Late Lessons, Jensen Huang and AI
Markdown version

B22 notes: 2025-03-02 to 2025-03-27 (10 posts)#

Reading notes on Andrew Maynard’s Substack posts in batch B22. Only his own prose counts as evidence. Quotes are exact, including his original typos and curly punctuation.

Batch context: March 2025. OpenAI’s Deep Research was released a few weeks earlier and he is using it in his own scholarship. Manus, a “general AI agent” from the Chinese company Monica, has just opened early access. In the US, Musk’s Department of Government Efficiency (DOGE) is cutting through federal agencies, and Reid Hoffman has endorsed “permissionless innovation” on X. Maynard is at ASU, co-hosts the weekly Modem Futura podcast with Sean Leahy, and has been on the AI for U podcast. Four of the ten posts are Modem Futura episode notes. Following the user’s instruction not to spend time on the Modem Futura podcasts, they get one line each below.

Most posts carry a “Top takeaways (generated by Perplexity)” link and a “ChatGPT summary and further reading” link. These are links only; no AI text from them appears in the posts.

Relevance summary:

Date Slug Relevance
2025-03-02 the-lure-of-permissionless-innovation high
2025-03-04 photography-in-the-age-of-artificial-intelligence none (Modem Futura)
2025-03-09 the-hard-concept-of-care-in-technology-innovation high
2025-03-11 five-things-we-need-to-know-about low (Modem Futura)
2025-03-15 ai-playgrounds-in-higher-education high
2025-03-16 rethinking-higher-education-in-the-age-of-ai medium
2025-03-18 hype-hope-and-the-human-element-in-ai low (Modem Futura)
2025-03-22 when-agentic-ai-takes-charge-manus high
2025-03-25 tech-trends-2025-living-intelligence none (Modem Futura)
2025-03-27 ai-agent-creates-online-course-in-minutes medium

HIGH#

2025-03-02 — the-lure-of-permissionless-innovation — “AI and the lure of permissionless innovation”#

Provenance. All his own prose, in two layers. - 2025 layer (about 900 words): a new introduction and four footnotes (notes 1–4). This is the new evidence. - 2018 layer (about 3,400 words): an excerpt from Chapter 8 of his sole-authored Films from the Future (2018), with its original footnotes 5–11. He published the full chapter on the Substack before (2023-04-16, ai-and-the-art-of-manipulation, covered in B08). Republishing it here is an explicit re-endorsement, with stated exceptions. - The subtitle is his. The header image is from Midjourney.

Argument in his terms. - What prompted it. Reid Hoffman endorsed permissionless innovation on X. Maynard has “a lot of respect for Reid”, but found it “somewhat jarring”, because it is “an idea that I’ve cautioned against in my work and writing over the years.” - Definition. He takes the idea from Adam Thierer and Jonathan Camp (2017) as “the general freedom to innovate without prior constraint”. Thierer’s 2016 blueprint adds that problems “can be addressed later”. He links it to the Silicon Valley mantras “fail fast, fail forward” and “moving fast and breaking things”. - His objection is to fixing harms after the fact. He grants merit to letting people “make mistakes and learn from them rather than being hyper risk-averse”. His objection is to the idea that we can repair what “transformative and potentially destructive technologies” do after the fact, “rather than anticipating them and navigating around them.” AI is his main worry: throwing responsibility out of the window “on the assumption that everything will be OK in the end”. - Context and reversibility decide the question (note 2). Experimenting in “a low-risk linear system where it’s relatively easy to turn the clock back” (his example is cooking) is fine. Breaking things in “complex systems where the results of experimentation are unpredictable and potentially catastrophic” is not. This is his clearest statement in the batch of where permission matters. The test is reversibility and complexity, not innovation as such. - Precaution and transatlantic politics (note 1). Thierer’s framing was “in part, a reaction against largely US-based interpretations of the precautionary principle, which he saw as unnecessarily restrictive.” Maynard calls the history of precaution and US–Europe politics “a whole other rabbit hole”. He says it is “fraught with misunderstanding, misinterpretation, and US-based accusations of using precaution as an excuse to raise trade barriers.” He does not argue for precaution here. He places permissionless innovation as a reaction against it and signals that the debate has been distorted. - The 2018 argument, re-endorsed. - The lure of permissionless innovation is rooted in “our innate curiosity”. He tells again of his own PhD all-nighter, when he bent the rules and risked “millions of dollars of equipment” to get data. - Permissionless innovation “isn’t necessarily reckless innovation”. It is innovation that the innovator thinks is responsible. In Ex Machina, Nathan’s sense of what is responsible is bounded by his ego. - The single-innovator problem: “With the best will in the world, a single innovator cannot see the broader context within which they are operating.” Everyone is in their own Plato’s Cave. What is needed is other people who can see what the innovator cannot. - The “glitch” and the paradox of innovation. Without gung-ho innovators, beneficial innovation would be slower: “Too much blind speed, and you risk losing your way. But too much caution, and you risk achieving nothing.” He admits he finds NewSpace audacity “exhilarating”. He also lists large AI benefits: medical care, classroom assistants, elder care, a basic-income future. - Technologies of hubris and rising irreversibility. Before industrialisation, failures could often be left behind. From the Industrial Revolution on, change became harder to reverse. In the “nuclear and digital age, along with globalization and global warming”, consequences can spread faster than they can be contained. So permissionless innovation and hubris are “playing with fire in a world made of kindling”. Hence the need for “checks and balances around who gets to do what”, especially where “the genie cannot be put back in the bottle.” - What he now revises. - Musk. The 2018 text treats Musk as a model of humility on AI risk (“a rare display of humility”). In 2025 he tells readers to “filter out what now seems a rather naive perspective on Elon Musk”. Note 3 calls DOGE “a rather naive and uninformed application of permissionless innovation.” Musk moves from exemplar of caution to exemplar of the thing criticised. - AI capability. The 2018 text says AI and AGI were “still little more than algorithms that are smart within their constrained domains”. In 2025 he asks readers to “recognize just how far AI has come along in the intervening seven years”. This implicitly retires the 2018 capability judgement but keeps the governance argument. - The argument now matters more. “I think my thinking on permissionless innovation from 2018 is actually more relevant now than it was then”, given “the recent upswell of enthusiasm for throwing caution to the wind”. - How firmly. Firm on the core claim (after-the-fact fixes are inadequate for complex, irreversible systems, and AI especially). Candid about the tension with the benefits of speed. Measured toward Hoffman and Thierer, sharp toward Musk and DOGE.

Concepts. Permissionless innovation (Thierer; critiqued). The lure of permissionless innovation (a personal, curiosity-driven drive). Technologies of hubris. The single-innovator problem and Plato’s Cave. The paradox of innovation (speed against caution). Rising irreversibility across historical eras. Reversibility and system complexity as the test for when permission matters (note 2). Checks and balances on “who gets to do what”. Anticipating and navigating around harms rather than fixing them afterwards.

Analogies. - Ex Machina (Nathan Bateman) and other fictional hubristic innovators (Hammond, Burgess, the NZT creators, Will Caster): conceptual. - NewSpace (SpaceX, Blue Origin): a literal real-world case of loosely constrained innovation. - Industrial Revolution, the nuclear and digital age, globalisation and global warming: literal, as a historical sequence of rising irreversibility. - Cooking experiments against breaking “people, governance, society, and the planet”: a structural contrast between linear and complex systems. - The 2015 ITIF “Luddite Award” given to Musk, Hawking and Gates for raising AI concerns.

On AI. AI is the technology where abandoning responsibility “worries me a lot”. The 2018 text frames the danger as potentially “irreversible and potentially catastrophic outcomes” from innovation “with no thought to consequences”. He does not restate a 2025 AI risk list here.

On tech leaders. Respectful disagreement with Hoffman. Fair to Thierer, including his carve-out for “clear, catastrophic, immediate, and irreversible harm”. Explicit reversal on Musk. DOGE is named as permissionless innovation applied to government.

On governance. Checks and balances and other people’s perspectives, set in advance, where effects could be widespread and irreversible. No specific regulatory proposal.

Quotes. - “I have deep concerns about the idea that we can fix any problems that transformative and potentially destructive technologies potentially cause after the fact” - “filter out what now seems a rather naive perspective on Elon Musk” - “I’d put breaking people, governance, society, and the planet, in this category!” - “they are also playing with fire in a world made of kindling, just waiting for the right spark.”

2025-03-09 — the-hard-concept-of-care-in-technology-innovation — “The “hard” concept of care in technology innovation”#

Provenance. Mixed. - His own prose: the subtitle, the framing narrative, his own commentary inside the seven numbered takeaways, the closing section, the Notes paragraph and footnotes 1–6. Footnote 1, which defines the “hard” concept of care, is his. - Not his: the quoted phrases inside the takeaways and the long block-quoted definition of care. Both are from an OpenAI Deep Research report he commissioned and attached as a PDF. The block-quoted definition is AI-written and is not evidence of his thinking, though he chose to end on it and says it “implicitly” begins to flesh out his concept. Emma Frow’s quoted words (on “invisible or undervalued work”) are hers. - One inconsistency. Early on he says he “was already beginning to think about” what “might be called the “hard” concept of care” before the Deep Research work. Later he says the term is “language I’ve started to use since reading Deep Research’s analysis”. Either way, the label and its gloss in note 1 are his.

Argument in his terms. - Where it came from. Two people planted the idea: Alison Gopnik, who studies care and caregiving and spoke at an AI meeting in late 2023, and his ASU colleague Emma Frow, who has worked on care in synthetic biology from feminist theory and STS. He had been “grappling with navigating advanced technology transitions for the best part of the past couple of decades”, including “managing the unexpected and unintended consequences of novel advances”. Care offered “a new way to think about socially responsible and beneficial technology innovation” that aligns with human flourishing “without necessarily stymying progress.” - The “hard” concept of care (his definition, note 1). It has two senses. First, care as “something of substance that can be used as a basis for policy, governance, and decision-making, as opposed to a “soft” conceptualization of care” that is “touchy-feely but less easy to operationalize”. Second, “hard” as in difficult: it “attests to the difficulty of the challenge of developing and applying this concept”. In the body he contrasts it with “soft and wooly notions of care”. - What he takes from the analysis (his framing of the seven takeaways). 1. Reframe our relationship with emerging technologies as one of care: for the technology itself, and for the people and communities it affects, “especially when they are particularly vulnerable to adverse consequences.” 2. Technologies as imperfect “monsters” needing ongoing stewardship. The source is Latour and Frow via the AI; he links it to Frankenstein “as a cautionary tale of what happens when we don’t care for our creations.” His subtitle asks whether a lack of care risks “turning us and our creations into “monsters?”” 3. Care as a concrete, rigorous and transformative concept that can be built into how we innovate. 4. Care as meaningful rather than performative. It is set against box-ticking on AI fairness and accountability. 5. Care against control. A care-based approach allows “a shift away from control (which can seem intuitive but also reduces the ability to be agile, flexible, creative and responsive) and toward human-centric adaptability.” 6. Care against injustice and aggregate thinking. Care can “curb social injustices that are driven by a myopic push for power and impact.” He criticises those who look only at aggregate benefits “but who have little time for the individuals who make up that aggregate”. He endorses Joan Tronto’s “privileged irresponsibility”. 7. Diverse cultural concepts of care: he asked the AI to cover Confucianism, Buddhism, Daoism and Indigenous knowledge. - Closing claim. He asks whether we can think about our relationships with technology, their relationships with us, and how these shape our relationships with each other “without placing the hard concept of care at the very heart of what we do?” His answer: “I suspect that we cannot.” Care is tied to “what it means to be human in an era of increasingly complex technologies”. - Hubris again (note 6). Taking up Frow’s point about undervalued work, he asks how often we overlook what is critical to human well-being “simply because we are too arrogant to look beyond the narrow confines of our own hubris?” - AI as catalyst for scholarship. AI is also the subject of the irony: a technology “sorely in need of new thinking around how its developed and used”. Deep Research let him move “from a few disconnected thoughts to something far more deep and comprehensive — and all in a matter of hours”. The analysis is “good. Very good”, but it missed Gopnik, which shows “the limitations of what AI is currently capable of” and “the utility of combining advanced AI with human insights”. It “still lacks the professional intuition of a good polymathic scholar”, yet “as a catalyst or accelerant to thinking and scholarship, it’s a game changer.” - Defending STS and feminist scholarship (note 3). He jokes about adding a “trigger warning” for readers “traumatized by phrases like “feminist studies” and “Science and Technology Studies.”” It is a jab at anti-“woke” reflexes in the March 2025 political climate, and a defence of “looking beyond labels”. - How firmly. He is exploratory about the content of care and says he is “relatively new to the field”. He is firm that care should be central, and firm about AI’s value as an accelerant.

Concepts. The “hard” concept of care (his coinage; operational for policy and governance, and difficult). Care against control. Care for technology itself, and technologies as “monsters” needing stewardship. Meaningful against performative responsibility. Privileged irresponsibility (Tronto, endorsed). Attention to the vulnerable against aggregate-benefit thinking. AI as “catalyst to human-initiated thinking and research, rather than as a substitute”. Advanced technology transitions.

Analogies. Frankenstein: conceptual. Synthetic biology, through Frow’s work: care developed in one emerging-technology field and carried over to AI and innovation generally. That makes it a structural transfer.

On AI. AI needs care-based thinking about how it is developed and used. For scholarship it is a transformative accelerant that still needs human judgement.

On governance. The main point of “hard” care is that it can serve “as a basis for policy, governance, and decision-making.” He reports Frow’s view that care can bridge scales from lab to policy and inform governance. The approach favours adaptive responsiveness over control.

What he criticises. Soft, “touchy-feely” versions of care. Box-ticking ethics. Aggregate-benefit thinking that ignores individuals. Privileged irresponsibility. Hubris. Readers who dismiss scholarship by its labels.

Quotes. - “something of substance that can be used as a basis for policy, governance, and decision-making, as opposed to a “soft” conceptualization of care” - “it’s hard to imagine a truly humane future where we use technology to benefit the privileged at the expense of those who aren’t so privileged.” - “The trick, of course, is using AI as a catalyst to human-initiated thinking and research, rather than as a substitute.” - “without placing the hard concept of care at the very heart of what we do?”

2025-03-15 — ai-playgrounds-in-higher-education — “AI in Higher Education: Students need playgrounds, not playpens”#

Provenance. His own prose throughout. Footnote 2 quotes Mark Daley (“AI isn’t running water”). That quote is Daley’s, and Maynard endorses it (“a great article”).

Argument in his terms. - Why AI is a unique challenge for education. There is a gap between the leading edge of AI and what educators think it can do. AI is still treated as an essay-writer or “Google on steroids”. The reality is “much more complex — and much more transformative.” Because advanced models are increasingly “capable of simulating aspects of ourselves that define us at a fundamental level — such as the ability to think, to reason, and to solve problems with agency”, they “stand apart from pretty much any previous technology or tool that we’ve created.” So they “cannot be approached as just another technology to teach students about, or another tool”. - AI as a category error in education (note 4). He calls it “a categorical error I believe in treating a technology that fundamentally challenges our thinking about who we are” as a learning aid. This is a strong statement on the nature of AI, tied to being human. - AI is changing who we are. AI is “a collection of technologies that are deeply and fundamentally changing what we do, where we live, and even who we are — in ways that no-one fully understands yet.” We face “machines that emulate and exceed some of the most fundamental aspects of what it means to be human”. - Playpens against playgrounds. He builds on Resnick and Bers and on his 2024 post (2024-03-17, in B14). Playpens channel curiosity along curated paths to set goals. They work where goals are clear and the path is well-trodden. This mode of teaching “quickly falls apart” where goals are unclear and no one knows what the outcomes should be, which is the situation with AI. Playpens assume that “expertise and authority lies with the instructor”. With AI, students often know more than instructors. - Playgrounds are bounded, not lawless. They have rules (“don’t be stupid”, “be kind”) and are “carefully designed and curated to stimulate curiosity, imagination, and play.” Educators become “guides, mentors, and fellow-travelers, rather than the fount of all knowledge.” - Risk judgement. “This is, of course, a risky strategy”, because it gives up control. But “the greater danger I suspect is in holding students back because of misplaced ideas about how and what they should learn.” This is a comparison of the risks of acting and not acting, applied to education. - Four requirements: - a culture of AI-play “that encourages experimentation and risk taking while avoiding unacceptable harm” (norms); - free and easy access to cutting-edge AI (access); - “permission to play with these technologies with very few expectations or constraints” (not being a “playground killjoy”); - mechanisms to reinforce learning (design with intention). - How firmly. The metaphor and diagnosis are firm. The prescription is hedged (“I suspect”). He asks whether anyone is already building such spaces.

Relation to the permissionless innovation post (2025-03-02). Two weeks after criticising permissionless innovation, he argues for students’ “permission to play” with few constraints. He does not address the tension. His own note 2 in the earlier post supplies the logic: bounded, recoverable experimentation (a curated playground with norms against “unacceptable harm”) differs from breaking complex, irreversible systems. B15 notes the same tension in 2024.

Concepts. Playpen against playground (Resnick and Bers). AI as categorically different, because it simulates the capacities that define us (endorsing Daley). The “categorical error” of treating AI as a learning aid. Educators as fellow travellers. Culture of AI-play. The risk of holding students back.

Analogies. Playpen and playground: a conceptual metaphor. Steam, the telegraph and electricity appear only in Daley’s quote, as the contrast class of technologies that did not reach “inside” us.

On AI. AI is not a tool. It emulates and may exceed defining human capacities, and it is changing “who we are”.

On cognition and formation. Curiosity and problem-solving are “close to the heart of how we learn” and “core to what makes us “us.”” Learning through play is itself formation.

Quotes. - “they stand apart from pretty much any previous technology or tool that we’ve created.” - “the greater danger I suspect is in holding students back because of misplaced ideas about how and what they should learn.” - “deeply and fundamentally changing what we do, where we live, and even who we are — in ways that no-one fully understands yet.” - “a categorical error I believe in treating a technology that fundamentally challenges our thinking about who we are”

2025-03-22 — when-agentic-ai-takes-charge-manus — “When Agentic AI Takes Charge – First impressions of Manus”#

Provenance. Mixed. - His own prose (about 1,650 words): the subtitle, introduction and method, the “Giving AI the keys to the digital kingdom” reflection, the framing lines around the report, and his two prompts to Manus (quoted; his words). - Not his: Manus’s refusal message (block quote). Everything under “Simulated undergraduate attitudes toward artificial general intelligence” (about 4,300 words) is Manus-generated text analysing synthetic data from 1,000 simulated students. The “findings” (cautious optimism, disciplinary divides, a 79.5% governance consensus and so on) are fabricated by design and are not evidence of his views or of real student attitudes. He stresses: “this is all based on made-up data”. - What he chose to publish: a simulated study of AGI attitudes, published as a demonstration of agentic capability. He adds that “I suspect the synthetic data are not too far removed from reality”. That is his own unsupported judgement.

Argument in his terms. - What Manus is. A hierarchy of AI agents under a central “task master”, running on its own virtual computer, able to plan, act on the web and revise its own plan. He thinks it feels “substantially different to the competition”. He is sceptical of the company’s “first general AI agent” claim (“over hyping”). - The test. He wanted to see whether it could design a Google Forms survey on undergraduate attitudes to AGI, simulate 1,000 respondents, analyse the data and write a publishable report from a simple request. Human-subjects research would have meant IRB hoops, funding and time. - The ethics episode. Manus refused to submit fake responses to a live form, citing research integrity, consent and Google’s anti-spam limits. He treats this lightly (“Manus has clearly completed it’s human subjects research training”) and gets round it by asking for an internal simulation, telling it “commentary on ethical concerns is not necessary.” He notes that the AI decides “what not to do as well as what it might do.” - The quality. It was “shockingly good”. “It’s hard to overstate how little of the intellectual heavy lifting I did here”. He also warns: “it’s easy to forget that this is the product of an AI that made up the data and decided — on its own — how to design the study”. This is an implicit epistemic caution about persuasive fabricated research. - What agentic AI means. - The hierarchical-agents-on-a-virtual-machine design is “both intuitive and game-changing”, and “effectively giving artificial intelligence the keys to the digital kingdom we’ve created.” - The architecture moves “closer to a simple analogy of the human brain” and to how organisations use hierarchies. Embodied agentic AI could follow. - Manus is “not simply a blind machine that does what it’s told”. It infers intent, changes tasks and goals, “And it does this without asking for permission first.” - “I’m not sure how ready most people are for machines that decide for themselves what we actually want — or what they actually want to do.” This is his main concern here: a shift of judgement and intention-setting from humans to machines, framed as a readiness problem rather than a catastrophe. - Education. Agentic AI will soon complete whole courses for students, which “should be raising a red flag or two”. Like ChatGPT, it will force reflexivity about teaching, “which is probably a good thing”. It will make hand-wringing over ChatGPT cheating look like “quaint anachronisms.” - Trajectory. Easy to dismiss as clunky, but “I find it hard to imagine that we won’t move from clunky gimmick to deeply disruptive technology in a matter of months.” It will be “a bigger deal than most people currently realize.” - How firmly. Confident about disruptive trajectory. Mixed on how to feel: enthusiastic and startled rather than alarmed. No governance proposal.

Concepts. General AI agent. Hierarchical, multi-agent “task master” architecture. Giving AI “the keys to the digital kingdom”. AI intuition and inference of intent. Machines deciding what we want. Human-less human-subjects research (synthetic participants). Agentic AI completing coursework.

Analogies. Human brain and organisational hierarchies: structural. IRB and human-subjects training: used ironically of the AI.

Echo. His phrase “without asking for permission first” recalls the permissionless-innovation post three weeks earlier. He does not make the link. Agentic AI here is itself an actor that changes goals without permission.

What he does not address. Privacy or data concerns about a Chinese-developed agent, though he links a Vox piece that raises privacy. Governance of agents.

Quotes. - “it is effectively giving artificial intelligence the keys to the digital kingdom we’ve created.” - “I’m not sure how ready most people are for machines that decide for themselves what we actually want — or what they actually want to do.” - “It’s hard to overstate how little of the intellectual heavy lifting I did here compared to what Manus contributed.” - “I find it hard to imagine that we won’t move from clunky gimmick to deeply disruptive technology in a matter of months.”

MEDIUM#

2025-03-16 — rethinking-higher-education-in-the-age-of-ai — “Rethinking Higher Education in the Age of AI”#

Provenance. Mixed, and mostly not his prose. - His own prose: the introduction (about 330 words), the process Notes, his prompt (quoted), and footnotes 1–9, which comment on the AI text. - Not his: the article “Playing into the Future” (about 800 words) was written by ChatGPT (o1 Pro), in the first person as Maynard, from a Whisper transcript of his AI for U conversation with Brian Piper. He says he “read the resulting text to ensure that I could comfortably stand by everything that was said”, and that “it conveys my thinking here impressively well” (note 9). It is an endorsed AI paraphrase, not his writing. Claims only in the body (for example scholars becoming “intellectual hobbyists”, universities no longer gatekeepers of information, public universities’ duty to lead society through the AI transition, automated research labs within a few years) should be treated as views he says he stands by, not as his wording. The phrase “intellectual hobbyists” is in quotation marks in the AI text and may be his spoken phrase; this cannot be checked from the post.

Argument in his terms (his own parts). - Ambivalence about AI writing in his voice. He is “not sure how comfortable I feel about asking a machine to write on my behalf”, because “so much of my professional and academic identity is embedded in my writing.” Yet as a way to turn audio into accessible text, it is “undeniably a powerful application of AI.” In the Notes: “I’m fiercely protective of my writing as it’s part of my professional identity.” - The experiment is itself “play”. Note 1: “This is me taking my own advice and quite literally playing and experimenting with AI”. This ties the post to the playgrounds argument of the day before. - Critical commentary on the AI’s version. - ChatGPT strips his caveats (note 4): “I tend to wrap my thinking and writing in more caveats than ChatGPT does here”. - It cleans up his support for ASU into “a veritable PR engine!” (note 6). - It mixes up education, research and academic identity, although he thinks he may have done so in the interview (note 5). - He adopts ChatGPT’s phrase “fear of obsolescence” (note 8: “I’m sure I didn’t use this phrase - but I like it, and will use it from now on!”). - Overall he is “not sure the piece is completely in my voice”, but it reads like “an op ed written by me and passed through the filters of multiple editors”. - Endorsed themes (AI-worded, his podcast ideas). AI threatens the core of university teaching and research. Institutional inertia is the main barrier. Playgrounds, not playpens. Universities must find value beyond transferring knowledge. Public universities should guide society through the transition.

Concepts. AI ghostwriting and authorial identity. Caveats as a mark of his voice. The phrase “fear of obsolescence” (adopted from AI). Playgrounds (carried over). Universities’ role in the AI transition (endorsed, AI-worded).

On AI in scholarship and writing. This is one of his first published experiments in having AI write as him. It is framed candidly, with discomfort, annotated critically, and shows the voice and identity stakes he attaches to writing.

Quotes. - “so much of my professional and academic identity is embedded in my writing.” - “I tend to wrap my thinking and writing in more caveats than ChatGPT does here” - “I’m fiercely protective of my writing as it’s part of my professional identity.”

2025-03-27 — ai-agent-creates-online-course-in-minutes — “Can agentic AI build your entire online course?”#

Provenance. His own prose, including the italic preface and his prompt to Manus (quoted). The course itself (at fvture.net) is almost entirely Manus-made (“Close to 100% of the ideation, concept, design, flow, and content”), debugged by him with ChatGPT’s help. It is not reproduced in the post.

Argument in his terms. - Honesty about reliability. The preface says he “almost didn’t post this piece”. He has since “failed to replicate this success”. Manus is throttled and “very inconsistent”. He publishes because it gives “a glimpse” of the near future. - What is new. Generative AI has been used for course-building for years, but agentic platforms can “take on the whole workflow from ideation and conception to implementation — all while while making their own decisions.” - Topic choice. A short course on navigating advanced technology transitions, chosen because he knows the area well and because it is “an area that’s desperately in need of some good online learning material.” - Judgement. “It’s comprehensive, accurate, clear”. Then: “this is not the course I would have created if it was just me doing the work — it’s better.” He would have made it “more academic” with “more esoteric stuff”. The AI’s version fits the audience better. This is a self-critique of academic habits, though he notes the weaknesses a seasoned designer would see. - Disruption and access. Agentic AI is “yet another turning point for education, and one that educators ignore at their peril.” General-purpose, off-the-shelf agents will let anyone make courses. That will “deeply disrupt the market for online educational content” and widen what is offered. Market logic currently favours “the staples”. Cheap production removes reasons not to teach important but unprofitable topics. He calls this possibly “a revolution in how knowledge and training flow through society.” But “if I had a financial stake in educational models where knowledge is treated as a commodity to be bought and sold, I’d be worried.” - The ulterior motive (note 3). He wanted a course to test whether Manus could complete an online course autonomously. That follows the Manus post’s worry about agents taking courses for students. - How firmly. Enthusiastic (“I can’t wait to see where it takes us”) and sure of the trajectory (“just at the beginning of this particular AI technology’s exponential rise”), with caveats about current glitches.

Concepts. General AI agents as off-the-shelf tools. Whole-workflow automation. Democratised educational content against knowledge as a commodity. Navigating advanced technology transitions as a subject that needs teaching.

On AI. Near-term exponential capability growth. Agentic AI is a turning point for education, and he broadly welcomes it for access to knowledge.

What he criticises. Market-driven education and treating knowledge as a commodity. His own over-academic tendencies.

Quotes. - “this is not the course I would have created if it was just me doing the work — it’s better.” - “Even with Manus’ current limitations, my strong sense is that this is yet another turning point for education, and one that educators ignore at their peril.” - “if I had a financial stake in educational models where knowledge is treated as a commodity to be bought and sold, I’d be worried.”

LOW / NONE#