting despite Expo’s evolution into a robust framework capable of handling sophisticated, high-performance applications with complex requirements.</p><h3>Understanding the Approaches</h3><p>React Native development offers two primary approaches. Bare React Native provides direct access to the native project files and build configurations, traditionally favored for its flexibility. Expo builds on top of React Native, providing a managed development environment with pre-configured build tools and a standardized project structure.</p><p>While Expo initially traded some flexibility for development speed, its modern iteration has evolved to offer the best of both worlds. Since React Native 0.74, Expo has become the officially recommended approach in React Native’s documentation. Yet, many teams still hesitate to use it for complex projects due to persistent myths about its capabilities. Let’s examine these misconceptions.</p><h3>Myth #1: “Expo Can’t Handle Complex Native Features”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gCWQ8CRM3RQrok4Ff4PjIw.png" /><figcaption>Built with Expo: <strong>Flexport’s</strong> iventory tracking app matches traditional React Native capabilities.</figcaption></figure><p><strong>The Old Reality:</strong> Projects requiring native platform features, hardware integration, or specialized SDKs traditionally demanded Bare React Native. This often meant navigating complex build processes, managing dependencies, and writing platform-specific code and customized bridging to Javascript.</p><p>Just a few years ago, when we started building the mobile app for <a href="https://www.whitespectre.com/work/spoton-case-study-gps-fence/">SpotOn’s pet containment system</a>, Bare React Native was the right choice for handling ultra-wideband (UWB) communication and advanced Bluetooth Low Energy (BLE) features. The same was true when developing Merryfield’s mobile receipt scanning platform — its specialized native SDK integration and precise image processing requirements aligned perfectly with Bare React Native’s capabilities at the time. These decisions were spot-on for their moment, but Expo’s capabilities have evolved dramatically since then.</p><p><strong>The Expo Reality: </strong>Expo now offers a robust suite of tools and APIs that make integrating complex native features a breeze:</p><ul><li><strong>Expo Modules API: </strong>Seamlessly integrate native modules for functionalities like Bluetooth communication, background tasks, sensor access, and more. This means accessing platform-specific features is no longer a Bare React Native exclusive.</li><li><strong>Custom Development Clients:</strong> Extend Expo’s capabilities by integrating custom native code while still benefiting from Expo’s streamlined workflow. This provides a flexible escape hatch for highly specialized needs without sacrificing the convenience of Expo.</li><li><strong>Config Plugins:</strong> Customize native configurations without leaving the Expo ecosystem. This allows for fine-grained control over your app’s interaction with the underlying native platform.</li></ul><p>With these capabilities, we now confidently choose Expo even for projects with complex native requirements.</p><h3>Myth #2: “Expo Sacrifices Performance”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O7fV7LDiowfM6Qqr4PjJlw.png" /><figcaption><strong>Brex’s</strong> lightning-fast Expo app proves performance fears are outdated.</figcaption></figure><p><strong>The Old Fear: </strong>Complex applications with high-performance demands required direct native optimization through Bare React Native. Developers worried Expo’s managed environment and JavaScript-based architecture might introduce performance bottlenecks.</p><p><strong>The Expo Truth:</strong></p><ul><li><strong>First-Class Support for the New React Native Architecture:</strong> Full support for Fabric and TurboModules, enabling enhanced performance out of the box</li><li><strong>Well-Implemented Built-In APIs: </strong>Reducing common performance pitfalls through well-implemented pre-built solutions.</li><li><strong>Performance Profiling: </strong>Expo integrates powerful performance profiling tools that allow developers to identify, analyze, and resolve potential bottlenecks within their applications.</li></ul><p>Expo’s performance optimization tools and architecture support ensure that developers can build complex, resource-intensive applications without compromise.</p><h3>Myth #3: “Expo Limits Customization”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_PZqgw6fVvixSE5cxzdq5A.png" /><figcaption><strong>Prettylittlething’s </strong>unique brand shines through Expo’s flexible customization options.</figcaption></figure><p><strong>The Old Perception: </strong>Expo was perceived as restricting developers to its pre-defined components and configurations, limiting the ability to create truly unique and customized applications.</p><p><strong>The Expo Flexibility:</strong></p><ul><li><strong>Styling Freedom:</strong> Expo completely controls UI elements and animations, allowing developers to craft unique brand experiences and tailor the app’s look and feel.</li><li><strong>Build Process Control:</strong> Customize build configurations while maintaining Expo’s workflow benefits, giving developers greater control over the final app output.</li><li><strong>Custom Component Integration: </strong>Create and integrate native components through a structured, maintainable approach.</li></ul><h3>Myth #4: “You Need Native Expertise”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pMJWobp7Q_b5GK4rgJ870Q.png" /><figcaption><strong>Bounce</strong> ships fast updates without native devs thanks to Expo’s EAS.</figcaption></figure><p><strong>The Old Belief:</strong> Building sophisticated React Native apps required extensive native platform expertise (iOS and Android) to handle everything from implementing features and integrating third-party services to configuring builds and publishing the app on app stores.</p><p><strong>The Expo Reality:</strong></p><ul><li><strong>Well-Documented Guidelines:</strong> Expo provides clear, comprehensive, and beginner-friendly documentation for its APIs and features. This ensures developers can seamlessly implement native functionalities without needing advanced platform-specific expertise.</li><li><strong>Expo Application Services (EAS):</strong> Expo offers powerful application services, such as EAS Build, Submit, and Update, to handle tasks like compiling code, deploying updates, and publishing apps to the app stores. These services streamline the entire development-to-deployment pipeline.</li><li><strong>Clear Upgrade Path for Maintenance and Evolution: </strong>Expo provides a structured approach to keeping projects up-to-date with the latest React Native and native platform (iOS and Android) upgrades. This ensures your app remains compatible with evolving technologies, making it easier to maintain, implement new features, and adapt to platform changes over time.</li></ul><h3>When to Consider Bare React Native</h3><p>While Expo elegantly handles most requirements, there are specific scenarios where Bare React Native might still be the preferred choice:</p><ul><li><strong>Existing Bare React Native Codebase: </strong>If your team is already working with a Bare React Native codebase, it might be more efficient to continue with that approach.</li><li><strong>Nonstandard Project Structure or Build Requirements: </strong>If your project requires a custom directory structure, specialized build configurations, or nonstandard native integrations (such as custom React Native CLI configurations, unique module linking, or complex asset handling), Bare React Native gives you full control over the project structure and build process, enabling you to implement and manage these customizations without restrictions.</li></ul><h3>Looking Forward</h3><p>The evolution of Expo represents a fundamental shift in React Native development. Rather than choosing between development speed and technical capabilities, teams can now have both. Our experience with complex projects has shown that starting with Expo today is not just viable — it’s often the smartest choice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=40513f902d8a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/debunking-react-native-myths-why-expo-is-now-our-go-to-even-for-complex-projects-40513f902d8a">Debunking React Native Myths: Why Expo is Now Our Go-To (Even for Complex Projects)</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[FlashList vs. FlatList: Understanding the Key Differences for React Native Performance]]></title>
            <link>https://medium.com/whitespectre/flashlist-vs-flatlist-understanding-the-key-differences-for-react-native-performance-15f59236a39c?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/15f59236a39c</guid>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Tue, 28 Jan 2025 12:19:29 GMT</pubDate>
            <atom:updated>2025-01-28T12:19:29.098Z</atom:updated>
            <content:encoded><![CDATA[<h4>In this article, we explain how we improved performance and UX issues using FlashList as a better alternative to FlatList. We’ll cover why FlashList’s cell recycling strategy outperforms FlatList’s virtualization for complex layouts and share practical tips to fine-tune performance, improve FPS, and avoid crashes.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z56-aBzHYmEdT8Gf0sFFew.gif" /></figure><p>Scrolling performance plays a crucial role in defining the quality of user experience in mobile apps, especially when handling complex or long lists.</p><p>Whether it’s an infinite social media feed with videos and images, an e-commerce catalog, or a meal menu with multiple categories: As complexity increases — with features like animations, auto-scrolling, custom filters, and sorting rules — the performance demands also grow. Keeping a smooth experience becomes progressively more challenging, particularly when managing large lists and ensuring good performance across all types of devices.</p><p>We faced this challenge when building a project where one of the core parts of the app featured categorized menus with nested lists, dynamic content, and animations. Despite using React Native’s FlatList, performance issues surfaced during our testing before the Android launch, even though the iOS version was already in production and running well. These issues were particularly noticeable on low-end Android devices, which resulted in laggy scrolling, animation stuttering, and even crashes.</p><p>In this article, we’ll explore how we addressed these challenges using FlashList, share tips and strategies to fine-tune scrolling performance, and explain why, in certain scenarios, FlashList can deliver a smoother and more efficient user experience, with its recycler view approach, than FlatList, which relies on a virtualization strategy.</p><h3>FlatList: A reliable choice, but…</h3><p>React Native has a built-in component for handling large lists, that you’ve probably already heard of, named FlatList. It uses a technique called virtualization to optimize rendering, that means that instead of rendering all items at once, like a View or ScrollView would, FlatList renders only the items currently visible, with a buffer at the top and bottom to support smooth scrolling. Items outside the viewport are unmounted to save memory and improve performance.</p><p>While this approach works well for most cases, it introduces some expected overhead because mounting and unmounting components consume device resources. When list items are complex, such as including animations or dynamic content, this overhead can lead to performance issues. These problems are even more noticeable on low-end devices, where limited processing power can result in laggy performance and a poor user experience.</p><h3>The Solution: Introducing FlashList</h3><p>Unlike FlatList, FlashList uses a cell recycling strategy instead of virtualization. It works by keeping a fixed pool of component instances in memory. When an item scrolls out of view, it reuses the same component with new data instead of destroying and re-creating it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*XwDOHpyS6LmeDitIqcukEQ.png" /><figcaption>Source: WiseITeach (2018)</figcaption></figure><p>This approach can drastically improve performance, especially in scenarios where the list is complex, making the mounting and unmounting very costly on low-end devices.</p><h3>Our Experience: From FlatList to FlashList</h3><p>When we encountered performance issues related to this complex nested list on the project we were working on, we decided to give FlashList a try. Since it is designed as a drop-in replacement for FlatList and uses a recycling approach, it was well-suited to improve performance on devices with limited resources, so the replacement process was very straightforward.</p><p>After the change, we immediately noticed a performance boost, and scrolling became significantly smoother. But, after further investigation and a closer review of the documentation, we realized there was still room for additional improvements and by applying a few optimizations, we were able to enhance performance and user experience even further.</p><h3>FlashList Fine-Tuning Tips</h3><p>Based on insights from the documentation, we implemented some key enhancements:</p><ul><li><strong>Data Reorganization:</strong> We merged multiple sub-lists into a single flat list, making it easier to handle recycling.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jIc6f7Au8VukVM7kKh-CAw.png" /></figure><ul><li><strong>estimatedItemSize:</strong> This mandatory prop helps FlashList determine how many items to render initially and while scrolling. Since our list contained different types of components, we followed the documentation’s recommendation to use an average or median value for estimatedItemSize to optimize performance.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/802/1*m0fQ8msDPE_5TmP40c-LVg.png" /></figure><ul><li><strong>getItemType: </strong>By transitioning from a nested list to a flat list, we were able to group items by type (headers vs. items), and use the <strong>getItemType </strong>prop, enabling smarter recycling.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/650/1*l51DcgbvUDxZp_kaG8nUVQ.png" /></figure><ul><li><strong>overrideItemLayout:</strong> This has a higher priority over <strong>estimatedItemSize</strong>. Assigning fixed heights for headers and meals improved scroll precision for the auto-scroll feature.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g_xaFdmAN11ZWvYPMwtjNw.png" /></figure><p>After implementing all these optimizations, our list component looked like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XAJERrUEZEYL3N8j2yANtg.png" /></figure><h3>Results</h3><p>After implementing FlatList and fine-tunning properties, we saw a significant improvement in performance, to measure it we used Flashlight a Lighthouse-like tool for mobile apps, with the same environment and Android device.</p><p>Average FPS improved from 36.9 FPS to 56.9 FPS, a 54% increase</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8KL5LvT4xcQ6FrwEwqceZg.jpeg" /><figcaption>FPS before FlashList implementation and optimization</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_G1xtM8W7AnhKFdFoAyYnA.jpeg" /><figcaption>FPS after FlashList implementation and optimization</figcaption></figure><p>The average CPU usage decreased from 198.9% to 36.5%, an 82% reduction. Improving app responsiveness, particularly on low-end devices</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E-zGOL2NmY_mYN0fmtSweQ.jpeg" /><figcaption>CPU usage before FlashList implementation and optimization</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RUw2IuqoSsXzEo71VA96WA.jpeg" /><figcaption>CPU usage after FlashList implementation and optimization</figcaption></figure><p>We also saw a substantial improvement on React Native JS thread CPU usage. Previously, it exhibited consistent utilization above 90%, leading to potential scrolling and transition bottlenecks. After using FlashList and optimizing it, the JS thread stayed consistently below 10%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BrtJEqqGfojeOJ1vLELSXg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bNMkz9iWPlh5num5AwcySA.jpeg" /><figcaption>Crash rates related to out-of-memory errors were eliminated.</figcaption></figure><p>Below, you can see a summary of the comparison results:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*os1Bkv08zQxeWoIQ2wCEdg.jpeg" /></figure><h3>Conclusion</h3><p>FlatList is a solid choice for most applications, especially simpler lists. But, it can struggle with performance when handling complex layouts and animations. That’s where FlashList might be a good choice, due to its cell recycling strategy.</p><p>That said, like almost everything in software development, it’s not a silver bullet. It works best when matched with the right use case and properly configured. When used in the right scenario, FlashList can deliver noticeable performance improvements and a smoother, more responsive user experience.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=15f59236a39c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/flashlist-vs-flatlist-understanding-the-key-differences-for-react-native-performance-15f59236a39c">FlashList vs. FlatList: Understanding the Key Differences for React Native Performance</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Behind the Product: How Tech Matters is Bringing Silicon Valley to Social Good]]></title>
            <link>https://medium.com/whitespectre/behind-the-product-how-tech-matters-is-bringing-silicon-valley-to-social-good-b722fb8576bc?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/b722fb8576bc</guid>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Fri, 24 Jan 2025 15:24:57 GMT</pubDate>
            <atom:updated>2025-01-24T15:24:57.547Z</atom:updated>
            <content:encoded><![CDATA[<h4>Jim Fruchterman, CEO and Founder at Tech Matters, explains why nonprofits shouldn’t chase the latest tech trends — and how to make technology actually work for social impact.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i4baui-CUdgeJwJluEkMcQ.jpeg" /></figure><p>You might expect a former Silicon Valley AI pioneer who founded one of the valley’s early machine learning companies to be championing the latest cutting-edge technology. Instead, Jim Fruchterman, CEO and Founder of Tech Matters, spends much of his time talking nonprofits out of building apps they don’t need or chasing AI solutions that won’t serve them.</p><p>It’s this practical approach — honed over three decades of bringing technology to the social sector — that’s made him one of the most trusted voices in tech for good. Through <a href="https://techmatters.org/">Tech Matters</a>’ flagship projects <a href="https://techmatters.org/project/1000-landscapes/">Terraso</a> and <a href="https://techmatters.org/project/child-helplines/">Aselo</a>, he’s proving that sometimes the most impactful solutions aren’t the shiniest or newest, but the ones that actually work for the people who need them most.</p><p>Just recently, we partnered with Terraso through our <a href="https://www.whitespectre.com/green-grants/">Green Grants</a> program to help bring offline functionality to their Land Potential Knowledge System App, enabling farmers in remote areas to access crucial soil data without internet connectivity — exactly the kind of practical solution Fruchterman champions.</p><p>From providing soil analysis tools to farmers in developing countries to modernizing child helplines across 15 nations, Jim sat down with us to share his insights on how tech companies can make a real difference in the social sector, why 95% of nonprofit apps fail, and his ambitious new initiative to revolutionize how organizations handle data ethically.</p><p><em>If you’re a nonprofit interested in partnering with Whitespectre as part of our </em><a href="https://www.whitespectre.com/green-grants/"><em>Green Grants</em></a><em>, </em><a href="https://www.whitespectre.com/green-grants/"><em>please get in touch</em></a><em> — and let us know how we can help.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*R_1s3XbOWNa_aR4BNlSF_Q.png" /></figure><p>We get into:</p><ul><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section1">How tech companies can get involved with more pro bono work</a></li><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section2">How to approach climate tech</a></li><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section3">How tech companies can make the biggest real-world impact</a></li><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section4">How AI can make a big impact for nonprofits</a></li><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section5">The ethics of using AI for social good</a></li><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section6">The high failure rate of non-profit apps</a></li><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section7">Getting end user adoption</a></li><li><a href="https://www.whitespectre.com/ideas/interview-behind-the-product-tech-matters-social-good/#section8">Data colonialism and the Better Deal for Data</a></li></ul><p><em>We started by asking how we could further our existing pro bono work at Whitespectre:</em></p><p><strong>What’s the best way for tech companies like ours — who want to be able to make a social impact, but have constraints of still having a business that works — to get more involved in pro bono work?</strong></p><p><strong>Jim:</strong> Well, the great news is there are now hundreds of non-profit organizations with tech teams. For obvious reasons, right? Just because software is behind everything these days. More and more, people who are ambitious about social change are realizing, “if I’m going to try to change the world for a million people, software and data just might be involved somewhere.”</p><p>And yet, these teams are generally tiny. They tend to be generalists. So that means they care a lot about the mission, but if you’ve got a three-person team it can be difficult to get the best from technology.</p><p>So we have two kinds of models that generally work for tech companies to collaborate with nonprofits.</p><p>The first model is when a nonprofit would like to talk to experts in their field about something they don’t know about. The thinking is “if we talked to someone who really knew a lot about that, they would have saved us a month of developer time from two or three hours of conversation.”</p><p>If you find someone who can say: “No, you don’t want to use that library. That’s terrible.” Or “have you thought about this? You haven’t? Well, maybe you should take a step back and think about that first because you shouldn’t actually build this until you’ve really decided.”</p><p>I had an early project, which is one of the first crypto-enabled software packages for human rights defenders, and being able to tap the crypto security field for expert advice was essential because we had just one developer working on this app. They said “using crypto to help human rights defenders is one of the coolest things you could be using crypto for, and we would love to advise you.”</p><p>And, of course, in a lot of cases what you need is to be told which technology infrastructure to use because a lot of modern technology is about picking the right infrastructure and your code is the 2% on top. So that’s the first model.</p><p>The second model is, where you have a longer term engagement with tech companies who can help you with specific things you’re working on. For example, Okta: we’re not going to reinvent multi-factor authentication for our sensitive information apps. So we work with Okta because they’re part of the <a href="https://www.pledge1percent.org/">Pledge 1%</a> crowd.</p><p>A Pledge 1% company can come either with a grant and/or with access to their products and team for a set period.</p><p>And of course, we’re big fans of <a href="https://www.techtotherescue.org/">Tech To The Rescue</a>, which also organizes these pro bono projects and I know you’ve had great experiences with them.</p><p>I think there’s a lot of opportunity here that’s consistent with the business imperatives, which is exciting and interesting.</p><p>It’s often one of the cooler applications of specific technology as developers working on climate change, working on kids rights, working on helping people grow more food. Usually our tech applications, I’m speaking for the Tech for Good field, are very visceral and understandable. It’s obvious why this matters.</p><p>And your average tech person would love to use their skill set to do something that’s socially important, yet probably isn’t willing to take a 50 or 70% pay cut!</p><p>So this is a great way where they can actually make a big difference. And, help the people who’ve made that kind of sacrifice to get on a better path because they’ve gotten expert advice or they got a project done that they couldn’t have done themselves but that they can build on and maintain.</p><p><strong>How do you find for-profit companies that are interested in doing that?</strong></p><p>Well, I think the first thing is to look at the companies behind your infrastructure when you’re deciding on a new product — those become your first targets. For example, our software for national child help lines, which operates in 15 countries through our Aselo, is built on top of Twilio Flex, Amazon Web Services, and Okta. Each of these companies has provided us with grants, free products, and technical assistance.</p><p>That’s your inner circle — the companies whose products you’re already using. Like when we needed help with PagerDuty for our operations monitoring, they not only gave us a grant but also provided staff support.</p><p>There are also organized volunteer programs like Tech To The Rescue, which started in Poland. They help connect nonprofits with volunteers based on specific needs — if you need someone who knows React, they’ll find that person. Many major companies, especially big banks and insurers, have their own tech volunteer programs.</p><p>Word of mouth in the field is crucial. Your peers in other nonprofits can tell you things like, “JP Morgan Chase did a great job on this specific project that matched their team’s expertise,” or warn you about companies that weren’t helpful and wasted a lot of time. It’s about finding the right fit.</p><p><strong>So it’s never too early for nonprofits to reach out to tech companies and start collaboration in order to avoid taking the wrong path…</strong></p><p>Exactly. In the tech industry, you learn what works and doesn’t work through various channels — your own experiences with failure, seeing clients struggle, or reading industry publications about why 95% of startups in a particular space failed. You add all of that to your mental model.</p><p>The nonprofit sector doesn’t have that same knowledge base. Tech people can help bridge this gap, but we often run into tech boosterism — overly optimistic promotion that drowns out practical voices. We’re seeing this with AI right now. Some AI companies, driven by financial interests, are drastically overstating their current capabilities. They’re betting on future developments and working to reduce costs and energy usage.</p><p>Will it work? I don’t know, but I’m glad for-profit companies are spending billions figuring it out, but I don’t want nonprofits wasting resources trying to be at that leading edge when even larger companies are struggling. Let’s wait for them to figure it out, and then we’ll implement the proven solutions three to five years later.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EaiXY6LcfqsfrpkwjX3d1A.jpeg" /></figure><p><strong>There’s obviously a disparity in how much bandwidth a smaller business can donate versus a much larger enterprise. How can we do more to encourage others — our competitors and our peers in the field — to do the same?</strong></p><p>I tried a start-up 15 years ago called Code Alliance where we’d take tech company volunteers and have them work on open source projects, and it didn’t work. It was hard. Google obviously has a big volunteerism program, but even though they’re big, it was often a challenge to get people to actually show up for volunteerism gigs. So it’s often hard to do.</p><p>The thing I like about Tech To The Rescue is we’re at the point where more and more nonprofits have a tech team so that the volunteerism actually matters over time. And their target for volunteerism is actually companies of your size.</p><p>If it turns out your team is mega-crunched and they have no bandwidth, you’re not going to do a volunteerism project in the next quarter because you’ve got some deadlines, right? But if you look and say “we’re a little soft in the summer and we got a couple of people at loose ends and we’d like them to use their skills and pump them up and we’re willing to continue paying them because we have a big contract coming up in September” then it’s actually a really good business case for maintaining your important human assets to give them something exciting to do rather than having them twiddle their thumbs. And they get to feel great about it.</p><p><strong>We’ve been working with Tech Matters on your Terraso app, as part of our pro bono </strong><a href="https://www.whitespectre.com/green-grants/"><strong>Green Grants</strong></a><strong> initiative. But speaking generally, how should a company like ours look to get involved when it comes to climate tech?</strong></p><p>The climate space is really difficult. Climate finance and hedge funds have tech people — or they can hire tech people directly or get an agency to do some work for them.There’s plenty of opportunities to do climate work for people who have money because our capitalist system works pretty well for that. There’s so interesting stuff going on in the supply chain and by and large it’s fundable in a capitalist way, with customers.</p><p>But we’re interested in the harder part — the other 90%. Your local farmer in rural Asia, Africa, or Latin America isn’t an attractive target for profit. That’s where our model with Terraso and our suite of tools comes in. We’re focused on the person on the front lines, or maybe the person managing the co-op, or the local NGO, or the land use planner in that county, province, or district.</p><p>If those are your targets, by and large, they’re not investible but they have a lot of the same needs as the equivalent person in a wealthy country. In rich countries, people use tools like Esri’s ArcGIS for mapping. But you can’t use that in most of the world. A lot of companies would use Salesforce for business management but you can’t use that in rural Zambia — I’m sorry but it’s just a bad idea. So that’s the gap we’re excited about filling as a nonprofit.</p><p>I’m not interested in competing with for-profit companies in viable markets — they’re generally better at that than nonprofits. But when companies can’t make enough money, or when local companies lack sophistication, that’s where we come in. Even though we’re a charity, most of the work on the ground is about people making money. Farmers want to increase their profits despite less rain and more heat on average. We’re trying to give them the missing tools and infrastructure.</p><p>We’re not alone — there are dozens of nonprofits like us tackling different parts of this problem. Many of us are open source, which encourages collaboration. For example, in Terraso, we needed data collection capabilities. The standard tech for Android data collection is Open Data Kit (ODK), and over the last 15 years, great tools have been built on top of that. KoboToolbox is one of the best of these — it enables you to collect data, pictures, maps, and more, all offline, which is crucial in many parts of the world. So we just built it in.</p><p>Another example is Precision Development, a terrific nonprofit — I think they’re UK-based. They provide advice to five million farmers in the developing world. An Iowa farmer can buy this kind of service, but a Kenyan farmer probably can’t afford it. So having an NGO provide this service, in partnership with the National Ministry of Agriculture, gives farmers the practical information they need.</p><p>From a volunteerism standpoint, these organizations often have their own small tech teams. The challenge is creating lasting impact. If you do tech work for a non-technical nonprofit, you might come back in a year and find they’re not using it anymore. But if they have even one tech person, you’ve got someone who can maintain and build upon what you created.</p><p>There’s another pattern we’re seeing: Universities and scientists are incredibly enthusiastic about climate solutions, building exciting models for various approaches. But university professors and their teams aren’t usually the best at developing practical, usable software. The science might be right, but how do you get that to a million farmers? That requires real product work, not just scientific work.</p><p>I see a big gap between proven research and actual implementation — getting it to the farmer or local policymaker who needs it. I think this gap will increasingly be filled by small tech-for-good nonprofits partnering with industry while getting scientific guidance. Not every professor wants to hear this, but just throwing technology at people without doing the product marketing and user experience work doesn’t stick.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zjRtpATfkVKCF5s1t2QiEg.jpeg" /></figure><p><strong>Are there areas where people could be making an impact on the climate crisis that may seem counter-intuitive — are there any unglamorous areas you’d identify that would yield a bigger impact if companies got into those spaces?</strong></p><p>The dirty secret of technology is that it’s usually just plumbing: Basic tools to help you get whatever you want to get done. Many of the AI tools people use today work without users even knowing they’re using AI. It’s just something that helps them get from point A to point B.</p><p>Making technology invisible is actually the secret to adoption. The shortest distance between me and my goal might be using technology, but I shouldn’t even have to think about it — it should just work and do what I need. The more energy you put into highlighting how cool or shiny it is, or how it connects to big controversial issues, the more you miss the point.</p><p>Take climate change: As a former scientist, I don’t think there’s much debate about humans causing it, but I don’t have to sell a farmer on that concept. I can just say, “Hey, if you change your farming method from this to that, someone will give you $10 per acre per quarter.” They’ll take that money without needing to buy into the larger debate.</p><p>Some of this stuff is really boring, but that’s actually how people use technology in real life. They use it to track their finances, plan their next steps, and communicate with others. Look at our two big programs — Terraso for climate change and Aselo for national health lines. What’s Aselo doing? It’s taking telephone-oriented helplines and adding text channels. In Silicon Valley, they’d say “That’s so 10 years ago!” But most helplines in the world are still taking telephone calls, while teenagers aren’t making many calls anymore.</p><p>Because of this trailing edge phenomenon, we don’t have to be at the leading edge or bleeding edge to be useful. We can take things that have been proven to work in the business sector and figure out how to apply these now-boring, standard tech infrastructure solutions to new contexts. That tech infrastructure and those benefits haven’t reached these people yet. If we can just make their lives that much easier, we’re doing what the tech industry does for a group of people who aren’t currently financially exciting.</p><p>If we do our job right, in five or ten years, these people are going to be middle class and part of a viable market. I think that one of the big dreams of the nonprofit sector is how do we graduate people out of poverty so they become functional members of the market economy and can buy that smartphone? Well, the smartphone is probably the one thing that has reached almost everybody, or at least the majority — but other tech solutions haven’t.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CvrHTrLGj5tvOz-XNTzC4w.jpeg" /></figure><p><strong>About six months or so ago, </strong><a href="https://www.youtube.com/watch?v=h1UjKJZpKso"><strong>you did a talk on AI for social good</strong></a><strong>. The AI space is progressing really quickly. You’ve spoken in the past about nonprofits tending to use older generations of technology. What’s your thinking about AI and how it might level the playing field?</strong></p><p>I actually started one of Silicon Valley’s early for-profit AI companies a long time ago — we were working in character recognition, which was then the leading edge and called machine learning at the time. Our particular breakthrough was being the first in the OCR field to use a database of millions of character examples to train the AI, rather than trying to design the algorithm from out of our brains.</p><p>I’ve spent over 30 years bringing AI to the nonprofit sector, and my number one piece of advice to nonprofits is: Don’t build an AI product. It’s hard and expensive. To create a product, you need a fair amount of data, which a lot of nonprofits don’t have. Over five years, you’re going to be spending seven figures, whatever your major currency is. That’s not what a million dollar nonprofit should be doing.</p><p>Even though I’m a complete enthusiast for AI — it’s been my career, especially bringing AI to social good — I spend a lot of my time talking people out of AI because right now the hype cycle is so powerful that people think it’s this magic wand. They think they can lay off half their staff and everything will be great. And I’m like “no, don’t do that!”</p><p>Last month I published something called the “Nonprofit AI Treasure Map.” It’s a single chart about how to find treasure with AI. The short answer is, if there’s a product: Try it. If it helps you far more than it hurts you: Use it. If not, wait a year and check again, because the dynamics could have changed. That’s what 99% of nonprofits should be doing.</p><p>Now, there are maybe a couple hundred nonprofits that have tech teams, know a lot about data, have a lot of data, and they’re doing really exciting stuff with AI. Some are pioneering within their organizations. Look at Khan Academy with the Khanmigo AI tutor. Will it work? I don’t know but it’s a bold experiment worth trying. Could a kid who could never pay for a tutor get an AI tutor that actually helps them significantly? Maybe, let’s try that.</p><p>Digital Green in India is doing agricultural advisory for farmers really well. MomConnect in South Africa focuses on maternal health advice, using AI to answer 20–60% of their questions with canned answers because, frankly, many questions from pregnant women are similar, and you can provide vetted, correct answers.</p><p>These are all great applications. They’re all about making the humans in the system more effective and powerful, reserving human interaction for the more challenging needs rather than routine ones. AI is great at eliminating drudgery. But if you’re going to invest a million dollars or more, you need to be eliminating enough drudgery to justify that investment.</p><p>Nonprofits have to justify their technology investments either through cost savings or massive expansion in the impact of their services — effectively lowering the cost of delivering a unit of social good. That’s what technology is for. How can we be 30% more efficient in two years with this technology? How can we take this one task and make it cost a quarter of what it costs now?</p><p>I often joke that your nonprofit should not be writing software any more than your corner restaurant should be writing software. Wait until there’s a product that actually solves your problem — one that solves the problem for 5,000 or 50,000 other organizations like you. That product is likely to be good, maintained, and improved over time, and affordable!</p><p><strong>Yeah, but it’s so interesting, right? Because it’s always the build versus buy for any business with what they’re trying to do and that kind of quality off-the-shelf software to meet the goals of whatever that the organization is trying to do is so much more cost effective than it was 15 years ago.</strong></p><p>And then if there’s one place where you go “oh my god, we have to do something much, much better than the status quo products,” that’s just one place rather than trying to do 15 things simultaneously.</p><p>You can buy or rent, 13 or 14 of those, and if you really have something that’s important, well then all you have to do is focus on that one missing piece that maybe what your effort turns on as opposed to basic things like security and servers and managing loads. People solve those problems for you and it’s affordable.</p><p><strong>A common criticism of using AI — or crypto, which you mentioned earlier — for moral good is that those technologies themselves have a very real and negative environmental impact. It would be fascinating to hear your outlook on that whole discussion.</strong></p><p>My view is that we’re usually using technology for harmful purposes before we use it for good ones. The ethical dilemma is often “why aren’t we using it for good?” Usually, the reason is simple: I can’t make enough money doing that, but I can make a lot selling technology to the oil and gas industry or whatever. People can use generative AI to create deep fake porn, for instance. So my challenge isn’t defending technology: It’s acknowledging that while technology is being used for something harmful here, it could be used for something beneficial there.</p><p>That’s actually why I got into AI in the first place. As a grad student, we were learning how to make smart missiles with camera systems in the nose that could recognize targets. When I was in grad school, most of my peers went to work for the military-industrial complex. I went back to my dorm wondering if there was a more socially beneficial application for pattern recognition than smart bombs, and thought, “Hey, you could use this to help blind people read.”</p><p>That’s where my career started. First building the technology as a venture-backed startup, raising $25 million in Silicon Valley. And when my investors vetoed the product for blind people, I said, “Well, I’m going to do it anyway,” and made reading machines as a nonprofit. That worked. So I’m always looking at the leading edge and trying to figure out how to adapt it for good.</p><p>Now, about the environmental impact of AI: We’re running into an issue where our arms race in the AI field is causing us to keep burning more coal. We can’t get enough energy fast enough. However, because the social good field is lagging behind, we’re actually in a different position. People ask us how we can store so much data for free. Well, poor people and small NGOs don’t have very much data. I’m not trying to be a Meta or Amazon or Google or Apple.</p><p>In many ways, the most energy-intensive uses aren’t going to be important in the social good sector. We can use cheap, thin models that are very computationally efficient that we can be running. We’re not going to run the most expensive model a thousand iterations to get the right answer.</p><p>I can give you two examples. In Terraso, our most advanced AI application is soil identification. By making maybe 15 or 20 measurements without a lab — just dig, take four measurements, go deeper, take four more, see how much rock there is — we have a little AI algorithm that tells you the soil type. If we can tell you the soil type, you know what will grow well on that hillside. That’s really important, but it runs on a smartphone! I’m not worried about the energy that little AI algorithm uses.</p><p>For Aselo, one of the best AI applications is analyzing chats. That’s kind of the slam-dunk thing. The bad idea would be to fire the humans and put a chatbot on the line — no empathy, no understanding, might accidentally cause someone to commit suicide. We’ve had a couple of news stories about that in the last few months. So don’t do that.</p><p>Instead, we use AI to understand what conversations are about — is it about bullying or assault or divorce — it’s pretty straightforward stuff as far as an AI use case goes and not computationally intensive. An increasing number of these helplines in different countries have enough data to run models without using LLMs at all. If you don’t have much data, you might need an LLM initially, but you could probably build a simple model that runs inexpensively once you figure it out. Use the expensive LLM for development, but when you go to scale, use something cheaper.</p><p>So while we do have a climate issue around big AI, I don’t think nonprofit use cases are going to be pushing the envelope in terms of being a significant part of that problem.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p6vCt5CZMpXfBmYP1U5YMA.jpeg" /></figure><p><strong>You’ve said elsewhere that 95% of the apps you’ve seen created for nonprofits fail. Why do you think that is?</strong></p><p>Being in Silicon Valley, I saw the app craze about 15 to 18 years ago. Investors did what they always do: They poured lots of money into lots of companies, and then there were winners and losers. People figured out what apps were good for, what they weren’t good for, and which apps worked. Some markets got locked up by incumbents, making it very hard to displace them.</p><p>But the nonprofit sector, to pick apps as an example, doesn’t get that memo. They’re not doing it for business reasons, and they’re often advised by people who think apps are still a thing. So I usually find myself talking people out of building an app because they’ve been told they need one, not because they actually have a use case for one.</p><p>Here’s a classic example: With Aselo, our helpline software, the founder of the movement — who isn’t technical — got a strategy plan from a management consulting company for the entire child helpline movement in 140 countries. Half of it was about an app. The founder came to me and said, “Jim, you’re a nerd I trust — what do you think?” I told them I didn’t think they had an app use case.</p><p>People call emergency numbers like 911 in the US because it’s well-socialized and they need help immediately. Your use case isn’t someone experiencing a dangerous situation thinking, “Let me pull out my smartphone, go to the Google Play Store, download an app, set it up, and then report an assault.” That doesn’t happen.</p><p>Yet nonprofits keep getting bad tech advice. Maybe it’s from a fresh MBA grad given an assignment by their boss, pulling down a 10-year-old playbook about apps being great. I joke that most of the time when I talk to nonprofit leaders about their tech idea, it’s usually anti-consulting.</p><p>In <a href="https://mitpress.mit.edu/9780262050975/technology-for-good/">my book</a>, there’s a chapter called “The Five Bad Ideas That Everybody Tries”: An app that no one will download; a gigantic database in the sky that people won’t use because they have search engines; blockchain or whatever the latest trend is. Eight years after the initial blockchain hype, I have yet to find a single nonprofit using blockchain effectively for social change at scale. Blockchain is like 0 for 1,000. But with AI, every wave brings something cool once we figure out what it’s good for.</p><p>I think it comes down to this: If you’re financially motivated, investors or your company’s investment committee won’t invest in a product that doesn’t make sense or goes against what everyone understands doesn’t work. But nonprofits don’t necessarily have that intelligence.</p><p>That’s part of my job — though I’m not really a consultant, I’m trying to build a couple of products here! But I spend a lot of my time advising people to not start with the technology, but to start with their problem and identify where it’s really painful.</p><p>If I was doing a startup, I’d go to Y Combinator and work on product-market fit. But if you’re doing technology in the social good sector, what you should be working on is product-user fit. If you haven’t got it, nothing else matters!</p><p><strong>Once you’ve completed one of these excellent tech initiatives, how do you get the adoption from end users?</strong></p><p>That’s the number one problem — distribution, implementation, adoption. It’s not the tech. The tech industry knows we can build anything, but getting people to use it is another issue entirely.</p><p>There are several models we use in the field. First, while you’re building the tech, you try to do more human-centered design to find out near the beginning that your product is completely unusable, rather than after you’ve spent all your budget. Let’s assume you’ve built a product that’s usable by your early adopters, and now you’re expanding to people who aren’t early adopters. You get a rude awakening that they’re very used to using the telephone, or collecting data on paper or in spreadsheets, and now you’re trying to get them to adapt to a SaaS platform or use a smartphone to collect data.</p><p>Like many developers, we don’t scale. If we want to have 50,000 users, we’re never going to be able to do that directly with a team of 15. So early on, we overinvest in early adopters, providing uneconomic levels of support to get insights, assuming that early adopter represents hundreds of customers in two years or thousands. We need to understand their problems and make sure we’re solving them.</p><p>For distribution, we’re always looking for existing channels we can use. In conservation work, it might be big orgs like UNDP, because none of the big nonprofits are particularly good at software development. We’re not writing software for big NGOs to use themselves — we’re helping their farmers. So we train the trainers, which is a very standard approach. In tech terms, we don’t want to be front-level support for our customers — we want to be second or third level support as a product maker — because many of our partners have infrastructure for training and support that we can leverage without recreating it.</p><p>Depending on your product’s maturity level, you end up spending more and more time trying to make adoption easier and smoother. Even though we know that rural people in developing countries aren’t as technically empowered as people in rich country cities, we’re often really reminded of this when we deploy a product and realize, “Oh, we thought they would know how to do this thing on their device, and they don’t.” So we have to add an even more basic level of training than we assumed.</p><p>Take our work with national child helplines around the world. During the peak of the pandemic, we were doing everything virtually, and it just wasn’t working well. We had underestimated the tech skill level of the average counselor in a child helpline in Africa. We ended up hiring software engineers in South Africa and sending them to Zambia, our first country, so we could actually see what was happening.</p><p>There are also cultural norms that make it very hard for people to tell you the blunt truth about your product. In South Asia, for example, they’re often too polite and will get more and more frustrated without telling you. You have to go out of your way to solicit feedback because if you wait for the squeaky wheel, people may be exploding in frustration before you’ve gotten wind that they’re unhappy with your product. So we never launch in a country without visiting and spending time with the frontline staff and planning for change management.</p><p>For individuals, you can’t afford that approach. Instead, you might go to a major conference in that area and talk to as many users as possible in one go. We can afford to do that even if we can’t talk to 300 farmers individually — if 300 of them show up at an event, we can have 25 meaningful conversations.</p><p>We also use tools like Fullstory, which blots out all the actual data entry fields but lets you watch what your users are doing. If they keep clicking on something that’s not clickable, you’ve got a product design problem. There are ways to gather insights without violating ethics or personal information. I’m always wondering how we can automate seeing what’s happening with our users to understand what’s working without violating privacy. It’s not always possible, but when it is, it’s usually smart. Of course, if you’re a big tech platform, this is second nature — you’re watching everything.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j5h6YezYqnVtkmpgyfzsiQ.png" /></figure><p><strong>We know you have a book coming out. What else are you most excited about in the next six to 12 months?</strong></p><p>We’re planning to launch a data governance initiative called the <a href="https://techmatters.org/the-better-deal-for-data/">Better Deal for Data</a>. Nonprofits and most businesses want to do the right thing with the data of the people they serve, and that doesn’t involve selling it to Meta or similar companies.</p><p>The Better Deal for Data is based on eight simple principles, modeled after the open source initiative, but for private data. The core concepts are straightforward: It’s your data, not ours. We’ll delete it, correct it, or transfer it somewhere if you want us to. If we hold it, we’ll secure and encrypt it. We won’t sell it. We’ll use it for your good and the good of the planet. If we do research with it, we’ll give you open access.</p><p>I’m really excited about this because nonprofits are looking for something like this. We’ve spent about two years looking for a lightweight approach to private data. Yes, we have regulations like GDPR in Europe, which are great, but this goes beyond that. We soft-launched it at Oxford this spring, and now we’re gathering feedback. We’re seeing a lot of excitement because data is important, and existing data sets are not very diverse.</p><p>We have to do something different, not only on the tech infrastructure, which is my traditional area of work, but also on the legal and social infrastructure — what are the norms, what’s an alternative to surveillance capitalism? I’m not trying to overthrow surveillance capitalism, I just think we need an alternative for mission-based organizations and companies that would rather be trusted by their users than selling them out to third parties.</p><p><strong>And is the plan to make that an international initiative?</strong></p><p>Definitely. One of our team members was in Kenya last week, gathering use cases from about 25 people on the ground, ranging from farmers to local leaders. Our work on data governance was actually inspired by a conversation I had five years ago in Tanzania, where someone explained data colonialism to me. As a guy from Silicon Valley, I was dismayed that I’d never heard the term. “You should look this up, Jim!”</p><p>When I looked it up, one of the scary things was that while we’re used to for-profit companies and governments being somewhat colonial, what was surprising was that the non-profit sector and academia are also very colonial in their approach to data. Part of the Better Deal for Data is not just about avoiding colonial practices, but ensuring data gets used for the benefit of the people it’s collected from. I’ve seen people take data about poor people and sell it to the fintech industry so they can make them poorer. I love data and I but I don’t think that’s the best use of data. That’s not the kind of use I want to be promoting.</p><p><em>If you’re a nonprofit interested in partnering with Whitespectre as part of our Green Grants, please </em><a href="https://www.whitespectre.com/green-grants/"><em>get in touch</em></a><em> — and let us know how we can help!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b722fb8576bc" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/behind-the-product-how-tech-matters-is-bringing-silicon-valley-to-social-good-b722fb8576bc">Behind the Product: How Tech Matters is Bringing Silicon Valley to Social Good</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mastering OpenSearch at Scale: A Practical Guide for Enterprise Search Solutions]]></title>
            <link>https://medium.com/whitespectre/mastering-opensearch-at-scale-a-practical-guide-for-enterprise-search-solutions-403450bf6eb9?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/403450bf6eb9</guid>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Wed, 15 Jan 2025 15:22:02 GMT</pubDate>
            <atom:updated>2025-01-15T15:22:02.073Z</atom:updated>
            <content:encoded><![CDATA[<h4>When queries take too long, they block real-time business decisions. We share architecture and optimization patterns drawn from real warehouse management systems processing millions of records.</h4><p><em>By Thales Kenne, Frontend Developer at Whitespectre</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ItGuqqIt3FYo7Vky-6v7MQ.jpeg" /></figure><p>Fast and effective search and analytics capabilities are critical to how businesses operate and make decisions. OpenSearch provides the scalability needed for searching millions of records while maintaining fast response times. We’ve proven this at scale, implementing it for Ruggable’s warehouse system tracking thousands of products globally. This guide shows you our proven indexing strategies and query optimizations, with practical examples from e-commerce and warehouse systems tested in production.</p><p>In this article, we cover:</p><ul><li><a href="#17df">Why OpenSearch?</a></li><li><a href="#dcf9">Real-World Applications</a></li><li><a href="#5f8f">Core Architecture</a></li><li><a href="#3be7">Hands-on Example: Building a Bookstore Search</a></li><li><a href="#06cd">Understanding Search Relevance</a></li><li><a href="#5b93">Query Types: A Practical Guide</a></li><li><a href="#0163">1. Match Query</a></li><li><a href="#fc49">2. Match Phrase Query</a></li><li><a href="#7cd0">3. Multi-Match Query</a></li><li><a href="#1e65">4. Term Query</a></li><li><a href="#7e0d">Query vs. Filter Context</a></li><li><a href="#4aa7">Advanced Implementation: Complex Queries</a></li><li><a href="#c874">Performance Optimization: Lessons from Production</a></li><li><a href="#0aa2">Conclusion</a></li></ul><h3>Why OpenSearch?</h3><p>OpenSearch shines when your system needs to search through millions of records, like server logs, product catalogs, or customer data, and return results in milliseconds. At its core, OpenSearch is a distributed search and analytics engine designed to handle large-scale data with remarkable efficiency.</p><p>If you’ve been in the search space before, you might wonder: “Why OpenSearch and not Elasticsearch?” This stems from recent changes in Elasticsearch’s licensing model. Elasticsearch was once the go-to tool for open-source data search. That changed when Elastic updated its licensing, moving away from open-source principles. In response, AWS forked the last open-source version of Elasticsearch and birthed OpenSearch, a fully open-source project. While the two tools started from the same foundation, they’ve since evolved independently, with distinct features and roadmaps.</p><p>When evaluating both options for our client implementations, we chose OpenSearch for these key reasons:</p><ul><li>Deep AWS Integration: Native AWS service integration reduces deployment complexity.</li><li>True Open Source: Apache 2.0 license ensures long-term availability and community access.</li><li>Community Innovation: Active contribution framework for custom feature development.</li><li>Enterprise-Grade Support: AWS backing while maintaining open-source flexibility.</li><li>Cost Optimization: Predictable AWS pricing with built-in infrastructure synergies.</li></ul><p>Beyond its licensing and community, OpenSearch offers deployment flexibility. You can opt for a managed service in the cloud, like Amazon OpenSearch Service, for a hassle-free experience where AWS handles the operational heavy lifting. Alternatively, an on-premises deployment provides greater control over your data and infrastructure, which might be necessary for organizations with specific compliance needs. The choice depends on factors like budget, technical expertise, and the level of control required.</p><h3>Real-World Applications</h3><p>At Whitespectre, we’ve leveraged OpenSearch to build large-scale, high-performance search solutions for our clients, including a comprehensive warehouse management system for Ruggable, a leading home goods manufacturer. This system manages inventories across multiple global warehouses, handling complex searches across materials, orders, and locations, resulting in improved operational efficiency and reduced order fulfillment time. While we’ll start with a straightforward example to establish core concepts, we’ll build up to the complex querying patterns we use in production systems like these.</p><p>Common applications include:</p><ul><li>Data Exploration and Visualization: Build interactive dashboards analyzing structured data (sales records) and unstructured data (customer reviews) with sub-second response times.</li><li>E-Commerce Search: Enable product discovery through typo-tolerant search, synonym matching, and relevance tuning, directly impacting conversion rates.</li><li>Log Analytics: Centralize and search millions of log entries in real-time, with automated alerting on error patterns and performance anomalies.</li></ul><h3>Core Architecture</h3><p>OpenSearch’s architecture is straightforward from an end user’s perspective, though it implements complex distributed systems architecture internally. The key concept to understand is indices, which act as OpenSearch’s databases. An OpenSearch index stores data in a JSON-like format, while more complex structures like Shards, Nodes, Clusters, and Inverted Indices work behind the scenes to ensure fast, reliable searches with built-in redundancy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KlgLXj7gmt-bHELU5VCFOQ.png" /><figcaption><em>Image Source: </em><a href="https://community.aws/getting-started-with-opensearch-models/introduction-to-opensearch-models"><em>https://community.aws/getting-started-with-opensearch-models/introduction-to-opensearch-models</em></a></figcaption></figure><h3>Hands-on Example: Building a Bookstore Search</h3><p>We’ll build a multi-store bookstore search system, starting with basic inventory management and adding store-specific search capabilities. First, we’ll populate our system with sample data in OpenSearch’s JSON-like format. Here’s our initial dataset of four books:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/358/1*20BhVOgujxjvOxmHDz8o6Q.png" /></figure><h3>Setup</h3><p>Setting up OpenSearch does not require a lot of boilerplate work. AWS provides templates that are easy to follow and will give configuration out of the box that will work for most people. Of course you can tweak all the configuration as you please, such as Node Availability, access control through IAM, instance types, the sort.</p><p>For our tutorial, we can run their <a href="https://docs.aws.amazon.com/opensearch-service/latest/developerguide/gsgcreate-domain.html">example</a> setup which uses simplified authentication with master password.</p><p>Side note: Make sure to select t2.small.search node instance when creating your service. That node is covered by the free tier and the tutorial baits you into creating a paid node.</p><p>Setting it all up won’t take more than 5 minutes. We just need to wait up to 30 minutes until our domain is up and running. After that, we’re ready to populate our first index!</p><h3>Populating Index</h3><p>Remember that json with books we just created? Let’s format it to create the payload that OpenSearch expects.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/642/1*zK4jhmPTAhXN7xQhIJxuJw.png" /></figure><p>From here, we have two options to populate the index: We can make a request (even through curl, if you want) to our domain or go to the dashboard. I personally prefer the dashboard, I feel more control over the data from there, so we’ll go with this option.</p><p>Navigate to your test domain and you’ll open your OpenSearch dashboard. Now navigate to DevTools. In there we can set the request type to POST _bulk and just paste our payload underneath it. Click the play button to the side and we are ready to start querying!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/602/1*y1YSONnIxHqQIk-mJqhkTQ.png" /></figure><h3>Query Types: A Practical Guide</h3><p>Now that we are ready to run these queries, we can either make requests from our app, which you’ll probably want for your production environment, or run them in the dashboard for testing. We can use the same DevTool dashboard where we just populated our index to run the queries below.</p><h4>1. Match Query</h4><p>Think of a Match query as your everyday search bar. It breaks down what the user types and looks for similar words in your documents that match any of the string’s terms. For example, if a customer types “Dark Tower”, Open Search will find any documents that meet “Dark” OR “Tower” and give them a relevance score.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/291/1*d9jg_01926kpByci7YqeDg.png" /></figure><p>For briefness, we will not include the query results for all of the sample queries in this tutorial, but this is what the response looks like for the query above. The response structure for all queries is always the same.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/336/1*yx2WEQeKu3ws89daFXHw6A.png" /></figure><h4>2. Match Phrase Query</h4><p>This one’s more strict — it looks for words in exactly the order they’re typed. If someone searches for “Dark Tower,” it’ll find your book, but “Tower Dark” won’t return any results. It’s essential for matching exact phrases like book titles or product names where word sequence affects meaning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/250/1*uAk-7f2Ntjm3tylRQ15D0A.png" /></figure><h4>3. Multi-Match Query</h4><p>Sometimes you want to search across multiple fields at once. Maybe your customers should find “White” whether it’s in the title or author name. Multi-Match lets you do exactly that. In our sample bookstore, this query will return two books. It matches the author Loreth Anne White and the title The White Tiger.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/307/1*v5ue1a1zvd0LSYz7XbnPvA.png" /></figure><h4>4. Term Query</h4><p>When exact matching is crucial — like searching for those ISBNs in our bookstore — Term queries are your friend. They don’t try to be clever with analysis; they look for exact matches. This would return only The White Tiger in our bookstore, since that’s our only book for store 456.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/239/1*twlxC7vA7FzLv8SBsi-bwg.png" /></figure><h3>Understanding Search Relevance</h3><p>When you search in OpenSearch, each result gets a “Relevance Score” — think of it as a matching percentage. The higher the score, the better the match to your query. This is what we call “Query Context.” For example, if you search for “White” in our bookstore database, you’ll find “In the Dark” by Loreth Anne White and “The White Tiger” by Aravind Adiga. Here’s where it gets interesting: you can decide whether matches in titles should score higher than in author names, tailoring the search to what your users value most.</p><h3>Query vs. Filter Context</h3><p>While the Relevance Score determines how well something matches, another important concept is “Filter Context.” Instead of asking, “How well does this match?” filters simply ask, “Does this match?” — a yes or no answer. Filters are great for narrowing down your dataset before running more complex queries, significantly improving performance. For instance, if you’re searching for “White” in our bookstore, you might want to filter for books only in a specific store or with stock available.</p><h3>Advanced Implementation: Complex Queries</h3><p>Our little bookstore is growing, isn’t it? We now have a few stores open, and the searches are becoming increasingly complicated. Say a customer wants to find books with “White” in the title or author name, but only in store123. Here’s how we’d write that:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/353/1*7xKrlDzwOYvbM_4MDrzpjg.png" /></figure><p>This query combines several concepts:</p><ul><li>Searches for “White” in titles and authors</li><li>Filters for uuid123</li><li>Returns results ordered by relevance</li></ul><p>Notice that the query above would not return The White Tiger, even if it has “White” in the title. That is because we filtered by store 123 and that title is only in store 456.</p><p>Boolean queries can use these parameters:</p><ul><li>must: All specified queries must match</li><li>should: At least one query should match</li><li>must_not: Excludes documents that match this query</li><li>filter: Like “must” but doesn’t affect scoring, making it faster</li></ul><p>Looking at our sample data, let’s try one final query:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/287/1*mXnk903BfxTs20h4F336Eg.png" /></figure><p>Can you guess which books this query would return? That’s right, it would give us “In The Dark” and “Dark Matter” but not “The Dark Tower”. The beauty of boolean queries is how they let you create complex search logic with straightforward, readable syntax.</p><h3>Performance Optimization: Lessons from Production</h3><p>Having scaled from basic bookstore searches to warehouse systems processing millions of daily queries, we’ve identified key performance patterns:</p><p>Filter Usage: Implement filters before relevance calculations to leverage result caching and boolean logic, reducing query time by up to 60%. Here’s what this looks like in practice.</p><p>Non-optimized query:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/386/1*0Q9_GXXlXg1taD0hUj6few.png" /></figure><p>Optimized query with filter:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/374/1*KbcnNVCqxC_wlMUX8T8v8w.png" /></figure><p>Beyond query optimization, we’ve found these practices essential for maintaining performance at scale:</p><ul><li>Proper Sharding: Match your shard size to your data volume. Too many small shards create overhead, while too few large shards limit parallelization.</li><li>Index Aliases: Use aliases to swap between indices during reindexing, preventing downtime during data updates.</li><li>Health Monitoring: Track query latency and cluster health metrics to catch issues before they impact users.</li><li>Data Lifecycle: Implement index lifecycle policies to manage growing datasets efficiently.</li></ul><h3>Conclusion</h3><p>These patterns scale from basic product searches to complex warehouse management. Building search functionality is iterative: start with simple match queries to establish a baseline, add filters where monitoring shows performance bottlenecks, and use real query patterns from your users to guide optimization. Every deployment will face unique challenges, but focusing on your core use cases and measurable performance metrics will keep your implementation on track.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=403450bf6eb9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/mastering-opensearch-at-scale-a-practical-guide-for-enterprise-search-solutions-403450bf6eb9">Mastering OpenSearch at Scale: A Practical Guide for Enterprise Search Solutions</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Behind the Product: How Keybotic’s AI-Trained Robot Dog is Changing an Industry]]></title>
            <link>https://medium.com/whitespectre/behind-the-product-how-keybotics-ai-trained-robot-dog-is-changing-an-industry-fb6ef998d074?source=rss-a894252ac1d1------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb6ef998d074</guid>
            <dc:creator><![CDATA[Whitespectre]]></dc:creator>
            <pubDate>Wed, 15 Jan 2025 14:11:36 GMT</pubDate>
            <atom:updated>2025-01-15T14:11:36.743Z</atom:updated>
            <content:encoded><![CDATA[<h4>Irene Gomez Alemany, CEO and Co-Founder at Keybotic, explains how they identified the market gap for their autonomous hazardous inspection robot — and their Robot as a Service model.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NqFfF_HTOcQlysquqfAhWg.jpeg" /></figure><p>You know how it goes: One day you’re studying international tax law; the next, you’re creating a company that builds autonomous robot dogs capable of entering hazardous environments and preventing industrial accidents.</p><p>Not a common story perhaps, but it’s the extraordinary career path of Irene Gomez Alemany, CEO and Co-Founder at Barcelona-based Keybotic, recently chosen as one of just five startups from across Europe invited by the European Commission to present in Brussels.</p><p>Irene sat down with us to explain how she, along with partner Hilario Tome, co-founded a robotics company that’s filling a neglected market gap, serving a genuine customer need — making industrial plants, and their employees, much safer in the process. She told us why they chose to bet on themselves, competing opposite the best and brightest MIT has to offer to win 1st prize in the DARPA Robotics Challenge — and how a $1 million cash prize helped them found their company and bring to life their first robotic dog, Keyper.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qOIGcDL8Ow_EBT1BlC-o8A.jpeg" /><figcaption>Keyper, the robotic dog enhancing safety. © Keybotic.</figcaption></figure><ul><li><a href="#5f07">How Irene’s tax and compliance background set the stage for her robotic start-up.</a></li><li><a href="#fd24">The importance of hiring the right people</a></li><li><a href="#979b">That market gap that drew Irene to robotics</a></li><li><a href="#43e7">The high-risk, high-reward gamble of DARPA competition</a></li><li><a href="#47f0">Building the MVP during COVID</a></li><li><a href="#88aa">How AI’s teaching robot dogs to walk — and jump!</a></li><li><a href="#5c47">Taking Keyper to market: the business model, sales, and implementation</a></li><li><a href="#e163">The importance of sustainability in robotics</a></li><li><a href="#ca72">Lessons for tech founders and IoT developers</a></li></ul><p><em>We start our conversation looking back on Irene’s interesting background in law, economics, and taxation to find out how she transitioned into robotics — and how she still applies the lessons learned in her past lives as a CFO and COO.</em></p><h3>Tell us about your background and how that led up to you founding Keybotic.</h3><p>The truth is I’ve always loved companies and finding ways to help them.</p><p>I actually studied law. That was my first degree and then I studied economics. After that, I started studying taxation and earned a master’s degree. From there, I moved to international taxes because that was the most complex part and I loved it.</p><p>It meant understanding different countries and how they worked. So not coming at it from just a national, domestic point of view but with international law so you know how things are taxed internationally.</p><p>That’s actually when I realized I liked business more generally, not just tax law. I liked the big picture of how companies worked. By that time, working as a strategic consultant, I had a lot of international customers — most of them startups and tech companies.</p><p>One of them, ALLOut Security, pushed me to go work with them, so I joined as CFO. However, two weeks later I became the CCO. Nobody was doing it so I took the role. And then, in six months, I was the COO because nobody was managing the business so I said “Okay, I’ll do it!”</p><p>I’d been on the board of directors for a few years before taking these roles, so I knew how everything worked — it’s not that I just flew in! I was at ALLOut Security for four years and I loved it very much.</p><p>At Keybotic, I’m doing the same type of work, but this time I’ve built it from the ground up and entirely in my own way. Instead of adapting something that already existed, I created the kind of company I always wished I could work at — that’s what Keybotic is.</p><h3>You make a difficult, complicated thing — founding an award-winning robotics company — sound extremely simple! What did you learn in your previous roles that influenced how you built your own company?</h3><p>The most important lesson I’ve learned is how essential employees are to a company’s success. In the past, I’ve often faced challenges that could have been avoided with the right team in place. So, at Keybotic, we have a very thorough interview process. For some positions, this has meant taking up to six months to find the right candidate.</p><p>We might have interviewed 200 candidates. We get a lot of applicants, and it’s not just me choosing the candidates, we probably take 3–5 people through the process. And we’re not just looking at their CV or the knowledge they have but their attitude.</p><p>We do a questionnaire and go through exams before we bring people in for an interview, but even if they score 10 out of 10 on the test, they don’t get in if we’re not convinced by their attitude.</p><p>I always want to go to work on Mondays. That’s my dream. And I think that’s a very important part — possibly the most important part — of a company because, at the end of the day, a company is just a bunch of people doing something together, right? What differs is what that thing is, but it’s ultimately just a bunch of people doing something.</p><p>So who this bunch of people is is really important.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KfOWNaRrOH0OL812fB5yeA.jpeg" /></figure><h3>So why did this bunch of people get together to focus on robotics?</h3><p>That’s a fun thing to talk about because my partner, Hilario [Tome, Keybotic’s CTO and Co-Founder], comes from creating humanoid robots, the first ones that were commercialized worldwide back in 2006–20 years before Tesla!</p><p>Basically, we noticed a gap: while most of us use automatic vacuum cleaners at home, similar automation hasn’t reached industrial settings. For instance, in chemical plants, automation is minimal, because robots have traditionally struggled to navigate complex environments, such as those with staircases and slopes.</p><p>We realized that a four-legged robot could overcome these challenges. However, to provide real value in this context, the solution must also be fully autonomous. There’s no return on investment if you have to babysit the robot.</p><h3>So you identified a gap in the market. How did you start addressing it?</h3><p>We started with the automation part and realized that, in the U.S., there’s the DARPA Robotics Challenge. It’s a big robotics challenge with enormous impact worldwide — autonomous cars, for example, were conceived in the first competition.</p><p>We took part in the third one, which was about robots finding people and objects in a 12-kilometer-long cave tunnel underground in place of police officers, soldiers, and firefighters who usually have to go in. We said, “Wow, this is exactly what we want to do.” We want robots that can enter a hazardous environment by themselves, find out whatever has to be found out, and come back, right? That’s what we have to build for the industry. <br>So we participated in 2020 and we won an intermediate prize. That’s when we registered the business. We kept working on the development of our first robot and, in 2021, we won the final prize of the challenge, which was worth around $1 million.</p><p>We invested this money fully into Keybotic, and that’s how we were able to further build the company into what it is today.</p><h3>You’ve mentioned in other interviews that entering the DARPA competition was a high-risk, high-reward strategy versus looking to get funding a traditional way. What were the risks involved?</h3><p>It was extremely high risk, high reward. Hilario worked super hard on the project for several years on his own. It’s not common to win in a world like that working alone!</p><p>We were competing, for instance, against MIT, which had a 40-person super team. I think it was ultimately his belief he could win that made him win, because if I’d known MIT was there I might’ve thought “It’s impossible, so let’s find another way to finance the company.” But he was sure that he’d at least win the third prize.</p><p>At that stage, nobody believed in us. Nobody would’ve thought we could compete against the big corporations we’re competing against nowadays. But we won!</p><p>So, besides financing, that’s what we achieved by winning this DARPA challenge: people started believing in us.</p><h3>What did those two years working on that project alone look like, without a team and support? Was Hilario just holed up in his room working on it?</h3><p>In a way, we were lucky that it was COVID time, because there was no social life. So Hilario worked every day from 10 am to 4 am. Every day.</p><p>We’d just bought a little Nissan Micra and, one day, he was driving back from the office at 4 am when the police stopped him. He must’ve seemed crazy, with Flamenco music blasting out the car really loudly so he didn’t fall asleep.</p><p>So they stopped him at a petrol station thinking he must have been out partying and he just told them “I’m working on a robotics competition for the US government!!”</p><p>That was kind of the atmosphere at the time. He was obsessed with the project.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h18k-EEZq1JZf0xkG5TbyQ.jpeg" /><figcaption>Hilario Tomé and Irene Gómez, co-founders of Keybotic.</figcaption></figure><h3>What did it look like at the MVP stage, with DARPA, versus what you have now with Keyper, your flagship robot?</h3><p>Well, we didn’t have any real robots at the competition — it was about the software, the navigation software that makes the robot navigate autonomously. What we had at DARPA was the first stage of what we have right now — it can do much more now, like working in larger spaces rather than just the confined space we had to work with for the challenge.</p><h3>Back when you were taking part in competition in — 2020, 2021 — the current AI boom wasn’t happening but it’s now part of your product. How has it helped you develop Keyper?</h3><p>For us, AI is just a technique to make things faster and better. We use it in many, many ways that you won’t even realize.</p><p>One of the main ways we use it is to help the robot understand the environment, for instance so the robot can read a gauge and understand if it’s outside a particular, expected range.</p><p>We were already using AI even back when it wasn’t fashionable to say we used AI. For example, we’d already been working on the robot being able to teach itself how to walk using AI for a year and a half before the AI conversation really accelerated.</p><p>Now, the new advancements are allowing us to work in a completely different way, with huge simulations running with thousands of robots walking.</p><p>While before Hilario was typing out various algorithms — math, it’s pure math — we can do things that math is not able to with AI, like teaching robots to walk when working in a plant with oil on the floor, with water, with slippery things, with ice.</p><p>We can use AI to train the robot to balance itself even on a moving floor. These are great things for us, because they enable our robot to work in very harsh environments.</p><p>So we use AI mainly for locomotion and perception, so our robots actually understand the environment. For example, now they can tell if an employee has a helmet on or not, as well as reading temperatures and things like that.</p><h3>When they’re learning how to walk in particular scenarios, is each robot learning on its own? Or does one robot learn how to walk on ice and then you kind of share that across all of the robots?</h3><p>Once one of them learns how to walk on the ice, it means that they’ve all learned it. So it’s always improving. They’re learning to walk with just two legs or even one leg. Or no legs when they’re jumping.</p><p>When you have a simulator, you can put certain constraints on them and get them to solve the problem. For instance, put them in the middle of a staircase and ask them to turn around. And then you increase the complexity of the problem. And once one of them does it right, it means that you have the right algorithm.</p><h3>A lot of people will think of Boston Dynamics when they see a four-legged robot walking around like Keyper does. What differentiates what you’re doing from the sort of viral robotics videos they’ve shown off in the past?</h3><p>Put simply: without them, we wouldn’t exist. They were the pioneers in robotics back in 1986 and they’ve created a superlab where they’ve developed so many impressive things, which ultimately help us incredibly in terms of marketing.</p><p>We wouldn’t exist without them because we didn’t have the resources to let people know that this type of product exists. So when people see a four-legged robot from Boston Dynamics that can dance, it’s great for the whole robotics industry.</p><p>What we add to that is that our robot has a final application for industry customers.</p><p>Our value proposition is that we provide a service to customers, and that’s actually why we also rent Keyper, we don’t just sell it. Our business proposition is saying that we’ll improve the process you have in your chemical plant, for example.</p><p>We’ll detect gas leakages in a way you couldn’t before. We’ll automate and digitize all your plants, and you’ll be able to monitor what’s happening from the control center. Until now, a person had to go around and check these things manually.</p><p>It’s so big that you can do that. If you wanted to do something like this before you’d have to have sensors everywhere, which isn’t possible across a big industrial plant. So we’re a mobile sensor platform. And, with artificial intelligence, we can actually detect anomalies in those plants.</p><p>We haven’t created the best robot in the world: We’ve created the robot that works in that environment and for that purpose. It’s cost-effective, and it’s modular so it’s very easy to transport. Everything’s integrated, even though you can add extra sensors on top, which aren’t ours.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*adzQacFhWEzh54L8oscWRg.jpeg" /></figure><h3>What made you decide to go down this road of the Robot as a Service (RaaS) model rather than only selling units?</h3><p>It’s a way to have zero barriers for customers to try it. So we want customers to try it and that model makes it very easy for us to establish a collaboration: they know we’re with them for the success of the project.</p><p>It’s not that we sell it and exit: We’re with them throughout the whole process.</p><h3>How do you think being a European company — a Spanish company — makes you different from if you were based in Silicon Valley with this concept?</h3><p>The truth is that it allows us to have European customers that otherwise we wouldn’t. Because of the data protection laws, we have to comply with the laws and certifications in Europe, which the other companies cannot always do.</p><p>So that’s why you can’t really find many robots from Boston Dynamics working in Europe, for example.</p><p>And then there’s the simple fact that we created the company in Barcelona because we wanted to! Hilario and I wanted to be home and we knew a bunch of good people that could start with us.</p><p>As I mentioned earlier, the first employees are the most important ones and actually we got the best of the best. If we’d gone to the US, even if we would be better financed, we wouldn’t have the people we want because we’d be competing against companies with 10x the funding.</p><p>And we’re very globalized so next year we could go to the US and start there too. That’s not a problem. I think it’s more complex to come to Europe because you need to understand how different cultures work across different countries.</p><p>The markets are smaller than in the US and that also made us create a robot that’s very cost-effective because otherwise, we wouldn’t have been able to start in Spain, where the budgets are lower, for example.</p><h3>What are the main problems you face getting quite conservative industries to embrace new technology?</h3><p>Already from the beginning it actually wasn’t very difficult when I cold called because we’d already gone viral within the industry and customers wanted to see what we had. They wanted to understand what was happening in the environment.</p><p>It’s not like selling software, for example, which is hard because there are lots of alternatives. In our case, they wanted to know what’s happening in the world of robots. They want to talk to you even if they don’t want to buy, which is a huge advantage.</p><p>They want to know what their neighbors are doing. So they answer 86% of the cold calls we make.</p><h3>What sort of impact have you seen in those hazardous environments where Keyper has been deployed so far?</h3><p>Right away we were doing a proof of concept and we detected a gas leak in a big plant that they didn’t know about — in an explosive environment.</p><p>This industry is very interesting because, when you go to a customer, they never want to tell you how many accidents they have. It’s not until you talk to them the third or fourth time.</p><p>We have a customer that has 366 leaks every year. Not all of them are dangerous, but when we got there, the previous week, two people were in the hospital because they thought a leak was water and it wasn’t.</p><p>So accidents happen in those environments. And what you want is to decrease the ratio. We’re detecting them earlier and in a way where people don’t have to go into hazardous situations to do it.</p><p>We also help them decrease costs.</p><p>For example, at one of the companies we work with, because the robot could take thermal images, we could prove that the plant had no leakages and it made a big audit pass in a single day.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Oz3jqT573C5QVuBpsDgncg.jpeg" /></figure><h3>So there’s savings in compliance too?</h3><p>Yes! In that particular case, because of the thermal images they had from all the plants, they could show that there were no problems and the customer was super impressed. Previously there wasn’t any way you could have a thermal map of an entire plant.</p><p>We actually increase safety standards in cases where the regulations aren’t there.</p><h3>What approach do you take around change management in the organizations you work with, because it could present a big shift in ways of working?</h3><p>Right now it’s not about reducing the number of people, even if that’s a long-term factor for our customers. Right now people who would have been monitoring the plants manually are redeployed and can do something else.</p><p>There’s a lot of work to do across these huge plants, so we take over the low-value activities they’re performing right now. Because making rounds 2–3 times a day to check gauges manually and then write a number on a piece of paper isn’t a great use of a person’s time. And the robot can actually do these jobs better.</p><h3>What limitations does Keyper still face in the scenarios you’ve described?</h3><p>So what we have right now is a passive robot. It goes around, but it does nothing other than report issues — it can’t take direct action to solve an issue. We’re trying to change that.</p><p>And we’re creating our second robot too, which is IP67 — which means it’s waterproof and dust proof. This means that it can go into very harsh environments.</p><p>Right now, where there’s a lot of dust and pollution in the air, or salt by the sea, it presents challenges to the robot. The new one will be able to function in those environments much better. And we’ll get our explosion proof certificates soon, so we can go into explosive environments.</p><h3>You’ve mentioned the importance of sustainability in robotics. What’s been your approach to environmentally friendly design?</h3><p>So we always need to balance sustainability and also cost effectiveness. The aim is always to make both things work at the same time.</p><p>Because the robot is modular, we can reuse all of it. So imagine that after three years, the computers don’t work anymore, because, like your computer, we’ve used it so much that it’s old.</p><p>Then, rather than build an entirely new one, we can reuse all the old robots for a new robot. So the legs are detachable, but you also have modules inside the robot. So you can plug and play them.</p><p>It’s not just about creating something in a way that’s environmentally friendly but it’s also about creating something reusable.</p><p>Everything we’ve designed for the robot has been designed to be as simple as possible.</p><p>And then the robot’s purpose is to detect gas and steam leaks in chemical environments very quickly, which is crucial for the environment.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FRzbUmvU04Xo%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DRzbUmvU04Xo&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FRzbUmvU04Xo%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/e64d9a852b327d5d5361863071449685/href">https://medium.com/media/e64d9a852b327d5d5361863071449685/href</a></iframe><h3>What fresh approaches did you bring to this industry coming from a non-traditional background that somebody wouldn’t have had if they’d spent their whole career in robotics?</h3><p>I think that there’s one important thing that I experienced before, which is the importance of easy implementation.</p><p>In my previous role, our software didn’t have any sort of complex implementation. From day one, from minute one, you could start using it, while competitors had a huge amount of implementation, consultancy, months and months of training — it was very complex.</p><p>In our robotics solution, it was important to me that there’s zero implementation.</p><p>The robot starts working from minute one. That’s very complex to reach and that’s why most companies don’t have that.</p><p>Right now we have videos to help our clients with everything from how to get the robot outside of the box to making it work, but in reality, if you click the button it starts. There’s no implementation required. The robot is very robust. Right now we go with them and show them how it works, but that’s about a client-customer relationship more than anything else.</p><p>It works right out of the box because you simply remote control operate it the first time so it can create a 3D map of the entire environment. You digitize the whole plant with that. And then you show the robot what has to be checked. For example, “now we go through these gauges. Stop here.” And then you stop the robot. The robot detects the gauge. And then you select the objective for that mission that the robot will follow in the future.</p><p>It’s super simple. That’s all the customer should do the first time. Right now, we help them do it. But it’s super simple to do and, in the future, we envisage a post-sales process where the customer can simply open the box and press start.</p><h3>Finally, what are the broad lessons you’ve learned with Keyper that might apply to other IoT products outside of robotics?</h3><p>I’d say, don’t try to create the best product from a technical perspective, but the product the customer actually wants. It’s difficult because all engineers want to create the best product. You have to stop them!</p><p>In our case, you always have to stop and say “how does this functionality help us reach our objective?” If our team wants to make the robot jump, for example, but the customer doesn’t want the robot to jump then we shouldn’t develop something because it’s cool or technically impressive.</p><p>If the answer is “we need to make the robot jump so it’s much more robust” then that’s ok, because you have a final application for customers. I think this is relevant everywhere because product owners want to create the best product and you have to be very close to the customer to understand what they actually need — and that’s the hard work.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fb6ef998d074" width="1" height="1" alt=""><hr><p><a href="https://medium.com/whitespectre/behind-the-product-how-keybotics-ai-trained-robot-dog-is-changing-an-industry-fb6ef998d074">Behind the Product: How Keybotic’s AI-Trained Robot Dog is Changing an Industry</a> was originally published in <a href="https://medium.com/whitespectre">Whitespectre Ideas</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>