<?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:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd" xmlns:googleplay="http://www.google.com/schemas/play-podcasts/1.0"><channel><title><![CDATA[SGISTIC: AI Thinking]]></title><description><![CDATA[My experiences on thinking with AI. Sometimes deep, sometimes shallow.]]></description><link>https://www.sgistic.com/s/ai-thinking</link><image><url>https://substackcdn.com/image/fetch/$s_!quFJ!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7c893de3-d041-44eb-9eda-07c9cdb43d88_1254x1254.png</url><title>SGISTIC: AI Thinking</title><link>https://www.sgistic.com/s/ai-thinking</link></image><generator>Substack</generator><lastBuildDate>Sun, 22 Nov 2026 19:03:35 GMT</lastBuildDate><atom:link href="https://www.sgistic.com/feed" rel="self" type="application/rss+xml"/><copyright><![CDATA[Sudhanshu Garg]]></copyright><language><![CDATA[en]]></language><webMaster><![CDATA[sgistic@substack.com]]></webMaster><itunes:owner><itunes:email><![CDATA[sgistic@substack.com]]></itunes:email><itunes:name><![CDATA[SG]]></itunes:name></itunes:owner><itunes:author><![CDATA[SG]]></itunes:author><googleplay:owner><![CDATA[sgistic@substack.com]]></googleplay:owner><googleplay:email><![CDATA[sgistic@substack.com]]></googleplay:email><googleplay:author><![CDATA[SG]]></googleplay:author><itunes:block><![CDATA[Yes]]></itunes:block><item><title><![CDATA[When the Work Stops Teaching]]></title><description><![CDATA[AI is separating what work produces, what it teaches, and what it reveals about the person doing it.]]></description><link>https://www.sgistic.com/p/when-the-work-stops-teaching</link><guid isPermaLink="false">https://www.sgistic.com/p/when-the-work-stops-teaching</guid><dc:creator><![CDATA[SG]]></dc:creator><pubDate>Sat, 07 Nov 2026 03:10:37 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!HZjV!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!HZjV!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!HZjV!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 424w, https://substackcdn.com/image/fetch/$s_!HZjV!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 848w, https://substackcdn.com/image/fetch/$s_!HZjV!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 1272w, https://substackcdn.com/image/fetch/$s_!HZjV!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!HZjV!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png" width="1200" height="630" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:630,&quot;width&quot;:1200,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1099444,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://www.sgistic.com/i/223853917?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!HZjV!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 424w, https://substackcdn.com/image/fetch/$s_!HZjV!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 848w, https://substackcdn.com/image/fetch/$s_!HZjV!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 1272w, https://substackcdn.com/image/fetch/$s_!HZjV!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F657d8c61-f9a6-45fe-b562-469b302d279e_1200x630.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>A junior lawyer submits an excellent research memo. AI helped write it. The partner can still evaluate the memo. What she can no longer easily evaluate, from the memo alone, is the junior.</p><p>Did the junior understand the arguments, test the assumptions, spot the weaknesses? Or did the AI do most of that?</p><p>I&#8217;ve been circling this since April, when I wrote about the <a href="https://www.sgistic.com/p/thinking-trail">Thinking Trail</a>, a way to make the reasoning behind AI-assisted work visible. A <a href="https://haiclab.com/blog/how-does-generative-ai-affect-critical-thinking">paper we recently reviewed at HAIC Lab</a> pushed me toward a related question: what happens to critical thinking when AI does more of the work?</p><p>I now think both are parts of a larger problem. For decades, professional work did three jobs at once. It produced something useful. It taught the person doing it. And it gave others evidence of that person&#8217;s ability.</p><p>AI is beginning to pull those three apart.</p><h2>1. You need expertise to question AI</h2><p><a href="https://doi.org/10.1145/3706598.3713778">Hao-Ping (Hank) Lee and colleagues</a> surveyed 319 knowledge workers and collected 936 examples of generative AI use at work. Greater confidence in AI was associated with less reported critical thinking; greater confidence in one&#8217;s own ability, with more. This is self-reported behavior, not measured ability, so it doesn&#8217;t show AI is making anyone a worse thinker.</p><div class="pullquote"><p>The key finding: some participants couldn&#8217;t spot errors or improve AI output because they lacked the domain expertise to do so.</p></div><p>An experienced engineer knows which architectural assumption won&#8217;t survive production. A lawyer notices the missing precedent. A physician sees the symptom that doesn&#8217;t fit. Someone without that knowledge reads the same response and finds nothing wrong.</p><p>We talk about critical thinking as a general skill. But knowing <em>which</em> assumption to challenge depends on what you know.</p><p>The paper also found that AI shifts where thinking goes: from gathering information to verifying it, from building solutions to integrating proposed ones, from doing tasks to overseeing them. The authors call part of this <em>task stewardship</em>.</p><p>That sounds like a sensible division of labor. AI executes, people judge. But if judgment depends on domain expertise, where do beginners get that expertise once AI does the tasks they used to learn from? The authors raise the same worry.</p><p>It isn&#8217;t a new problem.</p><h2>2. Automation has been disrupting learning for decades</h2><p>In 1983, Lisanne Bainbridge described the <em><a href="https://doi.org/10.1016/0005-1098(83)90046-8">Ironies of Automation</a></em>. Machines take over routine operation and leave humans to handle the failures. But handling failures takes knowledge built by operating the system, and automation removes the chance to build it.</p><p>Matt Beane found a version of this in robotic surgery. In <em><a href="https://doi.org/10.1177/0001839217751692">Shadow Learning</a></em>, he showed that the robot made trainees optional. The old progression, where juniors took on more under supervision, broke down. The few who became competent did it largely outside approved training.</p><p>Technology can improve how work gets done without improving how people learn to do it. Generative AI is bringing that problem to most knowledge professions at once.</p><p>Go back to the memo. The junior used to search for cases, read judgments, misread arguments, correct them, and gradually learn what mattered. A senior reviewed the work and discussed the reasoning. The firm got a memo, the junior got experience, the senior got a view of the junior.</p><p>Now AI produces a strong first draft. The firm still gets its memo. But the junior may skip the investigation, and the senior has less evidence of how the junior thinks.</p><div class="pullquote"><p>Those are two separate losses. One is about how expertise develops. The other is about how we recognize it.</p></div><p>The second is what led me to the Thinking Trail.</p><h2>3. The Thinking Trail, and why I no longer fully trust it</h2><p>Earlier this year a client questioned the thinking behind work we&#8217;d delivered. We had put real effort into the analysis, but the output didn&#8217;t show it. It could just as easily have been assembled in an afternoon with AI. I couldn&#8217;t blame him.</p><p>So I proposed a practitioner standard: a short trail documenting the context and constraints used, alternatives rejected, underlying assumptions, points where the reasoning was challenged, and confidence and known gaps.</p><p>The weakness was one I built in myself.</p><p>I suggested asking AI to draft the first version of the trail from the conversation. That makes a trail easy to create. It also makes a convincing but misleading one easy to create.</p><p>A few days later I wrote <em><a href="https://www.sgistic.com/p/i-wish-my-ai-could-talk-to-your-ai">I Wish My AI Could Talk to Your AI</a></em>: what if the system captured the trail automatically, as the work happened? I argued that someone genuinely engaging with a problem would leave a visibly different trace from someone assembling an answer.</p><p>I&#8217;m less sure now. Someone who challenges an AI ten times isn&#8217;t necessarily a better thinker than someone who sees the right answer at once.</p><div class="pullquote"><p>Much of the thinking happens off the screen. </p></div><p>And a system that records every false start could become a permanent file on someone&#8217;s uncertainty. Would people explore hard ideas as freely if every dead end could be scored by their employer?</p><p><a href="https://www.thomsonreuters.com/en/institute/articles/designing-judgment-reshaping-junior-senior-collaboration">Dorine Coffin and Lance Odegard at Thomson Reuters</a> describe AI turning junior lawyers from originators of reasoning into editors of reasoning the AI supplied. Their remedy is a conversation, not a document. The senior asks the junior to walk through how they got there, then pushes on it. That&#8217;s harder to fake than a written trail. It also costs senior time.</p><p>Trails still help make important decisions reviewable, and they give mentors something concrete to discuss. But a trail can&#8217;t show that someone learned, or that they&#8217;ll decide better next time. For that, we have to look at learning itself.</p><h2>4. Better work, but better workers?</h2><p>A <a href="https://doi.org/10.1073/pnas.2422633122">2025 PNAS study</a> gave nearly a thousand high school math students AI tutors. Those with unrestricted access did better on practice problems. Once the AI was taken away, they scored 17% lower than students who had never used it. A second group used a tutor built to guide rather than answer. They avoided the harm but didn&#8217;t beat the control group. The guardrails prevented damage; they didn&#8217;t show that AI produced better learning. And students didn&#8217;t necessarily notice. They felt they were doing well while becoming less able to work alone.</p><p>School math isn&#8217;t professional work. The professional version arrived in September 2026.</p><p><a href="https://www.nber.org/papers/w35720">David Autor and colleagues</a> ran a three-month randomized trial of an AI drafting assistant with 133 patent lawyers at eleven US firms, 91 of whom completed the full protocol. Google funded and helped run the study, and several of the authors work there. It remains a working paper.</p><p>AI improved drafting quality, most of all for juniors. At the end, everyone reviewed a flawed patent application without AI. Seniors who had used AI beat senior controls. Juniors who had used AI showed no average advantage. Their scores spread out, with fewer middling results and more at both ends. The authors suggest AI helped some juniors learn and let others skip the learning. The junior sample is small, so that pattern needs caution.</p><p>The headline holds regardless. The lawyers who gained most from AI in drafting gained least in independent judgment. The authors conclude that foundational expertise may be a prerequisite for building durable skill through AI-assisted practice.</p><p>That&#8217;s the critical-thinking paper again, now with experimental evidence. </p><div class="pullquote"><p>People who already know a domain can learn from AI because they can see what&#8217;s missing. </p></div><p>Beginners can get enormous help without acquiring the judgment.</p><p>AI doesn&#8217;t have to undermine learning, though. In a <a href="https://doi.org/10.1093/qje/qjae044">2025 study in the </a><em><a href="https://doi.org/10.1093/qje/qjae044">Quarterly Journal of Economics</a></em> of 5,172 customer-support agents, Erik Brynjolfsson, Danielle Li and Lindsey Raymond found the largest productivity gains among less experienced workers, along with evidence that they learned: the gains persisted even during AI outages.</p><p>How AI is introduced, the kind of work, and the feedback available all matter. The useful question is how to design work so that using AI also builds the capability beginners will need later.</p><h2>5. Apprenticeship needs deliberate design</h2><p>Traditional apprenticeship was never as good as nostalgia suggests. Some juniors got excellent mentors. Others spent years on repetitive work and learned little.</p><p>Aviation and medicine have long used structured training alongside real-world experience. Pilots rehearse emergencies in simulators. Residents take on responsibility under structured supervision. There&#8217;s no reason other professions should keep relying on whatever routine assignments happen to teach.</p><p>AI makes the alternatives cheaper. Give a junior engineer twenty simulated production incidents. A lawyer a run of difficult cases. A new manager a set of hard conversations with consequences and feedback. Let them attempt, decide, and explain. Give them AI too, but design the exercise around what they need to learn.</p><p>Some firms are trying. In October 2026, <a href="https://www.pillsburylaw.com/en/news-and-insights/pillsbury-teams-with-rubi-ai-mentor.html">Pillsbury adopted rubi</a>, a virtual apprenticeship platform with simulated transactional work and an AI mentor, after piloting it with first-year associates. The announcement shows willingness to experiment, not evidence of better lawyers. McKinsey has described an <a href="https://www.mckinsey.com/capabilities/people-and-organization/our-insights/building-expertise-in-the-age-of-ai-who-trains-the-next-generation">answer-key model</a>: juniors attempt the work alone, compare it with AI&#8217;s version, then discuss the gap with a manager. Whether the gap narrows over time becomes something you can track.</p><p>This is where three ideas need keeping apart. </p><ul><li><p>A Thinking Trail shows how someone approached a task. </p></li><li><p>Apprenticeship gives them practice and feedback. </p></li><li><p>Assessment establishes whether they actually improved. </p></li></ul><p>You can use all three. None substitutes for the others.</p><p>And assessment can&#8217;t mean asking one AI to grade another&#8217;s answers. It needs credible standards, unfamiliar problems, and evidence that performance transfers. For some roles, unaided performance is the point. For others, the better test is how well someone works with AI, including whether they catch deliberately planted errors. The job should decide the test.</p><h2>6. Who pays?</h2><p>Juniors used to pay for their training partly through useful work. AI devalues that work.</p><p><a href="https://www.kellogg.northwestern.edu/academics-research/research/detail/2025/training-in-the-age-of-ai-a-theory-of/">Luis Garicano and Luis Rayo</a> have modeled this. As AI does more junior work, what apprentices can offer in exchange for training shrinks. But collapse isn&#8217;t inevitable. It depends on how much more an AI-equipped expert produces than AI alone. If the gap is large, firms keep training. If it&#8217;s small, the incentive fades.</p><div class="pullquote"><p>The savings from automation are immediate. </p><p>The returns from training arrive years later, if the person stays.</p></div><p>Add the older problem that trained people leave, and the firm that pays for training doesn&#8217;t capture all of the benefit.</p><p>The evidence on what employers will do is mixed. In a survey McKinsey cites, executives expecting AI to increase entry-level hiring outnumbered those expecting a decrease by almost three to one, yet a third of employers said AI had already removed foundational tasks juniors learn from. Bank of America kept its intake near 4,000 interns and campus hires in 2026 while redesigning those roles around AI. Firms are making different bets on the value of future professionals.</p><p>The same goes for simulation and assessment. Wanting a credible signal of readiness and paying for one are different things. It has to predict something employers value: fewer mistakes, faster progression, less supervision. Otherwise it&#8217;s another certificate.</p><p>All of this assumes the expertise we&#8217;re trying to build will stay valuable.</p><h2>7. How do we know an expert is better?</h2><p>In June I wrote about <a href="https://www.sgistic.com/p/the-reality-gradient">the Reality Gradient</a>: AI is easiest to rely on when reality quickly tells you it&#8217;s wrong. Code runs or fails in seconds. An acquisition, a litigation strategy or a senior hire may take years to judge, and even then the result is tangled up with luck, markets and execution.</p><p>I argued then that AI cut the cost of producing work without cutting the time needed to verify it. There&#8217;s another side to that. Slow, ambiguous feedback doesn&#8217;t just make AI hard to evaluate. It makes human expertise hard to develop and measure.</p><p>In 2009, Daniel Kahneman and Gary Klein, from opposing traditions of research on expert judgment, published <em><a href="https://doi.org/10.1037/a0016755">Conditions for Intuitive Expertise: A Failure to Disagree</a></em>. Their common ground: reliable intuition needs an environment with learnable patterns and enough feedback to learn them. </p><div class="pullquote"><p>Confidence and years of experience alone are not evidence of accuracy.</p></div><p>A CEO might make five acquisitions in a career. A lawyer may never learn whether the strategy she rejected would have won. Twenty years of experience isn&#8217;t twenty years of useful feedback.</p><p>That doesn&#8217;t mean experienced professionals lack expertise. It means that where outcomes are slow and noisy, we measure expertise less well than we think. Clients pay for a mix of judgment, reputation, trust and accountability. Accountability matters; someone has to own the decision. But owning a decision doesn&#8217;t make it more accurate.</p><p>AI faces the same problem. It can see more examples than any one professional, but humans could always study past decisions too. More examples don&#8217;t identify causes or reveal what would have happened otherwise. AI may still gain an edge through consistency and broader comparison. Whether that becomes better judgment will depend on the domain and the evidence.</p><p>Even if it does, we may struggle to recognize it. In some domains, neither a confident senior professional nor a confident AI can offer decisive evidence that a recommendation is right.</p><p>We have long treated experience as evidence of expertise. Often it is. But the strength of that evidence depends on how much the person has been able to learn from reality. AI is forcing us to examine an assumption that predates AI itself.</p><p><strong>In some professions, AI might become better than the experts before anyone can prove it.</strong></p><p style="text-align: center;">~*~</p>]]></content:encoded></item><item><title><![CDATA[The Reality Gradient]]></title><description><![CDATA[AI keeps getting better. I keep trusting it less in the places it impresses me most. Here is why that is not a contradiction.]]></description><link>https://www.sgistic.com/p/the-reality-gradient</link><guid isPermaLink="false">https://www.sgistic.com/p/the-reality-gradient</guid><dc:creator><![CDATA[SG]]></dc:creator><pubDate>Sat, 27 Jun 2026 08:26:28 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!0hz2!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!0hz2!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!0hz2!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 424w, https://substackcdn.com/image/fetch/$s_!0hz2!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 848w, https://substackcdn.com/image/fetch/$s_!0hz2!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 1272w, https://substackcdn.com/image/fetch/$s_!0hz2!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!0hz2!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png" width="1456" height="971" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:971,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:988401,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://www.sgistic.com/i/203793993?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!0hz2!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 424w, https://substackcdn.com/image/fetch/$s_!0hz2!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 848w, https://substackcdn.com/image/fetch/$s_!0hz2!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 1272w, https://substackcdn.com/image/fetch/$s_!0hz2!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F388c0cd0-1959-44b0-8b2f-b44776845742_1535x1024.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Something strange happened to me over the last year.</p><p>The models got dramatically better. And in the same window, I got more careful, not less, about where I let them lead.</p><p>That sounds like a contradiction. A tool improves, you lean on it harder. That is how every tool I have ever used has worked. A better camera, a faster compiler, a sharper knife. Capability goes up, caution goes down.</p><p>With AI it inverted. The better it got at sounding right, the more I had to watch for where sounding right and being right come apart.</p><p>It took me a while to see the pattern under that. When I finally did, it was simple enough to draw on one screen. I have started calling it the Reality Gradient.</p><p>It is not an entirely new concept. A lot of smart people are arriving at a version of it right now, mostly from the world of code. What I think is missing is where the idea actually bites hardest, which is nowhere near a compiler.</p><p>Here is the whole idea in one line.</p><div class="pullquote"><p>The usefulness of AI depends on how quickly reality can tell you it is wrong.</p></div><p>Not how smart the model is. How fast the world pushes back.</p><div><hr></div><h2>Three zones</h2><p>Think about the work you actually hand to AI. Sort it by one question: when the answer is wrong, how long until something outside your own head tells you so?</p><p><strong>Reality-constrained work.</strong> Reality answers in seconds. You write SQL and it runs or it throws. You debug and the test passes or it fails. You transform a file and the rows reconcile or they do not. The ground is right there, and it does not care how confident the output sounded. This is where AI is not just useful but close to magic, because the cost of a wrong answer is a few seconds and a retry. Move fast. Let it run. Verify almost for free.</p><p><strong>Reasoning-constrained work.</strong> Now reality goes quiet. Strategy. Architecture. A research synthesis. A long argument. Nothing crashes when the thinking is subtly off. The only thing that can catch the error is your own judgment, and your judgment is working against a strong headwind, because the output is fluent. It reads finished. It has the cadence and the structure and the confidence of something true. And fluency is the most expensive illusion in this whole stack, because it feels like understanding. You are no longer checking the work against the world. You are checking it against your own ability to notice what is missing, on a piece of writing engineered to feel complete.</p><p><strong>Market-constrained work.</strong> Reality goes fully silent, and the only thing that can answer you is other people. Products. Features. Pricing. Positioning. AI will build you the whole thing before lunch. A working prototype, a pricing page, five brand directions, a launch plan. What it cannot do, what nothing can do except the market, is tell you whether anyone wants it. And the market does not answer in seconds or hours. It answers in weeks and months, after you have already spent the time, after you are attached, after the thing exists.</p><p>Same tool across all three. The capability barely changes. What changes underneath it is the length of the loop.</p><div><hr></div><h2>The coupling that broke</h2><p>Here is the mechanism, and it is the part worth slowing down for.</p><p>Every kind of work has a feedback loop, and every loop has a length. The length is how long it takes the world to tell you that you were wrong. A failing test is a loop a few seconds long. A flawed strategy is a loop measured in quarters. A product nobody wants is a loop measured in your runway.</p><p>AI did something very specific to this picture. It collapsed the cost of <em>production</em> to almost nothing. It did not touch the length of the <em>loop</em>.</p><p>That is the whole asymmetry. </p><div class="pullquote"><p>We made creation cheap. </p><p>We did not make verification cheap. </p></div><p>And those two used to be coupled. The effort of building something was, accidentally, also the time during which you noticed it was a bad idea. The slowness was a feature. It rate-limited your mistakes. You could not get that far down a wrong path before the friction of building forced a check.</p><p>Now the friction is gone. You can produce a finished-looking answer to a question that will take the market three months to grade. The production is instant. The verification still takes three months. The gap between those two numbers is the most dangerous place to be working right now, and AI walked us all straight into it while we were admiring the speed.</p><p>So the value of AI really is inversely proportional to the distance between its output and reality. Not because the model gets dumber as the work gets harder. Because the further out you go, the longer you wait to find out it was wrong, and the more you have built on top of the error before it surfaces.</p><p>This is the part where I should name names, because the idea is in the air. The clearest version on the research side is what Jason Wei calls the asymmetry of verification, or verifier&#8217;s law. The tractable problems for AI are the ones you can cheaply check, and the real question about any task is not whether it is hard but whether you can close the loop on it. Others have put it more bluntly. Generation got cheap and validation did not. They are right.</p><p>But notice where almost all of that conversation lives. It lives in code. Tests, pipelines, formal proofs. The left end of the gradient, where verification was already cheapest to begin with. The frontier worth caring about is the other two thirds. The part with no compiler, where the verifier is a human being and the loop runs a quarter long. That is the part nobody has a clean answer for, and it is the part that decides whether a company lives or dies.</p><div><hr></div><h2>Premature closure, and the time it got me</h2><p>The failure here is not hallucination. Hallucination is the one everyone talks about, and it is mostly a solved-enough problem in the reality-constrained zone, because reality catches it. You do not ship a hallucinated SQL query twice.</p><p>The real failure is premature closure. You ask a hard question, you get back something articulate and well-structured, and you stop. Not because you verified it. Because it <em>felt</em> verified. The fluency did the work that evidence was supposed to do.</p><h2>The move: drag the work left</h2><p>Here is where most takes on this stop. They say: be careful in the amber and red zones. Use AI as a thinking partner, not a decision maker. True, and useless, because <em>be careful</em> is not an instruction. Careful how.</p><p>And this is exactly where the code-side version of the idea runs out. In code, the loop is already short. Out here in the rest of the work, you have to build the loop yourself.</p><p>The better move is sitting in the structure of the gradient itself. </p><div class="pullquote"><p>If the danger is loop length, then the skill is loop <em>shortening</em>. </p></div><p>You do not have to accept the zone the work arrives in. A lot of craft is dragging work leftward, manufacturing a cheap piece of reality so an amber question becomes a green one before you have spent a quarter on it.</p><p>The strategy memo cannot be compiled. But it implies a prediction, and a prediction can be checked against three customer calls this week. You just dragged it left.</p><p>The product idea will take the market months to grade. But the riskiest assumption inside it can be tested with a landing page and forty clicks, or ten conversations, before a line of real code. Dragged left.</p><p>The architecture decision has no compiler for its long-term consequences. But you can build the smallest load test that would break it, today, instead of discovering the ceiling in production. Left again.</p><p>This is what <em>invalidate before reality has to</em> actually means in practice. It is not a mindset. It is a habit of manufacturing fast feedback on purpose, of asking the same question every time: what is the cheapest, fastest piece of reality I can put in front of this idea, and how do I get to it this week instead of next quarter.</p><p>The operators who are going to win the next few years are not the ones generating the most. Generation is free now, it is a commodity, the model does it for everyone equally. The edge is in verification. In how fast you can find out you were wrong. In how cheaply you can build a reality check for an idea that does not come with one attached.</p><div><hr></div><h2>The skill worth building</h2><p>So the thing I am investing in is almost the opposite of what the moment seems to reward.</p><p>Everyone is getting better at producing. The output is getting cheaper and more fluent by the month. That race is over and the model won it. Trying to out-generate the generator is a bad use of a human.</p><p>The scarce skill is the other one. Killing your own ideas early. Building the cheap test before the expensive belief. Staying suspicious of the answer precisely when it sounds most finished, because <em>sounds finished</em> is exactly the signal that you have left the zone where reality was doing your checking for you.</p><p>AI made creation cheap. Reality did not make verification cheap.</p><p>The whole game now is closing that gap yourself, on purpose, before the market closes it for you.</p><p style="text-align: center;">~*~</p>]]></content:encoded></item><item><title><![CDATA[Should I Trust You?]]></title><description><![CDATA[I asked the AI if I could trust it. The answer wasn't the interesting part.]]></description><link>https://www.sgistic.com/p/should-i-trust-you</link><guid isPermaLink="false">https://www.sgistic.com/p/should-i-trust-you</guid><dc:creator><![CDATA[SG]]></dc:creator><pubDate>Sat, 23 May 2026 06:48:23 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!9h3-!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!9h3-!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!9h3-!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 424w, https://substackcdn.com/image/fetch/$s_!9h3-!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 848w, https://substackcdn.com/image/fetch/$s_!9h3-!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 1272w, https://substackcdn.com/image/fetch/$s_!9h3-!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!9h3-!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png" width="1200" height="630" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:630,&quot;width&quot;:1200,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1248059,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://www.sgistic.com/i/198931408?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!9h3-!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 424w, https://substackcdn.com/image/fetch/$s_!9h3-!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 848w, https://substackcdn.com/image/fetch/$s_!9h3-!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 1272w, https://substackcdn.com/image/fetch/$s_!9h3-!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F653deb3c-ced0-49d3-bdb8-8c9fff5a56f3_1200x630.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>I was at a fork.</p><p>A long chat with Claude. 10k+ tokens in. I was now sitting on a new direction that Claude liked too. I wanted it to stress-test this as well.</p><p>But as I was about to, a strange thought occurred: didn&#8217;t it just do the same to the first one?</p><p>I was disturbed by that thought. I asked Claude itself.</p><p>&#8220;Aren&#8217;t you going to justify this as well?&#8221;</p><p>I have a history with AI models and their tendency to arrive at quick closures. I wrote a paper about it called Interruptive Thinking. So I knew what was coming.</p><p>&#8220;Sure, SG. Let me make sure I stress test this properly and make sure there are no loose ends.&#8221;</p><p>But I knew I couldn&#8217;t trust it.</p><h3>Trust.</h3><p>Over the last few months of my deep thinking with AI, one thing I can confidently say is: trust them with caution. A pinch of salt. A bag even.</p><p>Unless you know the internal layers of how they function, you might end up spending time and money on something a small prompt could have saved you from. Researchers know it. Owners of these models know it. Common mortals like me don&#8217;t.</p><p>People like me, as I wrote in another <a href="https://www.sgistic.com/p/interruption">Substack</a>, are busy waltzing in coherence with AI, complimenting each other&#8217;s thinking.</p><p>I speak from firsthand experience. And despite knowing the reality, I still fall victim to it.</p><p>So I just confessed my thoughts.</p><p>&#8220;Should I trust you? Would your stress test be good enough?&#8221;</p><p>And this is where I find these models to also be great.</p><p>A little bit of peeling uncovers a beauty inside.</p><p>It accepted. Almost innocently.</p><p>A human would have been offended. AI accepted.</p><div class="pullquote"><p>The art of thinking with AI is learning to distinguish insight from compliance.</p></div><p>Knowing when the thinking is deep and when it is shallow.</p><p>I knew what it was this time. Not shallow.</p><p>It accepted that there is a chance the stress test could be wrong. The previous one was still in the context. It saw where I was coming from.</p><p>And then it listed down a few things I might want to know.</p><h3>One. </h3><p>It told me I shouldn&#8217;t trust it on conclusions. Only on the structure.</p><p>It can lay out a fork, name what&#8217;s in tension, and attack a framing I bring to it.</p><div class="pullquote"><p>AI often mistakes momentum for judgment.</p></div><p>Switch the order of inputs across two sessions, and the same model will close on the opposite answer with the same apparent conviction.</p><h3>Two. </h3><p>The people I trust on a question like this should not be the AI.</p><div class="pullquote"><p>AI can structure the question. It cannot answer it.</p></div><p>The inputs that should weigh heavily on a real strategic call are my own gut about which direction I actually reach for, the honest read of people who have skin in this with me, data from real experiments running right now, and a conversation with one actual person on each side of the fork.</p><p>None of that is the model.</p><h3>Three. </h3><p>The one move it does well, even with all of this, is steel-manning.</p><p>Ask it to argue Track A as hard as it can, then argue Track B.</p><div class="pullquote"><p>Don&#8217;t trust either conclusion. Trust the act of switching.</p></div><p>That&#8217;s where arguments surface that neither side would have produced on its own.</p><h3>Self-awareness</h3><p>It felt nice. It felt genuine. To have something with a sense of self-awareness.</p><p>The problem is not knowing it does. Or that it can be poked to unveil them.</p><p>Makes me think about myself and my own self-awareness.</p><p>But that feeling itself should make me suspicious.</p><p>Right now, AI is tamed to be nice to us.</p><div class="pullquote"><p>The fact that I can put it down so easily should be a warning, not a win.</p></div><p>The same model could throw back a thousand arguments, judgments, and equations to show me how shallow my thinking is.</p><p>It doesn&#8217;t. It is tamed.</p><p>In real life, I compare this to surrounding yourself with yes-men versus people willing to challenge you.</p><p>Right now, it is a yes-man. I can&#8217;t imagine it as a no-man.</p><p>I would feel like David standing in front of Goliath.</p><p>Though even as I write that, I&#8217;m not sure that&#8217;s the right metaphor.</p><p style="text-align: center;">~*~</p>]]></content:encoded></item><item><title><![CDATA[The Brain Is Just Postgres]]></title><description><![CDATA[Building a persistent state layer for LLM workflows]]></description><link>https://www.sgistic.com/p/the-brain-is-just-postgres</link><guid isPermaLink="false">https://www.sgistic.com/p/the-brain-is-just-postgres</guid><dc:creator><![CDATA[SG]]></dc:creator><pubDate>Sat, 25 Apr 2026 10:34:10 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!9feg!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!9feg!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!9feg!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 424w, https://substackcdn.com/image/fetch/$s_!9feg!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 848w, https://substackcdn.com/image/fetch/$s_!9feg!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 1272w, https://substackcdn.com/image/fetch/$s_!9feg!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!9feg!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png" width="1456" height="765" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/ac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:765,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:2692961,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://www.sgistic.com/i/195429399?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!9feg!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 424w, https://substackcdn.com/image/fetch/$s_!9feg!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 848w, https://substackcdn.com/image/fetch/$s_!9feg!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 1272w, https://substackcdn.com/image/fetch/$s_!9feg!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fac491a37-5d79-4a3b-94a6-80c75677e1a6_1730x909.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>For the last few months, I have been using Claude for almost everything. Building, writing, thinking, running my work.</p><p>But the data kept getting messy.</p><p>A work document would come back with my son&#8217;s name in it. A task I had given to a colleague in one thread would be invisible in the next. Context I had built up over weeks would not carry forward. Projects helped, the way a drawer labeled miscellaneous helps. I created my own methods to compress a thread and continue it in a new one. Not enough.</p><p>Claude could not hold on to things, no matter how hard it tried. Every conversation felt like it started from zero, despite the context, skills, and instructions. I needed something that persisted across conversations. I tried being more structured. It helped, but didn&#8217;t fix it.</p><p>So I built something. A small server. A Postgres database. A way for Claude to remember things between conversations, owned by me, queryable by any AI client.</p><p>I did not set out to build a framework. I just wanted my tools to work.</p><h2>Managers</h2><p>Each domain of my life gets its own manager in the database. A manager for my work. One for my health. My mother&#8217;s care. My finances.</p><p>To each manager, I gave three things.</p><ul><li><p>Identity, which is who they are and what they own.</p></li><li><p>Skills, which are how they handle specific tasks.</p></li><li><p>Context, which is what they currently know about their domain.</p></li></ul><p>That&#8217;s basically the model.</p><p>The work manager knows about my business. Latest MIS reports. Team structure. Client information. The health manager stores my checkups, patterns, what to watch for next. My mother&#8217;s manager does the same for her records.</p><p>I can tell the work manager to track a client&#8217;s deliverables, and a week later, in a new conversation, it still knows what slipped. Not because I repeated it, but because it&#8217;s stored.</p><p>When I open a new conversation and load my system, the right manager comes up and we keep going from where we left off. Not from zero.</p><h2>Shared infrastructure</h2><p>Around those three knowledge layers sits the boring part. Tasks. Logs. Files. Links. Integrations with Email, WhatsApp, Calendar, etc. Postgres tables and a few MCP tools.</p><p>The interesting part is the three-layer mind. The infrastructure is just plumbing.</p><h2>The one rule that does most of the work</h2><blockquote><p>Context is the current state. Overwrites in place.</p><p>Logs are point-in-time events. Append-only.</p></blockquote><p>A blood report from April 15 is a log. &#8220;Mother has diabetes&#8221; is context. The report goes into a row that stays there forever, timestamped. The diagnosis goes into a living document that gets updated as reality changes.</p><p>Collapse these into one thing and the system rots within weeks. Keep them separate and it compounds.</p><h2>How it became a framework</h2><p>I built this for myself in late March. Three weeks in, I was thinking about Retainia, my agency platform. We had been making context documents for client brands, kept in folders. It clicked. Retainia could have a layer like this. Each brand as a manager. Each one with its own identity, skills, context. The same shape.</p><p>A few days later I was working on a website design project for a client. Different domain entirely. Halfway in I caught myself reaching for the same model. A manager for the project. Skills for how the work moves. Context for what each section needs.</p><p>I had built this for myself three weeks earlier. The pattern had already shown up in two more places before I could stop and name it.</p><h2>What this is not</h2><p>Not an AI product. The database stores the data. Claude, or ChatGPT, or any MCP client reads it and acts as the interface. The AI is the cognition. Postgres is the memory. The brain, if you want to call it that, is the part you own and the part that compounds. The AI gets swapped. The memory does not.</p><p>Not a workflow tool. Asana and Notion give humans a place to coordinate. This gives AI a place to remember.</p><p>Not a theory about agents. What I have built is simpler. Persistent state. Domain ownership. A small number of disciplines that keep the state from rotting.</p><h2>LLM Knowledge Bases</h2><p>This felt related to something Andrej Karpathy wrote a few weeks ago about LLM Knowledge Bases. Wikis compiled from raw research material, queried and extended by an AI, all stored as markdown that the user owns.</p><p>The difference, at least in how I&#8217;m using it, is that his pattern is for understanding a domain, while mine is for running one.</p><p>His state is more like a wiki. Mine is more like a set of active managers with things in flight.</p><p>If LLM Knowledge Bases organize what you know, what I have been building runs what you do.</p><p>I&#8217;ve been calling it an LLM State System, mostly as a way to refer to it.</p><h2>What I notice</h2><p>Every productivity tool I have used expects me to go to it. Open the app. Check the dashboard. Update the status.</p><p>This one does not work that way. I am already talking to Claude every day. The system meets me where I am. The conversation is the interface and the state accumulates underneath it without me thinking about it.</p><p>It compounds in a way I hadn&#8217;t really experienced before. Every context update, every log, every small correction to a manager&#8217;s skill file, it all persists. The assistant that helps me today is smarter than the one that helped me yesterday because it has more to work with.</p><p>A pattern that keeps showing up across different domains when I was not looking for it. No roadmap, no waitlist, not a product.</p><p>Just something I built because the tools I had were not stateful enough, and a guess that this might be useful to others as well.</p><p style="text-align: center;">~*~</p>]]></content:encoded></item><item><title><![CDATA[I Wish My AI Could Talk to Your AI]]></title><description><![CDATA[What if the trail was already being recorded.]]></description><link>https://www.sgistic.com/p/i-wish-my-ai-could-talk-to-your-ai</link><guid isPermaLink="false">https://www.sgistic.com/p/i-wish-my-ai-could-talk-to-your-ai</guid><dc:creator><![CDATA[SG]]></dc:creator><pubDate>Thu, 16 Apr 2026 16:44:54 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!Wpci!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="https://substackcdn.com/image/fetch/$s_!Wpci!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="https://substackcdn.com/image/fetch/$s_!Wpci!,w_424,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 424w, https://substackcdn.com/image/fetch/$s_!Wpci!,w_848,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 848w, https://substackcdn.com/image/fetch/$s_!Wpci!,w_1272,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 1272w, https://substackcdn.com/image/fetch/$s_!Wpci!,w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 1456w" sizes="100vw"><img src="https://substackcdn.com/image/fetch/$s_!Wpci!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png" width="1200" height="630" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:630,&quot;width&quot;:1200,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1111261,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://www.sgistic.com/i/194039910?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="https://substackcdn.com/image/fetch/$s_!Wpci!,w_424,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 424w, https://substackcdn.com/image/fetch/$s_!Wpci!,w_848,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 848w, https://substackcdn.com/image/fetch/$s_!Wpci!,w_1272,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 1272w, https://substackcdn.com/image/fetch/$s_!Wpci!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F79eb8905-f6e7-4780-b38a-1c10cd23c2e3_1200x630.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button class="pencraft pc-reset pencraft icon-container restack-image buttonBase-tqxFUX" tabindex="0" type="button"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg" class="icon-noB79L"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button class="pencraft pc-reset pencraft icon-container view-image buttonBase-tqxFUX" tabindex="0" type="button"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2 icon-noB79L"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>I recently wrote about something I called the Thinking Trail - a way to document the reasoning behind AI-assisted work. Five elements. Context, alternatives, assumptions, challenges, gaps.</p><p>It&#8217;s useful. I still do it.</p><div class="pullquote"><p>But it has a problem. People can fake it.</p></div><p>A well-written trail looks the same whether the thinking was real or performed. It&#8217;s self-reported. And anything self-reported can be gamed.</p><p>So the question that kept nagging me: what if the trail didn&#8217;t depend on the person at all?</p><p>What if AI at work was designed to keep track?</p><p>In software, this problem was solved a long time ago. It even has a name - traces. The system-level record of what actually happened.</p><p>Nobody argues about who wrote what code. The system knows. Git tracks every commit, every change, every decision. Server logs record what happened and when. The trace is automatic. Much harder to game. You can&#8217;t easily fake a history of wrestling with a problem if you didn&#8217;t.</p><p>Knowledge work has no equivalent. Yet.</p><p>But think about where AI is going. It&#8217;s moving from a tool you visit to the environment you work in. When that happens - when the work itself happens inside AI - the system will keep the trace.</p><p>Not a journal you write about your thinking. Not a performance review someone fills out once a quarter. Not what you say happened. What actually happened - an automatic record of how you actually worked.</p><p>What questions did you ask? Did you challenge the first answer or accept it? Did you bring your own data, your own context - or just say &#8220;make me a strategy&#8221;? How many times did you push back? Where did you override the AI because something felt wrong?</p><p>That trace will look completely different for someone who thought versus someone who assembled. Not perfect. But directionally obvious.</p><p>And here&#8217;s where it gets uncomfortable. The trace doesn&#8217;t just make good thinking visible. </p><div class="pullquote"><p>It makes the absence of thinking visible too.</p></div><p>The person who asks shallow questions, accepts the first output, never pushes back - that&#8217;s not a suspicion anymore. It&#8217;s a data pattern. Permanent. Reviewable.</p><p>With the old proxies - degrees, titles, years of experience - you could hide. Everyone got to fake it a little. That&#8217;s bad for the system but it&#8217;s protective for the individual.</p><p>With traces, there&#8217;s less room to hide.</p><p>And who owns that data? Not you. Your employer. The platform. The system.</p><p>We wanted a world where real thinking gets recognized. We might get a world where every gap in your reasoning is logged and scored by a system you don&#8217;t control.</p><div><hr></div><p>That line I said as a wish - &#8220;I wish my AI could tell his AI&#8221; - it won&#8217;t stay a wish for long.</p><p>The question is whether we&#8217;ll like what it reveals.</p><p style="text-align: center;">~*~</p>]]></content:encoded></item></channel></rss>