I think you have to introduce it at a reasonably large scale because you're going to need a number of AI-savvy software engineers who are going to be fairly expensive. So the whole thing will cost a lot to get going and needs to be funded somehow.
Agreed that it would have to be an entire school, either a large private high school or a private liberal arts college -- I am guessing larger school, especially a public university, would have too much institutional inertia. You would a group of wealthy trustees who wanted the prestige of doing something big. That's how I see it happening.
I am a fan of Hollis Robbins's heterodoxy when it comes to AI and higher education, not because I agree with her, but because Anectotal Value offers ideas that challenge the usual terms of what Henry Farrell calls AI Fight Club, which pits those who see AI-based educational tools as our inevitable future against those who resist AI in large or small ways.
In proposing that CSU automate general education through a microservices architecture consisting of AI applications, Robbins offers a concrete alternative (a genuinely awful one) to the already existing structure of gen-ed (also genuinely awful). Actually existing systematic general education is no longer viable because OpenAI built Shel Silverstein's "Homework Machine" and offers it to college students at no cost. AI, it seems, is the future of the modern state university system when it comes to providing the breadth of knowledge that a liberal arts education is supposed to provide. I appreciate the way the proposal presents a version of that future unflinchingly.
My primary objection is to Step 4: Develop the Project-Based Learning System and Step 6: Integrate Human Mentorship and Collaboration, because they are based on the increasingly dubious notion that credentialing and learning are meaningfully related. We have before us the opportunity to abandon this fiction by automating the former and leaving the latter up to the student to figure out, maybe with a few mentors around to help. Why muddy the waters by attempting to automate what remains of learning?
Far better, I say, to create loose structures of in-person connections where students who are interested pursue their interests via projects under the mentorship of experts. Let the machinery do the credentialing as efficiently and effectively as AI can using digital platforms. Let the faculty and students figure out the learning without all the bureaucratic apparatus, standardization, and government control. So, I say CSU should optimize all the other steps Robbins lays out, but give the faculty and the students steps 4 and 6.
I'll try to do justice to this response with a longer piece soon. But again, I appreciate the willingness to think systematically about what CSU's embrace of AI could do for general education.
I really enjoyed reading this, as I am, like you, invested so deeply in digital literacy, albeit mine being more general populace and yours more industry geared. Really like finding this kind of content.
I have a few questions, if you're open to kind of answering/talking about it a bit more for me. For context, I dropped out of university, studied computer science with a major in games design, but only lasted two years before I became self-motivated and taught myself online + reading + people in the space, so my understanding of legacy institutions like universities and the education system is limited to the direct experience of it. However:
I think it's plain to see even from my context, that the current system isn't working well. Particularly where I am in Australia, where going to university is essentially exclusive to people who have intergenerational wealth. Even middle-bracket earners are not able to justify the costs of attendance as Australia not only slips in global relevance, but structurally our education system was simply not built for longevity as a glorified colonial satellite for fossil fuel extraction, now very much overstaying (stolen land) welcome.
I understand the angle of looking at it your way, but to me it comes across as a somewhat algorithmic education factory, and that only reproduces legacy structures of education, of which are now sufficiently outdated in the way they cannot facilitate the scale, breadth and movement of how information travels these days.
There's deep acknowledgement of the current system's inefficiencies, quality control issues, and the disruption AI poses to traditional assessment, but the proposed solution reads more like a techno-solutionist fantasy than a workable educational reform. This is coming from a fellow tech-enthusiast as well.
The microservices architecture discussion where "Learner Profile Service" and "Project Recommendation Engine" come into play, I don't think quite grapple with the extraordinary complexity of actually building these systems. There's no plug-and-play simplicity, anyone who's worked on large-scale technology knows this is pipe-dream, even as these processes simplify or more people learn, the upkeep and complexity grows with it in other ways. Also, the cognitive diffrences in those raised "digital-native" are not even fully understandable as yet, and I don't think we are prepared to deal with the fact that institutional learning may simply not be possible at all at scale in the future.
I think the real issue, as I kind of mentioned above, is to address the real issues - funding, adjunctification, administrative bloat, and all of this through the lens of future-proofing or at least making all of these things dynamic and flexible enough to accomodate the challenges of AI - while preserving what makes education transformative: human connection, intellectual community, and the unpredictability of learning but for a new and endlessly diverse group of students.
This makes me think about the human element more so, From my understanding, universities are complex social institutions with multiple stakeholders, competing interests, and deep cultural significance. The breezy dismissal of "thousands of human teachers of widely varying qualifications" reveals a kind of technocratic blindness to the diversity and actual administering and maintenance of teaching those teachers actually do.
I see bias, equity, and accreditation as afterthoughts to be "addressed" rather than recognising them as central challenges that should shape the entire design. How exactly does an AI system assess "Cultural & Equity Fluency"? These things aren't bugs to be patched, there's already and has long been an educational inequity as feature of our current system, and patching hasn't worked. AI doesn't solve the problem of education being used as a sorting mechanism for class reproduction in this way, in fact, it potentially amplifies it.
I'm wondering about your thoughts on this precisely. I don't have a single part of me that believes AI is even eligible to be anthropomorphised as it so often is and is a mathematical space and should be approached as such, but the maths of bias, equity and accreditation has been cruel and exploitative for some time. E.g. the pricing out of potential students and the slashing of departments and budgets across learning institutions. These don't disappear or become irrelevant with AI, they provide residual issues / wear and tear to new systems, and I don't people consider the impact of putting a new wheel on a shit car.
I'm genuinely so curious to hear your thoughts, and hopefully have a conversation, I'm desperate to reach out to deep thinkers on AI and the future and I am loving your posts and commentary for doing precisely that with the maths and language both equally considered - to me that's so important. Thanks!
Education is, at its core, human knowledge sharing. It’s also a two way street, in that it’s a gift to those who want it. Ergo, the issues around education are human problems and the solutions can only be human ones. AI won’t change motivation of students, their parents or educators. Systematising it’s delivery with AI will only bring about atomisation and insincerity, even with human guardrails.
This 6 step framework for Gen Ed curriculum is well designed and ahead of its time, rather like Woody Allen’s Orgasmatron . In the latter we have a product that removes all the inefficiencies and mess rather quick touchless service which is now so in vogue. Why pay for that expensive dinner and wine, have to endure bad dates and disappointment? Isn’t it better to focus on those important goals with critical milestones without having to again wash bed sheets you just washed. Or is all that messiness truly essential? Great teachers and mentors eat, sleep, endure bad weather, transportation issues, go boogie boarding, get sunburned, sculpt, raise kids, get injured, take nature walks and gather things to build holiday wreaths, maybe drink too much on occasion, laugh, experience joy. They infuse their experiences and wisdom into the lesson. Their students are impacted and ultimately better at comprehending and retaining over the long term while building a foundation for exploration. So sure isn’t it lovely notion to outsource the first 25.2 miles to a 24/7 data consuming emotionless machine and then waltz in for the last mile with streets lined with people cheering you to the finish line barely breaking a sweat. But what value is a diploma, with an outstanding transcript ensured with neutral algorithm driven maysurable results from Rosie Ruiz U?
I love this, but I don't think it's bold at all. The market for AI-driven ed apps is already growing at a 30%+ cagr over the next decade. Many apps are already supplementing and in some cases replacing teaching in primary education (see Alpha School or Synthesis App). Applying the same model to higher-ed subjects is a no-brainer and will happen within a decade. Students are already agitating for better learning models, even and especially at top schools (see https://poetsandquants.com/2025/07/23/were-not-learning-anything-stanford-gsb-students-sound-the-alarm-over-academics/). The writing is on the wall when it comes to learning tech. What I like about this essay is how you put it in the language of higher-ed admins. Personally, I'm curious how this tech can and will be used to reinvent the traditional university model and cost structure. I would anticipate some new universities that do the equivalent of Alpha School at the university level.
I like the project based learning aspect for several reasons: 1) It forces students to integrate work from various disciplines. As you note, that's how things work in the real world. For example, it's how engineers have to work. But there's not enough (or even any?) of that in the current system. 2) It makes other students the "target" of learning activity, so that the AI becomes a vehicle through which one interacts with others. 3) This also allows you to separate the thinking-things-through aspect of writing from the (important) minutiae of crafting individual sentences. Let the AI craft the sentences, but the students themselves have to do the thinking-through in the way they prompt the AI so that the final result is useful to their collaborators in the project.
Yes, students need to learn how to craft sentences. But few of us need anything more than workman-like prose in our jobs. Once a student has learned basic sentence-craft let the AI do the bulk of that so the student is free to do the higher level exploration and integration. They might even find that fun now that they don't have to fiddle with sentence structure all time.
P.S. I say this as someone who has spent a LOT of time fiddling with sentence structure. But then I like doing that. Most people don't.
This seems really interesting. I wonder what sort of organization do you intend to be building this? A tech startup? It seems like you need some way to build one education platform and apply it at many universities, and as far as I can tell there isn't really a good way to do that other than a for-profit company. Then you have an extra sort of problem, like who are you selling this to, and what's your go-to-market strategy.
Basically, the pitch for the actual product is convincing. The question of how do you get distribution is unclear to me.
An ambitious agenda. (1) You might add in continuous improvement as a core feature. As the system tries different projects, class schedules and feedback mechanisms, an analysis AI can be distilling lessons about what works where, and teaching those lessons to each microservice. (2) Are you or others piloting? It would be great to start with one set of competencies and find out how to do this and how well it works.
AI aside, this proposal's main intervention is to dethrone seat-time in credit-bearing classes. In essence, you're pulling American GE in the direction of the Bologna Process, with e-portfolios taking the place of high-stakes exams. I suspect exams would quickly sneak back into the model you're proposing; some skills can in fact be tested efficiently without projects. Not every competency needs a Quest!
I would be interested to see a version of this suited to a private middle and high school.
Possibly easier to introduce at a smaller scale, and starting with students who are just entering academic life.
I think you have to introduce it at a reasonably large scale because you're going to need a number of AI-savvy software engineers who are going to be fairly expensive. So the whole thing will cost a lot to get going and needs to be funded somehow.
Agreed that it would have to be an entire school, either a large private high school or a private liberal arts college -- I am guessing larger school, especially a public university, would have too much institutional inertia. You would a group of wealthy trustees who wanted the prestige of doing something big. That's how I see it happening.
I am a fan of Hollis Robbins's heterodoxy when it comes to AI and higher education, not because I agree with her, but because Anectotal Value offers ideas that challenge the usual terms of what Henry Farrell calls AI Fight Club, which pits those who see AI-based educational tools as our inevitable future against those who resist AI in large or small ways.
In proposing that CSU automate general education through a microservices architecture consisting of AI applications, Robbins offers a concrete alternative (a genuinely awful one) to the already existing structure of gen-ed (also genuinely awful). Actually existing systematic general education is no longer viable because OpenAI built Shel Silverstein's "Homework Machine" and offers it to college students at no cost. AI, it seems, is the future of the modern state university system when it comes to providing the breadth of knowledge that a liberal arts education is supposed to provide. I appreciate the way the proposal presents a version of that future unflinchingly.
My primary objection is to Step 4: Develop the Project-Based Learning System and Step 6: Integrate Human Mentorship and Collaboration, because they are based on the increasingly dubious notion that credentialing and learning are meaningfully related. We have before us the opportunity to abandon this fiction by automating the former and leaving the latter up to the student to figure out, maybe with a few mentors around to help. Why muddy the waters by attempting to automate what remains of learning?
Far better, I say, to create loose structures of in-person connections where students who are interested pursue their interests via projects under the mentorship of experts. Let the machinery do the credentialing as efficiently and effectively as AI can using digital platforms. Let the faculty and students figure out the learning without all the bureaucratic apparatus, standardization, and government control. So, I say CSU should optimize all the other steps Robbins lays out, but give the faculty and the students steps 4 and 6.
I'll try to do justice to this response with a longer piece soon. But again, I appreciate the willingness to think systematically about what CSU's embrace of AI could do for general education.
I really enjoyed reading this, as I am, like you, invested so deeply in digital literacy, albeit mine being more general populace and yours more industry geared. Really like finding this kind of content.
I have a few questions, if you're open to kind of answering/talking about it a bit more for me. For context, I dropped out of university, studied computer science with a major in games design, but only lasted two years before I became self-motivated and taught myself online + reading + people in the space, so my understanding of legacy institutions like universities and the education system is limited to the direct experience of it. However:
I think it's plain to see even from my context, that the current system isn't working well. Particularly where I am in Australia, where going to university is essentially exclusive to people who have intergenerational wealth. Even middle-bracket earners are not able to justify the costs of attendance as Australia not only slips in global relevance, but structurally our education system was simply not built for longevity as a glorified colonial satellite for fossil fuel extraction, now very much overstaying (stolen land) welcome.
I understand the angle of looking at it your way, but to me it comes across as a somewhat algorithmic education factory, and that only reproduces legacy structures of education, of which are now sufficiently outdated in the way they cannot facilitate the scale, breadth and movement of how information travels these days.
There's deep acknowledgement of the current system's inefficiencies, quality control issues, and the disruption AI poses to traditional assessment, but the proposed solution reads more like a techno-solutionist fantasy than a workable educational reform. This is coming from a fellow tech-enthusiast as well.
The microservices architecture discussion where "Learner Profile Service" and "Project Recommendation Engine" come into play, I don't think quite grapple with the extraordinary complexity of actually building these systems. There's no plug-and-play simplicity, anyone who's worked on large-scale technology knows this is pipe-dream, even as these processes simplify or more people learn, the upkeep and complexity grows with it in other ways. Also, the cognitive diffrences in those raised "digital-native" are not even fully understandable as yet, and I don't think we are prepared to deal with the fact that institutional learning may simply not be possible at all at scale in the future.
I think the real issue, as I kind of mentioned above, is to address the real issues - funding, adjunctification, administrative bloat, and all of this through the lens of future-proofing or at least making all of these things dynamic and flexible enough to accomodate the challenges of AI - while preserving what makes education transformative: human connection, intellectual community, and the unpredictability of learning but for a new and endlessly diverse group of students.
This makes me think about the human element more so, From my understanding, universities are complex social institutions with multiple stakeholders, competing interests, and deep cultural significance. The breezy dismissal of "thousands of human teachers of widely varying qualifications" reveals a kind of technocratic blindness to the diversity and actual administering and maintenance of teaching those teachers actually do.
I see bias, equity, and accreditation as afterthoughts to be "addressed" rather than recognising them as central challenges that should shape the entire design. How exactly does an AI system assess "Cultural & Equity Fluency"? These things aren't bugs to be patched, there's already and has long been an educational inequity as feature of our current system, and patching hasn't worked. AI doesn't solve the problem of education being used as a sorting mechanism for class reproduction in this way, in fact, it potentially amplifies it.
I'm wondering about your thoughts on this precisely. I don't have a single part of me that believes AI is even eligible to be anthropomorphised as it so often is and is a mathematical space and should be approached as such, but the maths of bias, equity and accreditation has been cruel and exploitative for some time. E.g. the pricing out of potential students and the slashing of departments and budgets across learning institutions. These don't disappear or become irrelevant with AI, they provide residual issues / wear and tear to new systems, and I don't people consider the impact of putting a new wheel on a shit car.
I'm genuinely so curious to hear your thoughts, and hopefully have a conversation, I'm desperate to reach out to deep thinkers on AI and the future and I am loving your posts and commentary for doing precisely that with the maths and language both equally considered - to me that's so important. Thanks!
Education is, at its core, human knowledge sharing. It’s also a two way street, in that it’s a gift to those who want it. Ergo, the issues around education are human problems and the solutions can only be human ones. AI won’t change motivation of students, their parents or educators. Systematising it’s delivery with AI will only bring about atomisation and insincerity, even with human guardrails.
This 6 step framework for Gen Ed curriculum is well designed and ahead of its time, rather like Woody Allen’s Orgasmatron . In the latter we have a product that removes all the inefficiencies and mess rather quick touchless service which is now so in vogue. Why pay for that expensive dinner and wine, have to endure bad dates and disappointment? Isn’t it better to focus on those important goals with critical milestones without having to again wash bed sheets you just washed. Or is all that messiness truly essential? Great teachers and mentors eat, sleep, endure bad weather, transportation issues, go boogie boarding, get sunburned, sculpt, raise kids, get injured, take nature walks and gather things to build holiday wreaths, maybe drink too much on occasion, laugh, experience joy. They infuse their experiences and wisdom into the lesson. Their students are impacted and ultimately better at comprehending and retaining over the long term while building a foundation for exploration. So sure isn’t it lovely notion to outsource the first 25.2 miles to a 24/7 data consuming emotionless machine and then waltz in for the last mile with streets lined with people cheering you to the finish line barely breaking a sweat. But what value is a diploma, with an outstanding transcript ensured with neutral algorithm driven maysurable results from Rosie Ruiz U?
I love this, but I don't think it's bold at all. The market for AI-driven ed apps is already growing at a 30%+ cagr over the next decade. Many apps are already supplementing and in some cases replacing teaching in primary education (see Alpha School or Synthesis App). Applying the same model to higher-ed subjects is a no-brainer and will happen within a decade. Students are already agitating for better learning models, even and especially at top schools (see https://poetsandquants.com/2025/07/23/were-not-learning-anything-stanford-gsb-students-sound-the-alarm-over-academics/). The writing is on the wall when it comes to learning tech. What I like about this essay is how you put it in the language of higher-ed admins. Personally, I'm curious how this tech can and will be used to reinvent the traditional university model and cost structure. I would anticipate some new universities that do the equivalent of Alpha School at the university level.
I like the project based learning aspect for several reasons: 1) It forces students to integrate work from various disciplines. As you note, that's how things work in the real world. For example, it's how engineers have to work. But there's not enough (or even any?) of that in the current system. 2) It makes other students the "target" of learning activity, so that the AI becomes a vehicle through which one interacts with others. 3) This also allows you to separate the thinking-things-through aspect of writing from the (important) minutiae of crafting individual sentences. Let the AI craft the sentences, but the students themselves have to do the thinking-through in the way they prompt the AI so that the final result is useful to their collaborators in the project.
Yes, students need to learn how to craft sentences. But few of us need anything more than workman-like prose in our jobs. Once a student has learned basic sentence-craft let the AI do the bulk of that so the student is free to do the higher level exploration and integration. They might even find that fun now that they don't have to fiddle with sentence structure all time.
P.S. I say this as someone who has spent a LOT of time fiddling with sentence structure. But then I like doing that. Most people don't.
Around the corner at my blog I've posted a dialog I've had with ChatGPT about PBL and writing: https://new-savanna.blogspot.com/2025/07/project-based-learning-pbl-and-ai.html
This seems really interesting. I wonder what sort of organization do you intend to be building this? A tech startup? It seems like you need some way to build one education platform and apply it at many universities, and as far as I can tell there isn't really a good way to do that other than a for-profit company. Then you have an extra sort of problem, like who are you selling this to, and what's your go-to-market strategy.
Basically, the pitch for the actual product is convincing. The question of how do you get distribution is unclear to me.
An ambitious agenda. (1) You might add in continuous improvement as a core feature. As the system tries different projects, class schedules and feedback mechanisms, an analysis AI can be distilling lessons about what works where, and teaching those lessons to each microservice. (2) Are you or others piloting? It would be great to start with one set of competencies and find out how to do this and how well it works.
AI aside, this proposal's main intervention is to dethrone seat-time in credit-bearing classes. In essence, you're pulling American GE in the direction of the Bologna Process, with e-portfolios taking the place of high-stakes exams. I suspect exams would quickly sneak back into the model you're proposing; some skills can in fact be tested efficiently without projects. Not every competency needs a Quest!
No I’m playing chicken with the State.
It's a good question!