<?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 Luke Ong on Medium]]></title>
        <description><![CDATA[Stories by Luke Ong on Medium]]></description>
        <link>https://medium.com/@lukeyish?source=rss-c5145130c473------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*tv2ZtYcG-2BIKjMdn8DQBg.jpeg</url>
            <title>Stories by Luke Ong on Medium</title>
            <link>https://medium.com/@lukeyish?source=rss-c5145130c473------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 21:23:27 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@lukeyish/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[The Death of “Vibe Coding”: Why Deterministic Agent Runtimes are the New North Star]]></title>
            <link>https://lukeyish.medium.com/the-death-of-vibe-coding-why-deterministic-agent-runtimes-are-the-new-north-star-ac31bf584b35?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/ac31bf584b35</guid>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Thu, 16 Apr 2026 03:05:36 GMT</pubDate>
            <atom:updated>2026-04-16T03:05:36.223Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Published: April 16, 2026 | Category: AI &amp; Data Engineering</em></p><blockquote>The era of stateless, prompt-engineered AI is ending. In its place, we are seeing the rise of stateful agent runtimes that treat memory as a metabolic process and governance as a hardware-level primitive.</blockquote><h3>1. The Landscape: The Collapse of the Stochastic Gap</h3><p>For the past two years, senior engineers have struggled with the “stochastic gap” — the distance between a promising LLM demo and a production-grade autonomous agent. While 2024 and 2025 were about context windows and RAG tuning, 2026 has signaled a fundamental pivot.</p><p>We are witnessing the final collapse of <strong>“Vibe Coding”</strong> — where performance is measured by subjective quality — in favor of <strong>Decidable-by-Construction</strong> systems. The competitive advantage is no longer the model weights, but the robustness of the orchestration and governance layers that manage agent state.</p><h3>2. From Stateless APIs to Metabolic Memory</h3><p>The most significant bottleneck in autonomous systems has been the <strong>Long-Horizon Task Mirage</strong>: a phenomenon where agents lose coherence as operational state accumulates noise.</p><p>Instead of treating memory as a static append-only log, we are moving toward <strong>Metabolic Memory</strong>. Just as biological systems prune neural connections to maintain efficiency, enterprise agents must now calculate the <strong>Memory Worth (MW)</strong> of every artifact.</p><ul><li><strong>The Shift:</strong> Moving from cosine similarity to tracking the conditional success probability of retrieved memories.</li><li><strong>The Solution:</strong> The “File-as-Bus” workspace. Specialized agents re-ground themselves on durable, evolving artifacts rather than fragile conversational handoffs.</li></ul><h3>3. The Verification Tax and Structural Correctness</h3><p>As failure rates drop, the mathematical effort required to verify a model’s output increases exponentially — a phenomenon known as the <strong>Verification Tax</strong>. If a senior engineer cannot trust the output without a manual line-by-line audit, the productivity gains vanish.</p><p>To combat this, we are adopting <strong>CODESTRUCT</strong>:</p><ol><li><strong>Restricted Action Space:</strong> Agents no longer “write code”; they operate directly over <strong>Abstract Syntax Tree (AST)</strong> entities.</li><li><strong>Structural Integrity:</strong> This ensures output is syntactically and structurally correct by design, eliminating “silent” hallucinations.</li><li><strong>Reasoning Graphs:</strong> Moving beyond query-response pairs to evidence-centric feedback loops.</li></ol><h3>4. Perception-as-a-Program (P²)</h3><p>The industry’s greatest lie was that LLMs could understand structured data by “flattening” it into text. 2026 is the year of <strong>Perception Programs (P²)</strong>.</p><p>Instead of force-feeding raw vectors into a model, tool outputs are rewritten into language-native summaries that models can parse with high precision. This is often paired with <strong>Order-Aware Hypergraph RAG (OKH-RAG)</strong>, allowing models to reason over complex relational dependencies — like supply chain ripples — that were previously invisible to text-centric architectures.</p><h3>Summary: Vibe Coding vs. Agentic Systems Architecture (2026)</h3><p>FEATURE”VIBE CODING” (LEGACY)AGENTIC SYSTEMS ARCHITECTURE<strong>State Management</strong>Stateless / ConversationalStateful / Metabolic Memory<strong>Action Space</strong>Unconstrained (Raw Bash/Python)Structured (AST-Native)<strong>Reliability</strong>Probabilistic (Prompt Tuning)Deterministic (Decidable-by-Construction)<strong>Verification</strong>Manual Audit (“The Tax”)Evidence-Centric Reasoning Graphs<strong>Data Interaction</strong>Flattened Text / Raw VectorsPerception Programs (P²)</p><h3>5. The Directive: What This Means for Your Architecture</h3><p>The transition from “Model-as-a-Service” to <strong>“Agent-as-Infrastructure”</strong> is not a suggestion; it’s a requirement for scale.</p><ul><li><strong>Standardize on AST-native interfaces:</strong> Stop giving agents raw bash access. Restrict them to structured action spaces over your existing APIs.</li><li><strong>Implement ProbeLogits:</strong> Do not rely on “system prompts” for safety. Use kernel-level monitoring to detect unauthorized intent at the logit level.</li><li><strong>Adopt Memory Metabolism:</strong> Re-evaluate your vector store. Implement a multi-cycle buffer where agents prune their own knowledge base to maintain a high signal-to-noise ratio.</li></ul><p><strong>The goal is clear:</strong> Reliability in 2026 is an architectural constraint, not a model capability. Stop tuning prompts; start building runtimes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ac31bf584b35" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Engineering “The Drop”: Simulating the High-Stakes World of Hype Retail]]></title>
            <link>https://lukeyish.medium.com/engineering-the-drop-simulating-the-high-stakes-world-of-hype-retail-461ec7e0f411?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/461ec7e0f411</guid>
            <category><![CDATA[retail]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Thu, 12 Mar 2026 12:22:12 GMT</pubDate>
            <atom:updated>2026-03-12T12:25:20.850Z</atom:updated>
            <content:encoded><![CDATA[<p>In the modern digital economy, success for a certain breed of retailer isn’t measured in years or months — it’s measured in seconds. From limited-edition sneaker releases to “viral” streetwear drops, the Hype Retailmodel has transformed commerce into a high-stakes game of artificial scarcity, algorithmic marketing, and infrastructure-straining traffic surges.</p><p>To understand this phenomenon, I built the Hype Retailer Workflow Simulator — a Streamlit-powered engine designed to model the entire lifecycle of a “Drop.” Here’s a breakdown of the mechanics, the math, and why simulation is the ultimate tool for domain mastery.</p><h3>What is Hype Retailing?</h3><p>Hype retailing is a business model focused on selling high-demand, limited-edition products. Unlike traditional retail, which prioritizes steady-state inventory and “always-available” stock, hype retail thrives on artificial scarcity.</p><p>The operational workflow is built around three distinct phases:</p><ol><li>Pre-Drop: Identifying “viral” signals on platforms like TikTok, engaging “hypebeast” influencers, and building massive SMS/email waitlists.</li><li>The Drop: A high-volume sales event where traffic can spike 100x in milliseconds, requiring sophisticated bot protection (Edge AI) and fair sales models (FCFS or Raffles).</li><li>Post-Drop: Managing rapid fulfillment and monitoring the secondary market (like StockX) to see the true “market value” of the goods.</li></ol><h3>Building the Simulator: The Math of “The Drop”</h3><p>To make the simulation feel real, I moved away from static numbers and toward dynamic, stochastic modeling.</p><h3>1. The Hype Equation (H)</h3><p>Hype isn’t a feeling; in this model, it’s a quantifiable score. The simulator uses an Aggregate Hype Scorecalculated by combining marketing investment, scarcity levels, and a “Curation Intelligence Score” (I).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*xJxyiDFOLUzQX1vPJvlsew.png" /></figure><p>We prioritized TikTok Velocity over Instagram aesthetics for intelligence signals because TikTok’s “For You” page provides a higher-fidelity signal for viral discovery — a key requirement for forecasting demand in 2024.</p><h3>2. Simulating the Traffic Surge</h3><p>During “The Drop,” traffic isn’t a steady stream; it’s a vertical wall. The simulator uses Stochastic Poisson Modeling to generate requests. If your waitlist is 1M people, the server needs to handle 5M+ requests in a matter of seconds.</p><p>The simulator also factors in Edge AI (Anti-Bot Protection). High-tier protection (Akamai-style) filters out bot traffic, ensuring units go to “real” humans, which preserves long-term brand health even if it adds infrastructure complexity.</p><h3>3. The Secondary Market Proxy (StockX Prediction)</h3><p>One of the most exciting features is the StockX Prediction Model. It uses a heuristic proxy to predict the resale premium of a product:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/734/1*HiBmr0ZKyXQylIYhEWYnYw.png" /></figure><p>This illustrates the direct correlation between the quality of your Pre-Drop curation and the eventual market desirability.</p><h3>Learning Through Simulation: Why It Matters</h3><p>Building and using simulations like this provides insights that reading a textbook cannot.</p><h3>1. Risk-Free Experimentation</h3><p>You can see what happens when you spend $50,000 on influencers but fail to implement bot protection. In the simulation, your product sells out to bots in seconds, your “Brand Health” metric plummets, and your real customers are furious. Learning this in a simulator costs $0; learning it in the real world can cost a brand millions.</p><h3>2. Understanding Interdependency</h3><p>The simulator forces you to see how Marketing (Hype Score) affects Engineering (Traffic Volume) which affects Finance (Net Profit). In a siloed corporate environment, these departments rarely see their direct impact on each other. Simulation breaks those silos.</p><h3>3. Algorithmic Domain Mastery</h3><p>To build a simulation, you have to define the rules of the domain. You can’t just say “Hype matters” — you have to decide <em>exactly how much</em> it matters relative to scarcity. This process of “codifying the domain” is the fastest way to become a subject matter expert.</p><h3>Conclusion</h3><p>Hype retail is the intersection of culture, math, and infrastructure. As commerce continues to move toward algorithmic-driven demand, the ability to simulate and predict these “Drop” events will be the difference between a viral success and a total system collapse.</p><p><em>Built with Streamlit, Plotly, and Gemini CLI.<br></em><a href="https://hype-retailer-simulator-bngud8nqlcbkuj4q8qyuzn.streamlit.app/">https://hype-retailer-simulator-bngud8nqlcbkuj4q8qyuzn.streamlit.app/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=461ec7e0f411" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I built a Multi-Agent AI Researcher using LangGraph and Streamlit]]></title>
            <link>https://lukeyish.medium.com/how-i-built-a-multi-agent-ai-researcher-using-langgraph-and-streamlit-c94d17ffd1b1?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/c94d17ffd1b1</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Sat, 07 Mar 2026 15:15:45 GMT</pubDate>
            <atom:updated>2026-03-07T15:15:45.105Z</atom:updated>
            <content:encoded><![CDATA[<p>Building AI agents is easy. Building a <em>reliable</em>, multi-agent system that doesn’t spiral into an infinite loop and actually solves a business problem? That’s a different story.</p><p>Recently, I built a <strong>Multi-Agent AI Researcher</strong> — a tool that takes a business idea, scans the web for market competition, audits GitHub for relevant open-source tech, and synthesizes the findings into a professional SWOT analysis. I used <strong>LangGraph</strong> for the agent orchestration and <strong>Streamlit</strong> for the frontend.</p><p>Here’s a deep dive into how I built it and the three biggest architectural challenges I had to overcome along the way.</p><h3>1. Passing State Between Autonomous Agents</h3><p>In a monolithic LLM script, you just append everything to a list of messages. But in a multi-agent system, agents operate as distinct nodes in a graph. The Market Researcher needs to pass its findings to the Business Strategist, while the Tech Auditor works in parallel or sequentially.</p><p><strong>The Solution:</strong> LangGraph’s StateGraph and TypedDict.</p><p>Instead of passing raw context windows around, I defined a strict, centralized GraphState:</p><pre>class GraphState(TypedDict):<br>    query: str<br>    market_research_data: Optional[str]<br>    github_data: Optional[str]<br>    swot_analysis: Optional[str]<br>    retries: int</pre><p>Each agent is just a Python function that takes this state, performs its specific duty (like calling the Tavily API or GitHub GraphQL API), and returns a dictionary with <em>only</em> the keys it wants to update. LangGraph handles merging these updates back into the global state. This meant the Business Strategist agent always had clean, predictable access to market_research_data and github_data without worrying about how they were generated.</p><h3>2. Preventing Infinite Loops in Self-Correction</h3><p>One of the coolest features of AI agents is self-correction. If the Tech Auditor searches GitHub for “AI generated Singapore history lessons” and finds zero repositories, it shouldn’t just give up. It should broaden its search terms and try again.</p><p>However, naively looping back to the same agent is a guaranteed recipe for catching your system in an infinite loop, burning through API credits.</p><p><strong>The Solution:</strong> Hard limits and conditional edges.</p><p>I added a retries integer to the global GraphState.</p><pre>def route_tech_auditor(state: GraphState):<br>    if state.get(&quot;retries&quot;, 0) &gt; 0 and state.get(&quot;github_data&quot;) is None:<br>        return &quot;tech_auditor&quot;<br>    return &quot;strategist&quot;</pre><p>Inside the tech_auditor_node, if the GitHub search returns nothing, it checks the retry count. If retries &lt; 1, it broadens the search term (e.g., dropping the query to just the longest word) and increments the retry count. Crucially, if it fails <em>again</em>, it gracefully degrades by returning a fallback message rather than triggering another retry. The route_tech_auditor conditional edge then ensures the graph safely moves forward to the strategist node instead of looping endlessly.</p><h3>3. The “Human-in-the-Loop” Validation</h3><p>AI shouldn’t run entirely unchecked, especially when making business strategy decisions. I needed the system to pause after the Market Researcher finished so <em>I</em> could validate the competition data before letting the Tech Auditor and Business Strategist burn compute on a heavily saturated niche.</p><p><strong>The Solution:</strong> Checkpointers and Streamlit Session State.</p><p>LangGraph makes pausing execution straightforward using a MemorySaver checkpointer:</p><pre>memory = MemorySaver()<br>graph = builder.compile(checkpointer=memory, interrupt_before=[&quot;tech_auditor&quot;])</pre><p>The real challenge was integrating this seamless pause-and-resume flow into a stateless frontend like Streamlit.</p><p>By tying LangGraph’s thread_id to Streamlit&#39;s session_state, I could stream the graph up to the breakpoint:</p><pre>for event in graph.stream({&quot;query&quot;: query, &quot;retries&quot;: 0}, config):<br>    pass</pre><p>Once it hit the interruption, Streamlit naturally rendered a “Human-in-the-Loop” UI, asking the user to approve the findings or start over. If approved, the pipeline blindly resumed from exactly where it left off by passing None as the input:</p><pre>for event in graph.stream(None, config):<br>    pass</pre><h3>The Result</h3><p>The final product isn’t just a wrapper around an LLM. It’s a structured, resilient state machine that performs autonomous research with guardrails against hallucinations and infinite loops, all while keeping the human firmly in the driver’s seat.</p><p>Transitioning from prompt engineering to agent engineering requires a shift from thinking about <em>text generation</em> to thinking about <em>distributed state management</em>. LangGraph and Streamlit proved to be an incredibly potent combination for exactly that.</p><p>You are free to check out the app <a href="https://multi-agent-researcher-h2falzskxmdbitcp62azec.streamlit.app/">here</a> and to connect with me on <a href="https://www.linkedin.com/in/lukeong7/">LinkedIn</a>!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c94d17ffd1b1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[12 Myths of Product Management]]></title>
            <link>https://lukeyish.medium.com/12-myths-of-product-management-9b7937294fe1?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/9b7937294fe1</guid>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Sun, 16 Jun 2024 07:03:46 GMT</pubDate>
            <atom:updated>2024-06-28T00:04:25.675Z</atom:updated>
            <content:encoded><![CDATA[<h4>I have recently taken the “Managing Digital Products” course from NUS-ISS and the learnings have been phenomenal. They have significantly changed the way I think about product management and I wish to share how my perspectives have been informed and refreshed.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hxEN7opYBstyqgffxa1jYA.png" /><figcaption>Taken from NUS-ISS Resources</figcaption></figure><h3>Myth 1: Project Managers are also Product Managers.</h3><p>To launch a digital product, a series of project tasks need to be prioritised and executed well. A digital product provides at least one specific benefit to users, whereas a project or a series of projects can ensure that the product is launched on time. Managing a product involves defining the reason(s) for its existence and developing its components to meet user needs.</p><p>Project Managers ensure that tasks are executed on time to meet clearly defined project deliverables, whereas Product Managers define the scope of the projects and provide justifications for the scope. Product Managers are responsible for successful business outcomes that are enabled through the products they manage and new product development should be broken down into more granular tasks or projects.</p><p><strong><em>Proposed Reality: Project Managers can be Product Managers, but not all Project Managers are Product Managers.</em></strong></p><h3>Myth 2: We should focus on Agile processes more than the principles.</h3><p>Agile is a set of principles that can manifest itself in some frameworks or processes, including SCRUM and Scaled Agile Framework (SAFe). The choice of the process is not as important as the application of the principles of Agile. As we apply Agile principles, we will start to:</p><ol><li><strong>Favor individuals and interactions</strong> over processes and tools</li><li><strong>Working software</strong> over comprehensive documentation</li><li><strong>Customer collaboration</strong> over contract negotiation</li><li><strong>Responding to change</strong> over following a plan</li></ol><p><em>Proposed Reality: We should focus on Agile principles.</em></p><h3>Myth 3: A Product serves to solely meet user needs.</h3><p>Ideo — TriFecta framework proposes three lens through which we can think about product innovation and are equally important.</p><ul><li>Desirability: Do they want this? Does it solve a pain point and benefit the consumer?</li><li>Viability: Should we do this? Does it contribute to a sustainable business outcome?</li><li>Feasibility: Can do we do this? Do we have the necessary resources (financial, technical, etc.) to complete this?</li></ul><p><strong><em>Proposed Reality: A Product exists only if there’s a business case and a use case for it.</em></strong></p><h3>Myth 4: We do not need well-defined user roles and personas to build a great product.</h3><p>User roles and personas provide a vivid description of the user whose needs we are trying to meet through the product. If they are not well-defined, the product will not be able to meet specific user needs and will not have a reason to exist. Clearly defined user roles and personas will lead the product team to develop a thorough and thoughtful user story map, which will guide the product manager to define the core features required in the MVP.</p><p><strong><em>Proposed Reality: We need well-defined user roles and personas to build a great product.</em></strong></p><h3>Myth 5: Gap in the Market is equivalent to Market in the Gap.</h3><p>For a gap to exist in the market, usually some people or businesses have been facing a pain point that they cannot overcome with existing solutions. In addition, this pain points bothers them so much to the extent that they are willing to explore alternative solutions and perhaps willingly pay for them. Identification of the Gap in the Market usually follows after problem-solution validation is performed, i.e. the problem’s significance is validated and the solution objectives are well-defined.</p><p>After we are certain that there is a real gap in the market, we should start defining how our solution should look like. This is the time we take in user feedback to achieve product-market fit, by using the methodology recommended by Eric Ries in his book <strong><em>The Lean Startup</em></strong>. We will also adopt various product validation strategies, such as the Concierge Method and the Wizard of Oz (or more affectionately known as “Flintstoning”). At this stage, we are really asking ourselves the question “Is there a business or market opportunity in the gap we identified earlier?”</p><p><strong><em>Proposed Reality: We need to analyze the gap in the market before analyzing the market in the gap.</em></strong></p><h3>Myth 6: We can break easily into existing markets for broad customer segments defined for our product.</h3><p>For a new product to perform exceedingly well in an existing market, product managers are advised to be laser-focused in defining customer segments and the niche area that the product should compete in. As the saying goes, if everything is important, then nothing really matters.</p><p><strong><em>Proposed Reality: We will experience difficulty breaking into existing markets if our customer segments are defined too broadly.</em></strong></p><h3>Myth 7: We do not need a North Star metric for our product.</h3><p>A North Star metric should inform the product manager to what extent his product vision statement has been realised.</p><p>This product vision statement should be phrased in this way:</p><p>“ For <em>&lt;User Role&gt;</em> who needs to solve <em>&lt;Pain Point&gt;</em>, <em>&lt;Product Name&gt;</em> is a <em>&lt;Product Category&gt;</em> that can perform <em>&lt;Function 1/2/3&gt;</em>. Unlike <em>&lt;Competitor 1&gt;</em>, our product offers <em>&lt;Point(s) of Differentiation&gt;.</em>”</p><p>The North Star should be a quantifiable metric and should enable discovery of other causative or associative metrics that can be used to align the product vision with the overall business objectives and/or objectives from other business functions, including Marketing/Sales and Business Development.</p><p><strong><em>Proposed Reality: The North Star metric is a natural transition from a clearly defined Product Vision statement and quantifies the success of the product.</em></strong></p><h3>Myth 8: We need to build our MVP before performing product validation and customer experience design.</h3><p>Product validation and customer experience design should be necessary steps that precede the development of MVP, which is a working product for users. Product validation is a means to acquire user feedback early, so that hypotheses about the user behavior or mindsets can be tested — we can use Concierge or Wizard of Oz methods to do this in a cost-effective way. Customer experience design is the foundation that we rely on to determine how our prospective MVP will fit into the overall user journey when they are trying to engage the product to perform their Job to Be Done (JTBD).</p><p><strong><em>Proposed reality: We need to validate our product in the bigger scheme of customer experience design before deciding on the core set of features to build the MVP.</em></strong></p><h3>Myth 9: The Product Roadmap needs to be planned with clear and specific requirements for the 3–5 year time horizon.</h3><p>Near-term plans are usually more clear and specific, compared with long-term plans. This is the principle when it comes to charting product roadmaps. Clear and specific product requirements may not be achievable in the time horizon of a few years, but perhaps doable within the next few months.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yPjy1_Qbi_k-xNvQN2tJdw.png" /></figure><p><strong><em>Proposed Reality: The Product Roadmap can be planned with clear and specific requirements in 3–6 months’ time, but it can get more hazy and uncertain as we project plans further into the future.</em></strong></p><h3>Myth 10: The user story specifies how the solution will look like.</h3><p>The user story is not a requirements specification document. The user story serves to inform the product manager or developer on the intended outcome that solution will need to achieve and should not be overly prescriptive. This is an example of a good user story.</p><p><em>“As a teacher, I want to create a submission dropbox so that students can submit assignments.”</em></p><p><strong><em>Proposed reality: The user story specifies the “what”, but does not specify the “how”.</em></strong></p><h3>Myth 11: We do not need a product positioning statement to market the product.</h3><p>The positioning statement is the key foundation for <strong>all</strong> marketing messages. It is how you want your target market to perceive your product or service. It must be customized to your target market that must deliver on its promise and differentiate your product from competitors’ products. It is also helpful for internal alignment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gkFACQxuWlFisLtHB79mSw.png" /></figure><p><strong><em>Proposed Reality: The Product Positioning statement underpins how value to the customer should be conveyed and is integral in all marketing messages.</em></strong></p><h3>Myth 12: Product support is not related to customer service.</h3><p>Product support is a key component of good customer service. They should not be viewed in silo.</p><p><strong><em>Proposed Reality: Product Support is a subset of good customer service and the objective of the product support is to meet customer needs.</em></strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9b7937294fe1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unlocking User Engagement: A Step-by-Step Guide for Sustainable B2B Product Enhancement]]></title>
            <link>https://lukeyish.medium.com/unveiling-user-engagement-a-strategic-review-for-sustainable-b2b-product-development-e6bc61a8c879?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6bc61a8c879</guid>
            <category><![CDATA[product]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[strategy]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Sun, 24 Mar 2024 13:47:32 GMT</pubDate>
            <atom:updated>2024-03-24T14:05:26.101Z</atom:updated>
            <content:encoded><![CDATA[<h3>Main Body</h3><p>In today’s business landscape, where sustainability and user satisfaction are crucial, refining product strategies is key. This article delves into a comprehensive method to review and optimize product strategies, specifically tailored to boost sustainability and user engagement in the B2B realm. By prioritizing understanding of user behaviors, frustrations, and preferences, this approach aims to deliver tangible improvements in product performance.</p><ol><li>Selecting Diverse User Representatives for Interviews: Begin by selecting a range of users for in-depth interviews. These users should represent different types of businesses and how they use the product. This variety ensures a broad perspective on what users need and want.</li><li>Diving into User Questions: Ask key questions based on Jobs to Be Done Framework. These questions help understand what users hire the product for, what they compare it to, and what issues they face with or without it. It’s about getting to the heart of what users truly need and why.</li><li>Crafting Specific Questions for Insight: Each key question based on Jobs to Be Done Framework leads to more detailed discussions. These discussions aim to uncover precisely what users struggle with and what would make their experience better. It’s about digging deep to understand their pain points and goals.</li><li>Gathering Insights from Interviews: Conduct in-depth interviews and make keen observations. These interviews provide qualitative data, while observations offer additional context to understand how users interact with the product and where they face difficulties.</li><li>Using Numbers to Understand User Behavior: Go beyond interviews and observe user actions through quantitative data analysis. By examining metrics like how often users use the product or which features they prefer, we gain a clearer picture of what drives user engagement and what areas need improvement.</li><li>Addressing User Concerns: Synthesize insights from interviews and data analysis to address user concerns comprehensively. Identify common issues and opportunities for enhancement to guide product strategy effectively.</li><li>Continuously Improving Based on Feedback: Remember, this process is ongoing. Keep gathering insights from users, analyzing data, and refining the product strategy. It’s about constantly evolving to meet users’ needs and expectations.</li></ol><p>Elevating user engagement and sustainability in B2B product development requires a deep understanding of user preferences and frustrations. By following a structured approach that combines qualitative interviews with quantitative data analysis, businesses can unlock valuable insights to drive meaningful improvements in product performance. Through continuous iteration and refinement, organizations can build a product strategy that not only enhances user engagement but also fosters sustainable growth in the long run.</p><h3>Footnotes</h3><p>In Jobs to be Done framework, we seek to understand the user with the following tenets.</p><ol><li>People buy products and services to get a “job” done</li><li>Jobs are functional, with emotional and social components.</li><li>A Job-to-be-Done is stable over time.</li><li>A Job-to-be-Done is solution agnostic.</li><li>Success comes from making the “job”, rather than the prodcuts or the customer, the unit of analysis.</li><li>A deep understanding of the customer’s “job” makes marketing more effective and innovation far more predictable.</li><li>People want products and services that will help them get a job done better and/or more cheaply.</li><li>People seek out products and services that enable them to get the entire job done on a single platform.</li><li>Customer needs, when tied to the job-to-be-done, make innovation predictable.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6bc61a8c879" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Harnessing Behavioral Insights in Product Management]]></title>
            <link>https://lukeyish.medium.com/harnessing-behavioral-insights-in-product-management-7b973c76e1f6?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/7b973c76e1f6</guid>
            <category><![CDATA[behavior]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[insights]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Sun, 24 Mar 2024 10:56:31 GMT</pubDate>
            <atom:updated>2024-03-24T10:56:31.575Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Understanding user behavior is paramount for product managers seeking to optimize product performance and user engagement. While existing frameworks offer valuable insights, there remains a need to effectively bridge the gap between product usage data and user needs. In this article, we explore the application of Behavioral Insights (BI) in product management, focusing on its objectives, benefits, and potential applications.</p><h3>Problem Statement</h3><p>Current approaches to understanding user behavior often lack a direct connection between product usage data and user needs. Qualitative frameworks like the Jobs to Be Done method provide depth, while data-driven approaches like the HEART framework offer breadth. However, a framework specifically designed to identify users to interview and relate product usage data to user needs is missing.</p><h3>Objectives and Benefits of BI</h3><p>BI, rooted in empirical study and drawing from various disciplines, aims to uncover the underlying motivations driving user behavior. By integrating qualitative and quantitative data, BI facilitates the creation of sensible user segments and sheds light on why users engage with or neglect a product. This approach enables product managers to make informed decisions based on a deep understanding of user needs and motivations.</p><h3>Potential Applications and Use Cases</h3><p>BI holds promise in various product management scenarios, including managing business offerings platforms for B2B users. By understanding user needs, refining lines of inquiry and setting up data-driven experiments, product managers can enhance user experiences and platform relevance to the users. Key considerations include identifying pain points, evaluating user satisfaction, and refining product offerings based on behavioral insights.</p><h3>Conclusion</h3><p>Behavioral Insights offer a comprehensive approach to understanding user behavior, bridging the gap between product usage data and user needs. By leveraging BI methodologies, product managers can gain valuable insights to drive innovation, optimize user experiences, and ultimately achieve product success. In today’s dynamic market landscape, BI provides a powerful tool for unlocking actionable insights and staying ahead of evolving user preferences.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7b973c76e1f6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Untitled]]></title>
            <link>https://lukeyish.medium.com/untitled-8aaed1130222?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/8aaed1130222</guid>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Sat, 24 Feb 2024 15:19:16 GMT</pubDate>
            <atom:updated>2024-02-24T15:19:16.155Z</atom:updated>
            <content:encoded><![CDATA[<p>#myendgame</p><p>自分を成長させて自由になる方法<br>ー 神様（イエス）を信じて頼る<br>ー 体を大事にする<br>ー 技術を学び上達できる<br>ー 人間関係を促進できる（日本の方へ）<br>ー 他人の考えをよくわかることができる</p><p>我的理想<br>•⁠ ⁠成为一个众人能信靠的人<br>•⁠ ⁠成为一个对主耶稣信实的人<br>•⁠ ⁠成为一个能让人看的更清晰，更彻底，更辽阔的人<br>•⁠ ⁠成为一个能鼓励人活出真正自我的人</p><p>Develop myself in these areas:<br>•⁠ ⁠Skills to thrive in the marketplace and help others to do the same<br>•⁠ ⁠Character to do the hard things and courage to do the right things<br>•⁠ ⁠Spirit to trust God to make the way and submit my plans in exchange for His plans</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8aaed1130222" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Logistics System Solutions are Key — Pain Point 1: Resource Management]]></title>
            <link>https://lukeyish.medium.com/logistics-system-solutions-are-key-pain-point-1-resource-management-6046a8969b6c?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/6046a8969b6c</guid>
            <category><![CDATA[logistics]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Fri, 07 Jul 2023 08:34:49 GMT</pubDate>
            <atom:updated>2023-07-07T08:34:49.286Z</atom:updated>
            <content:encoded><![CDATA[<h3>Logistics System Solutions are Key — Pain Point 1: Resource Management</h3><p><strong>Introduction to Logistics</strong></p><p>Logistics is an industry that relies on the people-augmented process of planning and running operations. The reliance on people purportedly allows logistics companies to react to uncertainty and quickly follow up with solutions that are effective.</p><p>For clarity, a group of items that are shipped in the same ‘container’ (this is a generic term and does not mean the TEU term that we commonly use) will be referred to as a parcel. Parcel represents the simplest unit of shipment in logistics, which cannot be further broken down in the entire parcel journey (or logistics flow).</p><p>This article discusses the importance of having system-based solutions to assist on-the-ground staff in logistics businesses to tackle the key pain points that I have identified, given my experience in PSA Singapore, Lazada and Janio Asia. Broadly, the benefits of having system-based solutions are:</p><ul><li>Reduce the turnaround time for implementing solutions in uncertainty</li><li>Reduce the learning curve for on-the-ground staff to implement solutions in uncertainty</li><li>Instil a intentional habit of continuous improvement and lean operations in logistics</li></ul><p><strong>Pain Point 1: Resource Management</strong></p><p>Objectives for seller are:</p><ul><li>Ensure that 99% of parcels are delivered on time</li><li>Ensure that parcels that are delivered beyond must-deliver time do not exceed must-deliver time by 1% of overall parcel journey</li></ul><p>Fixed constraints:</p><ul><li>Must-deliver time of parcels (to customer)</li><li>3PLs involved in parcel journey</li></ul><p>Flexible constraints:</p><ul><li>Time-windows for 3PL pickup at parcel journey level</li></ul><p>Cost control levers for 3PL:</p><ul><li>3PL parcel schedule planning: can be low/high parcel geographical pickup density per vehicle. Note: 1) Schedule = parcels to pick up broken down into shifts/batches/routes — e.g. AM/PM. 2) Highest uncertainty in 3PL parcel schedule for first-mile and last-mile</li><li>3PL fleet deployed in each shift/batch/route</li></ul><p>Parcel-level alert prioritisation challenge:</p><ul><li>If parcel is delayed at first-mile, it’s not reasonable to ask the next-mile service provider to prioritise the parcel handover for that specific parcel over other parcels</li><li>It’s not reasonable but if this delay can be predicted, time savings can be achieved even before the delay practically happens</li><li>This leads to us solving the parcel delay prediction challenge</li></ul><p>Parcel delay prediction challenge:</p><ul><li>If parcel delay to customer can be predicted, then what can be done about it?</li><li>I can trace the root cause of the predicted parcel delay to customer as a specific 3PL delay</li><li>I can arrange for earlier first-mile pickup — subject to seller availability; I can ask first-mile to prioritise pickup and enforce stricter service levels for downstream 3PLs before the specific 3PL delay actually happens; I can perform a cost-benefit analysis of the possible pre-emptive solutions, assess the tradeoffs and proceed with mitigating the delay risk or ignoring that</li><li>If preventive action is taken, all downstream 3PLs in the parcel journey have to be involved. They have to adjust accordingly to the results of the preventive action. That leads us to another challenge: how can we forecast the results of the preventive action solutions that we can implement?</li><li>Going ahead with the second approach above requires recalculation of time-windows for 3PL pickups, which has to be supported by system.</li><li>If system can do this, parcel delay to customer can be prevented and the revised parcel delivery date can be obtained and updated to the customer</li></ul><p>Preventive Action Results Forecast:</p><ul><li>This helps in cost benefit analysis of the possible preemptive solutions</li><li>This also helps the other downstream 3PLs in the parcel journey to adjust their plans according to the projected results of the preventive action</li><li>Only with their adjusted plans, the mitigation of the parcel delay can be achieved</li></ul><p>Hub inventory lead time challenge:</p><ul><li>If parcel is first transported to a inventory hub instead of cross-dock, the parcel lead time might be higher.</li><li>For cross-dock to be effective in reducing lead time, the prediction of parcel drop-off to the takeover 3PL needs to be accurate.</li></ul><p>Air Freight flexibility Challenge:</p><ul><li>Air freight are usually arranged in big batches at lower frequency. There is less flexibility to book in parcels for next flight at the last minute.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6046a8969b6c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What makes them different?]]></title>
            <link>https://lukeyish.medium.com/what-makes-them-different-9e2806fc7dd3?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/9e2806fc7dd3</guid>
            <category><![CDATA[difference]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Mon, 12 Jun 2023 08:03:18 GMT</pubDate>
            <atom:updated>2023-06-12T08:03:18.128Z</atom:updated>
            <content:encoded><![CDATA[<p>On the train, there was a really haggard old man that did not seem to have showered for many days. There was a foul stench coming from his body and I was seated right next to him. At the corner of my right eye, I can see a group of lively youths engrossed in their conversations and laughter, standing in the middle of the door entrance. Sitting opposite me was a lady in her 40s but she seemed to have a crooked spine and one of her bones jutted out of her back. The look on her face suggested that she was not entirely comfortable, but was thankful that she got the seat in the crowded train — that might be a short respite for her knees.</p><p>Did you visualise the scene with my felt senses as you were reading? What were the thoughts that ran through your mind? What did you feel in my position?</p><p>The old man, the group of youths and the lady in her 40s — we can relate to them to different extents. Even if I did not manage to strike a conversation with them on the train back then, I could notice how we were different as I engaged my senses — sight, smell and hearing.</p><p>In that moment, I could feel a gush of thoughts and emotions as I came to terms with these differences. I felt bad for the old man; I felt like I wanted to sit or stand further away from him to avoid smelling the stench; I did not understand why the old man had to live for many days without showering. I felt that the youths were in their own world and oblivious to their surroundings; part of me wanted to find a chance to remind them to be considerate of the needs of people around them, especially those that were trying to enter the train. For the 40-year-old woman, I felt sympathy and I thought of my own mother and grandmother, for they were lucky enough to live freely without the spinal issue that the woman had experienced.</p><p>Differences between people exist everywhere and they trigger a variety of emotions and thoughts in us as we perceive these differences. Behind these emotions and thoughts are a complex mechanism of foundational aspirations, values and beliefs that also make ourselves unique from the view of the world.</p><p>Let’s not always be quick to respond to these differences in negativity, but instead use these differences to understand what makes them different and what defines us.</p><p>Thanks for hearing me out and allowing my voice to also resonate a little in your inner world.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9e2806fc7dd3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ownership to me]]></title>
            <link>https://lukeyish.medium.com/ownership-to-me-f6c1874f095?source=rss-c5145130c473------2</link>
            <guid isPermaLink="false">https://medium.com/p/f6c1874f095</guid>
            <category><![CDATA[ownership]]></category>
            <dc:creator><![CDATA[Luke Ong]]></dc:creator>
            <pubDate>Fri, 09 Jun 2023 15:26:42 GMT</pubDate>
            <atom:updated>2023-06-09T15:26:42.172Z</atom:updated>
            <content:encoded><![CDATA[<p>Everyone prizes ownership. We like to be owners of things, for example, property, money, luxury products and many more tangibles.</p><p>I like to be an owner too. Before talking about the commonly coveted tangibles, I would like to share something more intangible, personal and perhaps influential. I want to share about how I am taking ownership of my voice.</p><p>In my career break, I have come to realise that having a voice that I can own and freely express keeps me healthy and sane. As an example, I started to own my unique perception of what a successful career means to me. Having predecessors that taught me what success meant to them is great, but I found it important to define what success meant to me, lest I lost ownership of my voice of what success actually meant.</p><p>I remember in a small 2022 hackathon in Singapore, I formed a team using this prompt. From one-man team, we grew to become four.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8loloG6869j30MkoEO7e8w.png" /><figcaption>Prompt written to gather hackathon members</figcaption></figure><p>What struck me, even at this time, is my focus on how every team member has an equal voice in driving ideas to real success. Why do I perceive this modus operandi as a core culture changer?</p><p>I’ve worked in big and small companies. I’ve struggled through bad ones and managed teams with cultures of varying strengths. There has been a consistent pattern that I’ve observed in these companies. Without perceiving that it’s a safe space to voice opinions, employees most likely perform badly in productivity and creativity. Constructive discourse do not happen; instead, seeds of discord and distrust are easily sown. People become guarded and stop bringing their full selves to work; they ‘mirror’ more often then ‘showing up’. We wonder, how can these companies still create this safe space where opinions are respected regardless of rank, influence or authority? We wonder how proper intellectual discourse can be conducted in the workplace. We wonder how something as simple as being taught in academic settings become so far off from the general workplace culture that we have all become so desensitised towards.</p><p>This article is a mere rant. But through this rant, I hope you can start to see what we have conventionally accepted in the workplace as normal as something that needs work to improve — that is the consequence of people disregarding ownership of their own voices, or worse, of their colleagues’ voices.</p><p>I share this point of view on the importance of owning our voice, not to empower everyone with a ticket to boast in front of everyone else that we have the right to our voice, but instead, to encourage the non-conformists in our midst to own their unique perspective, share it at the right time and in the right way, and with the right people. Owning your voice does not mean that you have to be a loudhailer. It can be as simple as giving your voice space to resonate in your life.</p><p>Maybe by getting in touch with that still small voice in you, you can make this world a better place, one voice at a time.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f6c1874f095" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>