Can AI create a comprehensive degree program proposal in the time it takes to grab a coffee?
What started as an idle question got me thinking about how artificial intelligence stands to upend how we best-serve students in higher education
If emerging AI systems can design, refine, and help deliver, undergraduate degree programs that are far better designed and far more effective than those created by faculty committees alone, do we owe it to future students to ditch tradition in favor of emerging capabilities?
This, I must confess, is not the question I started out with as I began working on this article. But it’s one that I’m finding it hard to let go of, having spent the past couple of days working with Claude Code on designing a comprehensive plan for a new degree program.
To be honest, I didn’t even set out to design a new degree. It was just an idle exercise in seeing what’s possible with the latest wave of agent-based AI platforms. But having asked Claude Code the question, I’m finding it increasingly difficult to ignore the result.
Before I bury the lede any further, you can open/download the complete two hundred and twenty three page degree program proposal that Claude Code produced below — complete with design philosophy, program architecture, learning outcomes, career pathways, value propositions to students, parents, and employers, and detailed syllabi for 18 core courses and a capstone:
As you read it, take it from me as someone who does this for a living, this is impressive.
The real story here though is what I learned working with Claude Code on this, and why it got me thinking more deeply than I expected on what we owe our students.
Setting the scene
I’ve been using the desktop version of Claude Code for a few weeks now, and I’ve been impressed by its ability to break projects down into sequential tasks, assign multiple agents to these tasks, and orchestrate them toward delivering a final product.
For anyone who’s use the latest iteration of Claude Code, OpenAI’s Codex, or similar systems, you’ll know already that these are nothing like your average browser-based AI or AI bot. Here, I must confess I’m a bit of a novice user compared to some of my grad students. But I’ve intentionally kept things that way as I’m interested in seeing what users with little time or patience for engaging with technical wizardry can achieve with easily accessible AI platforms.
With this in mind, I’ve been exploring different ways of leveraging Claude Code as something of a side project.
The desktop version of Claude Code allows you to engage with it via a text box, much like Claude in a web browser. But there the similarities end. Claude Code (and similar systems) can read and write files to your computer, draw on “skills” that allow them to achieve an increasingly wide array of tasks, plan strategic approaches to exploring and executing projects, launch and coordinate multiple AI agents to carry out specific tasks, write and execute code as they do this, and much more.
As a result, they are far more powerful at executing complex tasks than a simple browser-based AI or a single AI bot or agent.
They also, it has to be said, only represent the tip of the iceberg when it comes to what more advanced agent-based AI implementations are capable of. Which means that everything I’m writing about here is just the tip of the iceberg of what a more sophisticated agent-based approach can achieve in the right hands.
The Process
Against this background, a couple of days ago I somewhat idly asked Claude Code to develop a “mature plan for a new undergraduate degree” (I’ve included the complete prompt below).
This wasn’t completely out of the blue. A few years ago I looked into the idea of developing a degree program around navigating advanced technology transitions, and started to flesh out some ideas. But I never got much further than jotting down a few thoughts and concepts.
This was well before the current emergence of easy-access LLM-based AI systems though.
Revisiting this, I was curious just how far — and how fast — Claude Code could take the idea and run with it.
And so I opened a new project, and asked the following (bad grammar and all — and I’ve only just spotted “curse-specific learning objectives!”):
I want you to develop a mature plan for a new undergraduate degree. This is a complex task that will require multiple steps. I have given you some of them below (as well as context), but you will have to research what the elements of a very strong undergraduate program are and build the program around these.
Context and guides:
The undergraduate degree should be a 4 year degree at a research university - assume it is a venture between engineering and a business school that also intersects with arts and humanities
It should be grounded in pedagogical and learning/education design best practices, including having clear, outcomes-aligned and assessable learning objectives and skills development
It should go deep on core courses, including detailed syllabi and and curse-specific learning objectives/skills
It should leave placeholders for electives - maybe categories
It should align very closely and realistically with career opportunities and pathways.
And above all it should be grounded in deep research so that the program is implementable and not just a paper exercise
As an area of focus, start with the idea of navigating technology transitions in a technologically complex world and hone this to something that is likely appeal to prospective students, parents, and employers, provide sufficient content, quality and heft over 4 years of study, and differentiate itself from the competition.
This is a very large task that will require a detailed and multi layered plan and multiple agents.
Any questions?
After a couple of clarifying questions on process, Claude Code got stuck in, and started down the process that led to the document above.
The process itself was pretty straight forward:
The prompt (above) followed by four clarifying questions
Me grabbing a coffee (that wasn’t just rhetoric in the title) — around 20 minutes to get to the nearest campus Starbucks and back
Claude conducting initial research, launching 5 agents, and delivering the first version of the proposal — captured across 23 markdown files and 48,000 words after undergoing several self-initiated internal reviews and edits. (As I was not being that sophisticated, this took a little longer than 20 minutes as I had to manually grant access to my laptop for various operations, but the actual work time was well within the coffee run window.)
Me launching a second project in Claude Code and asking it to conduct four detailed reviews of the first draft from the following perspectives:
Academic/pedagogical — does the degree program hold together pedagogically
Prospective employers
Prospective students
The parents of prospective students
Me providing the initial Claude Code project with the reviews and asking for an updated proposal.
Me asking Claude Code to package the revised version as a MS Word document (I was still finishing my coffee at this point)
Me providing feedback to Claude Code on formatting and key components of the content and focus, and asking for an updated draft (there were a couple of iterations here).
Me going through the final proposal manually and refining the formatting while making the occasional edit.
Far and away the longest part of this process was my editing — something like a 10:1 ratio of my time to Claude Code’s.
At the end of the process I asked Claude Code for a session audit. The final markdown files before their translation to a Word doc and my final editing contained over 66,000 words spread over around 215 pages, were the culmination of 334 files generated by Claude Code, and were the product of 7 autonomously designed and deployed sub-agents addressing curriculum structure analysis, external research, change audits, internal consistency (2 sub-agents), codebase exploration, and research synthesis. Across the project, Claude Code called on specific tools over 300 times to complete the tasks it assigned itself.
This represents a level of complexity and orchestration that no web-based LLM or single-agent chatbot could get close to. And the actual time Claude Code spent on this was just an hour or so tops.
But was the proposal any good, or did I just end up with 200+ pages of AI slop?
I’m sure some readers will disagree with me here — on principle if for no other reason — but based on well over a decade of teaching, developing courses and programs, and academic leadership in higher education, I am comfortable saying that the resulting document, while far from perfect, far surpasses most degree-planning documents I have seen emerge from more conventional processes..
So what does this all mean?
In some ways this was a relatively straight forward task for a multi agent LLM-based AI. Effective undergraduate degree program development isn’t rocket science. but it is hard work. And it does require knowledge of pedagogical and program design best practices, a good handle on domain knowledge and how to integrate across domains while tying this to learning process and outcomes, a professional understanding of degree-to-careers pathways, and the ability to expertly and simultaneously coordinate research, development, and drafting, across multiple dimensions.
These are challenging for groups of educators to achieve — especially if they represent just one of many responsibilities they are juggling. And they are well beyond the capabilities of single AI agents and bots (although these are adept at producing content that looks good, but is not).
But they are, in principle, relatively straight forward for multi-agent AI systems.
And the resulting proposal supports this.
If you haven’t looked at it (and I would encourage you to do so), the proposal builds on sound design and learning principles to deliver a deeply integrated program that has all the hallmarks of providing students with successful career pathways. It even includes information on how to market to prospective students, parents, and employers, while addressing how students might position themselves to demonstrate their new skills and abilities.
Of course, it’s not perfect. With time (and I had to resist working on it further) I would want to go through a few more iterations to further develop/refine the program. Some of the syllabi definitely need some work. And this is a blank-sheet proposal that assumes any university where it is implemented will build it up from scratch with the necessary faculty and staff — a luxury that few universities have.
That said, as a starting point, it is very good indeed. And this is where it got me thinking about what we owe our students.
What we owe our students
Here, I must confess that worry that many existing undergraduate degree programs are not as good as they could be — especially at research-focused universities, where something more akin to a “trickle down” model of education from world experts in their fields to young, open minds has been adopted.
This, of course, is a gross over-statement, and many universities — my own included — take education very seriously. And yet the reality is that many faculty are thrust in front of student with no training on how to teach, placed on degree committees with no knowledge of program design best practices, and charged with creating career pathways having never had a career outside of academia.
The result is a system that is functional but not necessarily optimal.
What, though, if using agent-based AI systems could help address these shortfalls. Maybe by producing robust drafts that reflect best practices. Or ensuring clear programmatic through-lines from entry to career. Or addressing value propositions to multiple stakeholders. And, of course, helping ensure student success comes before academic hubris.
These are all options that don’t replace humans, but rather vastly enhance their professional capabilities. And here it’s worth noting that the Claude Code generated proposal above isn’t good because Claude Code in isolation knew what a great degree program looks like, but because I was able to provide expert direction, feedback and evaluation along the way.
Even if such capabilities are just used to increase the quality of program development, surely we owe this to our students. Otherwise I worry that we risk selling them something that is far inferior to what it could be — remembering that this is a life-investment for many students who can barely afford it (or not in some cases) — while kidding ourselves that they are benefitting from our tricked-down wisdom.
Of course, we could just fall back on arguments around the sanctity of human intellectual labor and the inviolable standing of academics. But at some point students are going to start voting with their feet.
And given the choice between a human-made degree that seems to go no-where and an AI augmented one that does, I suspect I know which way they’ll lean.
But even before we get there, I’d like to think that we owe it to them to put their success before our own traditions and egos, and utilize emerging capabilities in ways that provide them with the education, experiences and insights that will enable them to thrive.
And of course, if we can achieve much of this in the time it takes to grab a coffee, so much the better!