<?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 Jon Innes on Medium]]></title>
        <description><![CDATA[Stories by Jon Innes on Medium]]></description>
        <link>https://medium.com/@innes09?source=rss-59b4bb0e5f62------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*-KjKtNoeHr2yQtWQMCiO6Q.jpeg</url>
            <title>Stories by Jon Innes on Medium</title>
            <link>https://medium.com/@innes09?source=rss-59b4bb0e5f62------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 22:12:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@innes09/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[If the Singularity Is Not That Near, What Is AI Really Good For?]]></title>
            <link>https://medium.com/@innes09/if-the-singularity-is-not-that-near-what-is-ai-really-good-for-da04927fd267?source=rss-59b4bb0e5f62------2</link>
            <guid isPermaLink="false">https://medium.com/p/da04927fd267</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[machine-learning-ai]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[future-of-work]]></category>
            <dc:creator><![CDATA[Jon Innes]]></dc:creator>
            <pubDate>Tue, 02 Dec 2025 15:52:30 GMT</pubDate>
            <atom:updated>2025-12-02T15:52:30.864Z</atom:updated>
            <content:encoded><![CDATA[<h4>What will move us toward a world where AI becomes a trusted, powerful, and positive force in our daily lives?</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DaqKIveXDqmilQ9fCb5RCA.png" /></figure><p>We are arguably beyond the point where AI can pass the long-standing test proposed by Alan Turing in his 1950 paper <a href="https://www.csee.umbc.edu/courses/471/papers/turing.pdf">Computing Machinery and Intelligence</a>. Seventy-five years later, in 2025, we have many examples of chatbots that can plausibly pass for humans in casual conversation. The “Can a machine imitate a person?” question has, in a practical sense, been answered.</p><p><strong><em>So what’s next?</em></strong></p><h4><strong>What AI Is Actually Good For in 2025</strong></h4><p>LLMs today are extraordinary in some ways and surprisingly mundane in others. They can hold conversations, summarize intricate topics, generate functional code, draft legalese, create images, and even to a limited extent reason through multi-step problems.</p><p>But <em>they are still weakest in areas that humans intuitively excel at</em> — physical perception, long-horizon planning, deep contextual understanding, and embodied interaction with the real world.</p><p>LLMs also tend fail to calculate accurately, are prone to fabrication (hallucinations), and have short-term memory issues (context window limits). Failings not shared with traditional computer programs.</p><p><strong><em>AI today is far better at manipulating symbols than understanding reality, or doing traditional computing.</em></strong></p><p>It’s important to be grounded in reality. We have to keep in mind what Geoffrey Moore calls “The Chasm” in his famous Product Adoption Curve where many tech innovations die an early death.</p><p>Symbol manipulation is the foundation of every digital system we’ve built: software, documents, databases, designs, simulations, and code. And if we look at where AI is actually driving meaningful impact today, one area stands above the rest:</p><p><strong><em>AI is becoming astonishingly good at helping us build new software.</em></strong></p><h4>Developer Tools Have Always Been the Hidden Accelerators</h4><p>Behind almost every major shift in computing over the past 40 years lies an inflection point in <em>developer tooling</em>:</p><ul><li><strong>Object-oriented programming</strong>, borrowed from Xerox PARC, enabled modular, large-scale systems and paved the way for personal computing at Apple and Microsoft. It made code reuse easier.</li><li><strong>IDEs </strong>like Visual Studio and Xcode allowed developers to build complex graphical applications like Microsoft Office with unprecedented speed.</li><li><strong>Content management systems</strong> like WordPress democratized the web — so much so that two decades later WordPress still powers 40% of all websites in 2025. Wordpress made website development easy.</li><li><strong>GIT,</strong> the distributed version control for managing source code. Developed by Linus Torvalds to support open source projects which were challenged due to the geographicly distributed team of volunteer contributors on Linux. It dramatically improved code collaboration.</li><li><strong>CUDA</strong>, originally developed for graphics processing in video games, allowed developers to finally take advantage of parallel architectures proposed in the 1980’s by researchers like computer scientist <a href="https://en.wikipedia.org/wiki/Danny_Hillis">Hillis</a>, and psychologists <a href="https://www.researchgate.net/publication/243672900_A_General_Framework_for_Parallel_Distributed_Processing">Rumelhart and McClelland</a>.</li><li><strong>Kubernettes</strong> developed by Google allows developers to easily deploy software to clusters and manage it, making parallel computing a near commodity and hyperscale computing possible — and LLMs.</li></ul><p>In every case, the breakthrough wasn’t just the technology — it was the tools that empowered people to use it.</p><p><strong><em>AI is now entering this lineage.</em></strong></p><h4>LLMs: Syntax Engines With Emerging Semantics</h4><p>For all the talk of intelligence, today’s LLMs are just astonishingly efficient pattern-matching engines. They excel at:</p><ul><li>suggesting syntax (boilerplate code, API usage, framework configuration)</li><li>identifying latent patterns learned from enormous datasets</li><li>natural language processing and reformulation</li><li>semantic search, which is rapidly replacing keyword-based retrieval</li></ul><p><strong><em>This may not sound revolutionary, but modern software development is dominated by configuration and wiring work — precisely the tasks LLMs are now excellent at.</em></strong></p><p>Developers spend significant time setting up databases, syncing cloud services, configuring authentication, wiring up front-end components, or orchestrating build pipelines. These tasks take up much of the time spent by developers building modern software, which makes LLMs surprisingly good at automating the grunt work.</p><p>And that matters, because…</p><h4>The Real Acceleration: AI That Helps Us Build More AI</h4><p>If the biggest catalysts of past computing revolutions were developer tools, then the implications of AI-native tooling are enormous. We might be in a reflexive moment where:</p><p><strong><em>AI helps humans build better software — including better AI — which then helps humans build even better software.</em></strong></p><p>This feedback loop may be the real engine pulling us toward the Singularity. Or at least faster software development…</p><h4>Why Design Now Matters More Than Ever</h4><p>As powerful as AI has become, it still fails when the human experience is neglected. Gavin Lew and Bob Schumacher put it succinctly in <a href="https://doi.org/10.1007/978-1-4842-5775-3">AI and UX </a><em>“Plenty of smart AI has failed because nobody thought about the person using it.”</em></p><p>Similarly, Akshay Kore emphasizes in <a href="https://a.co/d/2pQuBiT">Designing Human-Centric AI Experiences</a> that <em>“Designing for uncertainty, trust, and cross-team collaboration is not optional — it’s the work.”</em></p><p>And when AI systems act on our behalf, the mindset shifts again. Christopher Noessel describes agentive tools in <a href="https://a.co/d/bRDGGlI">Designing Agentive Technology</a> as <em>“Technology that quietly gets things done — without taking over.”</em></p><p>Finally, Chip Huyen writes in <a href="https://a.co/d/6iKwfju">Designing Machine Learning Systems</a><br> <em>“Scalable, resilient, and responsible ML products emerge from iterative, systems-level thinking — not one-off models.”</em></p><p>These four perspectives all point to a shared truth: the future of AI depends as much on thoughtful design and systems thinking as it does on model size or training compute.</p><h4>The Expanding Role of Product Designers in an AI-Driven World</h4><p>While engineers push the boundaries of what’s technically feasible, designers increasingly shape what’s usable, trusted, and socially acceptable.</p><p>One of the most valuable functions designers serve today is exploring the possible futures unlocked by emerging technologies — and giving them enough form that others can evaluate, debate, critique, and ultimately build them.</p><p><strong><em>Designers don’t just draw interfaces. They imagine the future in high fidelity and communicate that through prototypes, mockups and specs</em></strong>.</p><p>As AI becomes more capable, designers must:</p><ul><li>imagine how AI can augment (not replace) human work including collaboration</li><li>define reliable interfaces between human agency and machine initiative</li><li>prototype complex human–AI systems</li><li>reveal hidden risks</li><li>surface unarticulated opportunities</li></ul><p><strong><em>Designers have become translators between abstract technical capability and concrete human value.</em></strong></p><h4>So What Really Accelerates Us Toward the Singularity?</h4><p>Not fully autonomous agents.<br>Not emergent superintelligence.<br>Not machines spontaneously developing general reasoning.</p><p>The accelerant is far more grounded. It’s the degree to which AI helps us do things better. Including building new, better technology — faster, safer, and more importantly…better than any generation before.</p><p>The Singularity may one day arrive. But for now r<strong>eal progress depends on human experts that can leverage what AI can already do well today. AI assistance vs. fully autonomous agents for now, and in the near future.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=da04927fd267" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Entering the Age of the Intention Factory]]></title>
            <link>https://medium.com/@innes09/entering-the-age-of-the-intention-factory-82ad23a29527?source=rss-59b4bb0e5f62------2</link>
            <guid isPermaLink="false">https://medium.com/p/82ad23a29527</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[service-design]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[software-design]]></category>
            <dc:creator><![CDATA[Jon Innes]]></dc:creator>
            <pubDate>Wed, 19 Nov 2025 02:07:37 GMT</pubDate>
            <atom:updated>2025-11-19T02:07:37.099Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oiqYjZKNP51j5ebfyQakyQ.png" /></figure><p>Over a century ago, Henry Ford and Charles Sorensen revolutionized industrial production. Their innovations increased manufacturing throughput by over <strong>700% between 1913 and 1914</strong>, establishing the framework for modern industrial engineering. Today, software engineering looks to be on the cusp of a similarly disruptive transformation driven by <strong>large language models (LLMs)</strong>, <strong>vibe coding</strong>, and <strong>Spec-Driven Development (SDD)</strong>.</p><p>It’s been 14 years since Marc Andreesssen’s article <a href="https://www.wsj.com/articles/SB10001424053111903480904576512250915629460">Why Software Is Eating The World</a> was published in the Wall Street Journal. Marc, co-founder of Netscape and co-author of the 1st graphical web browser Mosaic just over 20 years ago helped usher in the dot-com boom. I read his thoughts carefully as I pondered my next career move in San Francisco at the end of the 2008 financial crisis. I had just bought a condo and had a vendor badge from the New York Stock Exchange on my chest as things unfolded.</p><p>Working in Silicon Valley I’ve been here to witness this incredible transformation from the ground zero perspective. I probably use that term because I arrived here after studying the intersection of cognitive science and computer science in graduate school in New Mexico not far from Trinity Site. I arrived there after studying business and psychology, paying my way through college working in the factories south of Detroit. They say wisdom is the ability to apply knowledge effectively, not simply the accumulation of knowledge…and <strong>wisdom in the age of AI might be all that differentiates humans from LLMs at this point.</strong></p><p>In the early 20th century, Henry Ford and Charles E. Sorensen transformed manufacturing from a craft practiced by specialists into a scalable system capable of delivering reliable products to millions. Their core insight — that complex things can be built from standardized, interchangeable parts through a carefully orchestrated flow of work — remains one of the most influential ideas in industrial history. <strong>What might be less obvious is how deeply this same philosophy might be applied to software engineering</strong>.</p><p>At Ford’s Piquette Avenue Plant, Sorensen experimented with the idea that a product could move while workers remained stationary. His now-famous “rope-and-chassis” test previewed the moving assembly line, which would soon revolutionize production at Ford’s new Highland Park plant. <strong>The auto assembly line was invented.</strong> There, instead of artisans crafting an entire car, the Model T became the result of hundreds of small, repeatable tasks executed with precision on the assembly line.</p><p><strong>Standardized parts and assembly lines transformed manufacturing.</strong></p><p>Manufacturing in an orchestrated fashion, while the product under construction is guided past the specialized workers, adding parts along “the line” to completion changed everything. Sub-assemblies of magnetos, engines, and axles were built separately and delivered ready to install on “the line”.</p><p>Standardized parts ensured a perfect fit. Quality checks were embedded throughout. <strong>The result was an explosion in productivity: Ford reduced vehicle build time from over 12 hours to roughly 93 minutes, lowering costs dramatically. </strong>Ford would produce close to 15 millon Model T cars.</p><p><strong>This transformation has a direct, almost one-to-one analogue in digital engineering in the AI era.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-AmQsku1xMcaY9PRTs7RBw.png" /><figcaption>Ford Manufacturing Innovation Compared to Software Development</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JgncYpc6S88QdnJ1tpj5jQ.png" /><figcaption>Design Systems, Design Tokens &amp; Sorensen’s Standardization</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ukzr5GNsjfilpclS5H7bnQ.png" /><figcaption>Assembly-Line to Software Equivalent</figcaption></figure><p>Where Ford standardized bolts, pistons, and body panels, today’s engineers standardize UI elements, API contracts, and software modules. Sorensen’s interchangeable parts live on in <strong>object-oriented programming</strong>, where classes and methods are designed to be reusable building blocks. Sub-assemblies map cleanly to <strong>microservices</strong>, <strong>RESTful APIs</strong>, and <strong>component-based UI frameworks</strong> like React, in which functionality is built in isolation and “bolted on” through well-defined interfaces.</p><p>Ford’s relentless focus on workflow and predictable task times echoes strongly in <strong>CI/CD pipelines</strong>, where code flows continuously from commit to build to test to deploy. Just as the moving line demanded early detection of defects, modern software uses <strong>unit tests</strong>, <strong>linting</strong>, and <strong>static analysis</strong> to prevent flaws from advancing downstream. The precision jigs and fixtures Sorensen developed for manufacturing are mirrored in <strong>test automation tools</strong> and <strong>Storybook component testing</strong>, ensuring that a UI component renders consistently across environments the same way a magneto had to function identically in every car.</p><p>Ford’s obsession with reducing variability also finds a modern parallel in design systems. <strong>What standardized parts did for mechanical manufacturing, design systems and tokens (now a formal W3C standard) do for digital design.</strong> They define a common, machine-readable vocabulary of colors, spacing, typography, and interaction patterns. Libraries like Tailwind or React component systems built from these tokens ensure that every digital “part” behaves consistently, just as Ford ensured every bolt matched every thread.</p><p>Even vertical integration at the scale of the River Rouge complex, where raw ore entered one end and a completed car exited the other, has its digital counterpart. <strong>Today’s cloud platforms and developer toolchains provide integrated environments where code, dependencies, builds, storage, compute, and deployment pipelines are tightly connected</strong> — an end-to-end ecosystem not unlike Ford’s “ore to assembly” River Rouge Plant.</p><p>Most tech workers today are unfamiliar with Sorensen’s wartime achievements at Ford. When applied to producing close to 650,000 Jeeps and later a B-24 bomber an hour during WWII, they demonstrated that the principles of modularity, process discipline, and automation were key to reducing costs while retaining quality. <strong>Could we apply AI to software engineering to achieve a similar leap in productivity?</strong></p><p>Take a complex problem. Break it into repeatable parts. Standardize the interfaces. Automate the flow. Iterate relentlessly. Scale indefinitely.</p><p><strong>The world of servers, APIs, React components, and design tokens may seem far removed from early automobile factories, yet the conceptual lineage is unmistakable. </strong>Ford and Sorensen industrialized the physical world; today’s engineers unknowingly apply the same principles to the digital world…</p><p><em>What’s </em><a href="https://sundaylettersfromsam.substack.com/p/where-is-all-the-collaboration-in?r=vzxne&amp;triedRedirect=true"><em>missing still is AI orchestration of the work</em></a><em>. </em>In the last few years since Co-Pilot and Claude arrived on the scene, we’ve seen AI-enabled coding.</p><p><strong>The next logical step is AI-orchestrated product development where the AI facilitates more than code generation. AI that learns your team’s patterns and proactively suggests improvements across disciplines.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=82ad23a29527" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>