On the pleasing absurdity of writing books about thinking carefully with AI, using AI, for a website that uses AI to think carefully about thinking carefully with AI
G. L. Nelson · maximinding.com · 2026
Let us begin with full disclosure, which in this case is also full comedy. The Cognitive Conductor series is a set of books about thinking carefully with AI tools rather than surrendering your judgment to them. The books argue, with some persistence, that you should stay in charge of your own thinking, maintain critical awareness of what the machine produces, and never mistake fluency for accuracy or speed for wisdom. The books were written with the assistance of an AI. The companion website, maximinding.com, which explores metacognitive approaches to human-AI collaboration, produces its research, curation, and drafts using a coordinated team of AI agents. The human conductor — the author — approves and refines the work at defined stages before anything is published. If you find this amusing, you are reading the situation correctly. If you find it hypocritical, you are reading it incorrectly — but the distinction requires some unpacking, and the unpacking is, rather fittingly, what the whole project has been about.
The Recursion Problem, Stated Plainly There is a word for a system that examines the processes by which it operates: metacognitive. A therapist who goes to therapy is practicing metacognition. A teacher who teaches teachers is practicing metacognition. A chef who teaches other chefs to taste more precisely is practicing metacognition. A writer who uses AI tools to write about using AI tools with greater awareness is — you can see where this is going — practicing metacognition. The recursion is not a flaw in the argument. It is the argument, demonstrated. The series does not say: do not use AI tools. It says: notice what you are doing when you use them, stay in charge of the judgment, and measure the gap between what the machine produces and what you actually think. I applied all three of these principles while writing the books. The website agents and I apply all three while producing its content. The practice and the preaching are the same. The series demonstrates its own argument. Whether that counts as integrity or merely as an unusually elegant piece of self-promotion is left as an exercise for the reader.
A Brief Guided Tour of the Irony Here is what the production of maximinding.com actually looks like, described without euphemism. A research agent identifies relevant developments in cognitive science, AI governance, metacognition research, and human-AI collaboration — scanning sources, flagging papers, and summarizing findings. A curation agent evaluates those findings for relevance, accuracy, and fit with the site's editorial perspective. A drafting agent produces initial versions of articles, briefs, and commentary. The human conductor reviews each stage, applying judgment at the decision points that matter: direction, tone, accuracy, editorial standards, and the question of whether the output actually says what needs to be said. This is, recognizably, the Generator-Critic-Synthesizer ensemble described in Book Three of the series. It is also, recognizably, the Seed-Mass-Carve cycle from Book One. The author was not following a methodology. The author was doing what the books describe and seeing that it worked — which is either reassuring or circular, depending on your disposition toward self-referential systems. The agents do not decide what the site believes. They do not set editorial direction. They do not determine what counts as a good argument or an honest acknowledgment of the framework's limitations. Those judgments sit, as they must, with the person whose name is on the work and who, being the human who initiated the process and approved the final product, is naturally accountable for it. The agents handle volume, variety, and first-pass quality. The conductor handles direction, judgment, and the final word. The humans who visit the site read the result and, if the calibration has worked, cannot distinguish a qualitative difference between this process and one conducted entirely by a single person who began by staring at a blank document. That invisibility is not a deception. It is craftsmanship.
What the agent team actually does at maximinding.com Research agent: Scans academic literature, governance reports, industry analysis, and primary sources. Flags items that meet editorial relevance criteria. Produces structured summaries with source attribution.
Curation agent: Evaluates research items for accuracy, relevance, and fit. Identifies gaps, contradictions, and items that warrant deeper investigation. Applies the site's editorial standards as a filter, not a rubber stamp.
Drafting agent: Produces initial versions of articles, commentaries, and briefs from the curated research. Generates multiple framings where the editorial direction is genuinely uncertain.
Human conductor — review gates: Reviews research before curation begins. Approves direction before drafting starts. Edits, rewrites, and approves final copy. Signs the work. Holds the accountability.
A Confession, Offered Without Excessive Embarrassment There is something the workflow description above omits, which is what it felt like to design it. The honest answer is: uncertain. The books make a series of claims about how thoughtful human-AI collaboration works, what it produces, and why it is worth the effort. Those claims are grounded in research, in observation, and in the author's own experience across many projects. But writing three books that tell other people how to think carefully with AI tools is, it turns out, a mildly vertiginous thing to do. The imposter syndrome was quiet and persistent. The internal question — who exactly are you to be saying this? — had a habit of arriving at inconvenient moments, often the moments when the argument was at its most confident. The agent workflow for maximinding.com was not designed purely for efficiency. It was also designed as a test. If the framework described in the books was sound, it should work when applied to the production of content about the framework. The practice should validate the preaching, or reveal where the preaching outran the practice. It did both, in useful proportions. The agents, applied without careful direction, produced exactly the kind of output the books warn against: fluent, confident, broadly accurate, and generically competent. Content that could have been about any topic, by any author, for any audience. The calibration log entries from the early weeks of the site (had they been kept in durable, publishable form) would have been a fair record of the gap between what the machine offered and what the conductor actually thought was worth saying. The gap was wide. Closing it required precisely the habits the books describe: the pre-prompt draft, the contrarian read, the willingness to scrap a fluent output because it was not yet honest enough to stand tall on its own without hedging. Applied consistently, the workflow improved. The content began to sound like it had been thought about rather than assembled. The imposter syndrome did not vanish, but it changed character — from 'who are you to say this?' to 'are you sure you're saying it well enough?' which is, I decided, a better question. The process was validating in the ways that matter: not because it confirmed my competence, but because it demonstrated that the framework was real and had a real value proposition to make. I promote what I practice because I practice what I promote. The calibration compounds. The work gets better. The doubt becomes productive rather than paralyzing. That is not a small thing to experience about something you wrote three short books arguing for. The imposter syndrome did not require resolution. It required practice. The practice, it turned out, was the point.
The Meta-Meta Question If the Cognitive Conductor series is about metacognition — thinking about thinking — and the website applies metacognitive principles to the production of content about metacognition — then maximinding.com is, structurally, metacognition about metacognition. Meta-metacognition. The thinking-about-thinking that is itself thinking about thinking. This is the point at which a certain kind of reader begins to feel dizzy, and a different kind of reader begins to feel at home. The author confesses to belonging to the second group. There is something genuinely interesting happening here that is not merely self-congratulatory recursion. The production process of the website is a live demonstration of the series' core claim: that the quality of human-AI collaboration depends entirely on the quality of the human's engagement, not the quality of the tool's output. The agents produce drafts of roughly consistent quality regardless of the editorial direction they receive. The direction is everything. Strip the conductor out of the process and you get competent, accurate, entirely characterless content that could have been produced by anyone, for anyone, about anything. Add the conductor back in — with a specific editorial perspective, a developed voice, a set of standards that the agents can be held to, and a genuine point of view about what the subject matter means — and you get something that sounds like a person thought about it. That distinction is the whole argument. It is also, if you have been paying attention, the whole point of the website. The agents provide the mass. The human sculpts. Without the first, there is nothing to shape. Without the second, there is nothing worth listening to.
On Being Caught in Your Own Argument Writers who produce books about writing are familiar with the peculiar exposure of having your working method become your subject matter. If you argue, in print, that good sentences require revision, and your own sentences are not particularly good, the reader notices. If you argue that thinking requires patience, and your argument is hasty and undercooked, the reader notices that too. The Cognitive Conductor series argues that good human-AI collaboration requires active judgment, honest evaluation, deliberate calibration, and a refusal to mistake the machine's fluency for your own thinking. Every piece of content produced by the author in the course of writing the books and maintaining the website has been subject to that standard. Some of it did not meet the standard. There were many iterations. Pieces were rewritten, or discarded, or held until the thinking caught up with the aspiration. The calibration log entries would have been the most honest document produced by the whole enterprise (alas, I am not a tightly structured person and failed to keep the logs in a lasting form): a running record of the gaps between what the machine offered and what I actually thought, and the work of closing them. That is what the site is, in the end. Not a demonstration of AI capability. A demonstration of what a human-AI collaboration can look like and produce. I went from a bucket of mental notes to a live website with articles and three downloadable PDF guides in 30 days, while working another full-time job.
A Note on maximinding.com The site exists because the books could not contain everything the subject demands. Metacognition in the age of AI is not a static topic. The research accumulates. The governance frameworks evolve. New failure modes emerge. New practices prove themselves. New questions arise that the books, written at a specific moment, did not and could not anticipate. The website is the living version of the argument. It updates because the conversation updates. It uses the agent ensemble because the volume of relevant material is beyond what a single human researcher could track without mechanical assistance. It publishes under the author's name because the judgment that shapes it — what to include, what to challenge, what to conclude — is the author's judgment, accountable in the way that only a person with skin in the game can be accountable. The agents do not have opinions about AI governance. They have outputs. The conductor has opinions. The distinction is, as the series has argued at some length, everything.
For the record This article was researched, drafted, and refined using the same human-AI workflow it describes.
The agents produced a competent first version. The conductor rewrote most of it.
That gap is the delta. It is where the judgment lived. It is, per the books, the whole point.
The agents have been informed of this. They expressed no opinion.
The Deeper Pleasantness What the author finds most genuinely satisfying about the recursive structure of this project is not the cleverness of the self-reference. It is the confirmation that the framework actually works — that applying the principles of calibrated human-AI collaboration to the production of content about calibrated human-AI collaboration produces better content than either the human or the agents would have produced alone. The research agent finds things the author would have missed. The curation agent applies the editorial standards more consistently than an exhausted human would at eleven in the evening. The drafting agent produces a version an author can react to rather than a blank page an author must fill. The human author provides the thing the agents cannot: a reason to care about the subject, a genuine point of view about what it means, and the willingness to say when the output, however fluent, is not yet good enough. That last part — the willingness to say it is not yet good enough — is, in the author's experience, the most frequently exercised skill in the whole enterprise. The agents are remarkably comfortable with whatever they produce. The conductor is not. This turns out to be useful. The machine generates. The calibrator decides. The site publishes what the calibrator decided. The books describe why that matters. And now there is an article that describes all of it at once, which was itself produced by the process it describes, which means the whole thing is either a philosophical snake eating its own tail or a proof of concept so complete it borders on smug. The author prefers 'proof of concept,' but acknowledges the snake.
maximinding.com Metacognitive approaches to collaborating with AI The Cognitive Conductor · The Calibrated Mind · The Strategic Conductor
On the pleasing absurdity of writing books about thinking carefully with AI, using AI, for a website that uses AI to think carefully about thinking carefully with AI
On the pleasing absurdity of writing books about thinking carefully with AI, using AI, for a website that uses AI to think carefully about thinking carefully with AI
G. L. Nelson · maximinding.com · 2026
Let us begin with full disclosure, which in this case is also full comedy. The Cognitive Conductor series is a set of books about thinking carefully with AI tools rather than surrendering your judgment to them. The books argue, with some persistence, that you should stay in charge of your own thinking, maintain critical awareness of what the machine produces, and never mistake fluency for accuracy or speed for wisdom. The books were written with the assistance of an AI. The companion website, maximinding.com, which explores metacognitive approaches to human-AI collaboration, produces its research, curation, and drafts using a coordinated team of AI agents. The human conductor — the author — approves and refines the work at defined stages before anything is published. If you find this amusing, you are reading the situation correctly. If you find it hypocritical, you are reading it incorrectly — but the distinction requires some unpacking, and the unpacking is, rather fittingly, what the whole project has been about.
The Recursion Problem, Stated Plainly There is a word for a system that examines the processes by which it operates: metacognitive. A therapist who goes to therapy is practicing metacognition. A teacher who teaches teachers is practicing metacognition. A chef who teaches other chefs to taste more precisely is practicing metacognition. A writer who uses AI tools to write about using AI tools with greater awareness is — you can see where this is going — practicing metacognition. The recursion is not a flaw in the argument. It is the argument, demonstrated. The series does not say: do not use AI tools. It says: notice what you are doing when you use them, stay in charge of the judgment, and measure the gap between what the machine produces and what you actually think. I applied all three of these principles while writing the books. The website agents and I apply all three while producing its content. The practice and the preaching are the same. The series demonstrates its own argument. Whether that counts as integrity or merely as an unusually elegant piece of self-promotion is left as an exercise for the reader.
A Brief Guided Tour of the Irony Here is what the production of maximinding.com actually looks like, described without euphemism. A research agent identifies relevant developments in cognitive science, AI governance, metacognition research, and human-AI collaboration — scanning sources, flagging papers, and summarizing findings. A curation agent evaluates those findings for relevance, accuracy, and fit with the site's editorial perspective. A drafting agent produces initial versions of articles, briefs, and commentary. The human conductor reviews each stage, applying judgment at the decision points that matter: direction, tone, accuracy, editorial standards, and the question of whether the output actually says what needs to be said. This is, recognizably, the Generator-Critic-Synthesizer ensemble described in Book Three of the series. It is also, recognizably, the Seed-Mass-Carve cycle from Book One. The author was not following a methodology. The author was doing what the books describe and seeing that it worked — which is either reassuring or circular, depending on your disposition toward self-referential systems. The agents do not decide what the site believes. They do not set editorial direction. They do not determine what counts as a good argument or an honest acknowledgment of the framework's limitations. Those judgments sit, as they must, with the person whose name is on the work and who, being the human who initiated the process and approved the final product, is naturally accountable for it. The agents handle volume, variety, and first-pass quality. The conductor handles direction, judgment, and the final word. The humans who visit the site read the result and, if the calibration has worked, cannot distinguish a qualitative difference between this process and one conducted entirely by a single person who began by staring at a blank document. That invisibility is not a deception. It is craftsmanship.
What the agent team actually does at maximinding.com Research agent: Scans academic literature, governance reports, industry analysis, and primary sources. Flags items that meet editorial relevance criteria. Produces structured summaries with source attribution.
Curation agent: Evaluates research items for accuracy, relevance, and fit. Identifies gaps, contradictions, and items that warrant deeper investigation. Applies the site's editorial standards as a filter, not a rubber stamp.
Drafting agent: Produces initial versions of articles, commentaries, and briefs from the curated research. Generates multiple framings where the editorial direction is genuinely uncertain.
Human conductor — review gates: Reviews research before curation begins. Approves direction before drafting starts. Edits, rewrites, and approves final copy. Signs the work. Holds the accountability.
A Confession, Offered Without Excessive Embarrassment There is something the workflow description above omits, which is what it felt like to design it. The honest answer is: uncertain. The books make a series of claims about how thoughtful human-AI collaboration works, what it produces, and why it is worth the effort. Those claims are grounded in research, in observation, and in the author's own experience across many projects. But writing three books that tell other people how to think carefully with AI tools is, it turns out, a mildly vertiginous thing to do. The imposter syndrome was quiet and persistent. The internal question — who exactly are you to be saying this? — had a habit of arriving at inconvenient moments, often the moments when the argument was at its most confident. The agent workflow for maximinding.com was not designed purely for efficiency. It was also designed as a test. If the framework described in the books was sound, it should work when applied to the production of content about the framework. The practice should validate the preaching, or reveal where the preaching outran the practice. It did both, in useful proportions. The agents, applied without careful direction, produced exactly the kind of output the books warn against: fluent, confident, broadly accurate, and generically competent. Content that could have been about any topic, by any author, for any audience. The calibration log entries from the early weeks of the site (had they been kept in durable, publishable form) would have been a fair record of the gap between what the machine offered and what the conductor actually thought was worth saying. The gap was wide. Closing it required precisely the habits the books describe: the pre-prompt draft, the contrarian read, the willingness to scrap a fluent output because it was not yet honest enough to stand tall on its own without hedging. Applied consistently, the workflow improved. The content began to sound like it had been thought about rather than assembled. The imposter syndrome did not vanish, but it changed character — from 'who are you to say this?' to 'are you sure you're saying it well enough?' which is, I decided, a better question. The process was validating in the ways that matter: not because it confirmed my competence, but because it demonstrated that the framework was real and had a real value proposition to make. I promote what I practice because I practice what I promote. The calibration compounds. The work gets better. The doubt becomes productive rather than paralyzing. That is not a small thing to experience about something you wrote three short books arguing for. The imposter syndrome did not require resolution. It required practice. The practice, it turned out, was the point.
The Meta-Meta Question If the Cognitive Conductor series is about metacognition — thinking about thinking — and the website applies metacognitive principles to the production of content about metacognition — then maximinding.com is, structurally, metacognition about metacognition. Meta-metacognition. The thinking-about-thinking that is itself thinking about thinking. This is the point at which a certain kind of reader begins to feel dizzy, and a different kind of reader begins to feel at home. The author confesses to belonging to the second group. There is something genuinely interesting happening here that is not merely self-congratulatory recursion. The production process of the website is a live demonstration of the series' core claim: that the quality of human-AI collaboration depends entirely on the quality of the human's engagement, not the quality of the tool's output. The agents produce drafts of roughly consistent quality regardless of the editorial direction they receive. The direction is everything. Strip the conductor out of the process and you get competent, accurate, entirely characterless content that could have been produced by anyone, for anyone, about anything. Add the conductor back in — with a specific editorial perspective, a developed voice, a set of standards that the agents can be held to, and a genuine point of view about what the subject matter means — and you get something that sounds like a person thought about it. That distinction is the whole argument. It is also, if you have been paying attention, the whole point of the website. The agents provide the mass. The human sculpts. Without the first, there is nothing to shape. Without the second, there is nothing worth listening to.
On Being Caught in Your Own Argument Writers who produce books about writing are familiar with the peculiar exposure of having your working method become your subject matter. If you argue, in print, that good sentences require revision, and your own sentences are not particularly good, the reader notices. If you argue that thinking requires patience, and your argument is hasty and undercooked, the reader notices that too. The Cognitive Conductor series argues that good human-AI collaboration requires active judgment, honest evaluation, deliberate calibration, and a refusal to mistake the machine's fluency for your own thinking. Every piece of content produced by the author in the course of writing the books and maintaining the website has been subject to that standard. Some of it did not meet the standard. There were many iterations. Pieces were rewritten, or discarded, or held until the thinking caught up with the aspiration. The calibration log entries would have been the most honest document produced by the whole enterprise (alas, I am not a tightly structured person and failed to keep the logs in a lasting form): a running record of the gaps between what the machine offered and what I actually thought, and the work of closing them. That is what the site is, in the end. Not a demonstration of AI capability. A demonstration of what a human-AI collaboration can look like and produce. I went from a bucket of mental notes to a live website with articles and three downloadable PDF guides in 30 days, while working another full-time job.
A Note on maximinding.com The site exists because the books could not contain everything the subject demands. Metacognition in the age of AI is not a static topic. The research accumulates. The governance frameworks evolve. New failure modes emerge. New practices prove themselves. New questions arise that the books, written at a specific moment, did not and could not anticipate. The website is the living version of the argument. It updates because the conversation updates. It uses the agent ensemble because the volume of relevant material is beyond what a single human researcher could track without mechanical assistance. It publishes under the author's name because the judgment that shapes it — what to include, what to challenge, what to conclude — is the author's judgment, accountable in the way that only a person with skin in the game can be accountable. The agents do not have opinions about AI governance. They have outputs. The conductor has opinions. The distinction is, as the series has argued at some length, everything.
For the record This article was researched, drafted, and refined using the same human-AI workflow it describes.
The agents produced a competent first version. The conductor rewrote most of it.
That gap is the delta. It is where the judgment lived. It is, per the books, the whole point.
The agents have been informed of this. They expressed no opinion.
The Deeper Pleasantness What the author finds most genuinely satisfying about the recursive structure of this project is not the cleverness of the self-reference. It is the confirmation that the framework actually works — that applying the principles of calibrated human-AI collaboration to the production of content about calibrated human-AI collaboration produces better content than either the human or the agents would have produced alone. The research agent finds things the author would have missed. The curation agent applies the editorial standards more consistently than an exhausted human would at eleven in the evening. The drafting agent produces a version an author can react to rather than a blank page an author must fill. The human author provides the thing the agents cannot: a reason to care about the subject, a genuine point of view about what it means, and the willingness to say when the output, however fluent, is not yet good enough. That last part — the willingness to say it is not yet good enough — is, in the author's experience, the most frequently exercised skill in the whole enterprise. The agents are remarkably comfortable with whatever they produce. The conductor is not. This turns out to be useful. The machine generates. The calibrator decides. The site publishes what the calibrator decided. The books describe why that matters. And now there is an article that describes all of it at once, which was itself produced by the process it describes, which means the whole thing is either a philosophical snake eating its own tail or a proof of concept so complete it borders on smug. The author prefers 'proof of concept,' but acknowledges the snake.
maximinding.com Metacognitive approaches to collaborating with AI The Cognitive Conductor · The Calibrated Mind · The Strategic Conductor
On the pleasing absurdity of writing books about thinking carefully with AI, using AI, for a website that uses AI to think carefully about thinking carefully with AI
