The public conversation keeps circling around whether university is becoming obsolete. I think that question is too broad to be useful. University is not obsolete in the simple sense. It still gives people time to read, write, argue, mature, meet peers, and build intellectual confidence. Those things matter.
The sharper question is whether university is still training the reflexes that the economy will reward as AI gets better. That is a different question. It is not about whether learning matters. Learning matters more than ever. It is about whether the things universities commonly evaluate are becoming easier to simulate.
Historically, university signaled that someone could perform legible cognition. They could read material, synthesize it, structure an argument, write clearly, cite sources, and present a coherent view. That was a valuable signal because producing that kind of artifact used to require a certain amount of intelligence, effort, and discipline.
AI weakens that signal. Not because it makes intelligence irrelevant, but because it makes the appearance of structured thought cheap. A report, deck, memo, market map, positioning document, literature review, or strategy outline no longer proves what it used to prove.
So the real question becomes: when polished cognition becomes abundant, what remains scarce?
My answer is: context, judgment, taste, ownership, and evidence.
I am saying this from a specific vantage point. I am a founder building in a fast-moving AI market, where attention is expensive and where useful work has a very clear shape: it reduces uncertainty, reduces workload, or creates evidence.
That is not every environment. A university, law firm, hospital, government agency, research lab, or large enterprise may reasonably reward slower forms of analysis, documentation, and coordination. Not every institution should behave like a startup.
So this should be taken with a grain of salt. I am not arguing that everyone needs to become a startup operator or that universities are worthless. I am describing a mismatch I see from the founder side, where the cost of attention is high and the cost of vague work is much higher than people realize.
But I also think this founder lens is becoming less niche. AI is making small teams more capable. It is compressing the cost of production. It is forcing larger organizations to care more about speed, ownership, and proof. In other words, more of the economy is starting to inherit startup-like expectations.
That is why this matters beyond startups. The old bridge from university to work assumed that institutions could absorb low-context artifact production while people slowly learned judgment. AI is putting pressure on that bridge.
University often trains people to make their thinking visible to an evaluator. This is understandable. A professor cannot grade your private understanding. They need an artifact. So students learn to create artifacts that make cognition legible.
This creates a very specific habit. When uncertain, expand the frame. Add context. Add caveats. Show research. Map stakeholders. Propose next steps. Demonstrate that you understand the complexity.
Inside university, that often reads as maturity. It proves you are not naive. It proves you know the problem is complex. It proves you can hold multiple variables in your head.
But in work, especially early-stage work, this can become a trap. The person receiving the artifact does not need proof that you can see complexity. They need part of the complexity to go away.
Slides are the easiest place for this habit to hide. A sequence of bullets can feel like movement even when nothing has been decided. The counter-move I keep coming back to was blunt: "We don't do PowerPoint."2 A page has fewer hiding places. The reasoning has to survive before the meeting can move.
This is the distinction universities often under-teach: making thought legible is not the same as making thought useful.
AI is extremely good at legibility. It can make half-formed thoughts sound structured. It can turn a vague prompt into a coherent memo. It can generate headings, tradeoffs, frameworks, examples, objections, and recommendations.
That means the economic value of merely looking structured is falling. The artifact may still be useful, but it is no longer a strong signal by itself. A polished deck can now mean insight, or it can mean prompt fluency plus light editing. The reader cannot tell from polish alone.
This creates a provenance problem for knowledge work. Before AI, effort was embedded in the artifact. If someone produced a long, coherent analysis, you could infer that they had done some amount of thinking. Now the artifact no longer carries the same proof of thought.
The market will respond by moving the proof closer to reality. It will ask not "does this look intelligent?" but "did this change anything?" Did a user reply? Did a customer convert? Did a bug get fixed? Did a decision become easier? Did the work reduce someone else's burden?
AI does not remove the need for thinking. It removes the subsidy for treating an artifact as sufficient evidence of thinking.
A lot of people still talk about AI as if the main question is productivity: can we produce more code, more copy, more documents, more designs, more analysis? That is true, but it is not the most important point.
The deeper shift is that production is becoming less bottlenecked than selection. The hard question is not "can we make something?" The hard question is "what should be made, why, for whom, and what would prove that it mattered?"
This is where judgment becomes more valuable. AI can generate options. It cannot automatically know which option fits the current company, customer, product, timing, culture, constraints, and risk. That knowledge lives in context.
This is also why generic strategy becomes less valuable. A model can produce generic strategy. A human has to know which generic strategy is irrelevant, dangerous, premature, politically impossible, technically impossible, or simply not worth the attention.
A credential proves you learned the accepted map. But valuable work often starts when the map becomes suspect. "What important truth do very few people agree with you on?"1 is not answered by polish. It requires noticing where consensus has become cover for not looking closely. AI can summarize consensus all day. It cannot make a person willing to stand behind a live disagreement.
The scarce person is not the person who can produce more artifacts. The scarce person is the person who can decide which artifacts should not exist.
The uncomfortable implication is that AI is pressuring many traditional entry-level tasks. A lot of junior knowledge work used to be artifact work: summarize this, research that, make the first draft, build the deck, prepare the brief, write the notes.
Those tasks were not perfect, but they served a function. They gave junior people a way to be useful before they had deep judgment. They also gave institutions a way to train them gradually.
AI makes that bargain weaker. If the first draft, summary, scan, and deck can be produced by software, the junior person has to justify themselves somewhere else. They need to bring taste, context, initiative, or contact with reality sooner.
That is a much harsher world for graduates. The market is asking for ownership earlier, while universities often still train people to produce evaluable artifacts rather than own consequences.
Ownership is the part that does not fit cleanly inside the artifact. It is choosing the next move when the brief is incomplete. It is noticing that the customer did not reply, that the metric moved for the wrong reason, that the handoff created more work than it removed. It is staying with the thing after the memo looks finished.
This is why the "university is obsolete" debate misses the sharper issue. The degree may still have value. But the protected ramp from school-style output to workplace value is getting shorter.
The common assumption is that startups are uniquely demanding and big companies will continue to absorb everyone else. I am not sure that holds.
Big companies still have major advantages: distribution, trust, capital, procurement, legal infrastructure, customer relationships, and regulatory knowledge. Those advantages are real. But AI changes their incentive to tolerate low-leverage work.
Historically, large organizations could absorb vague cognitive labor because they had layers. One person made the deck. Another interpreted it. Another turned it into a plan. Another managed execution. The system could hide inefficiency because coordination was already part of the institution.
AI makes that less defensible. If a human is mostly producing summaries, slide structures, market scans, and first drafts, the organization will increasingly ask why software cannot do most of it.
So even big companies will push people toward higher agency. Not because they become culturally like startups, but because the economics of human cognition change. Humans need to justify themselves closer to judgment, trust, accountability, and reality contact.
The future worker is competing on two fronts: against AI for artifact production, and against AI-native operators for speed.
The useful worker in this environment is not necessarily the person with the broadest analysis. It is the person who knows how to convert uncertainty into evidence.
They do not need the whole map before moving. They find a small path. They do not respond to every unknown by expanding the scope. They narrow until action is possible.
If they do not understand the whole market, they pick one segment. If they do not know the whole strategy, they test one message. If they lack internal context, they use available context. If the problem is too large, they make it smaller until they can own it.
This is the core distinction.
University often rewards expanding the problem until understanding is visible. Work increasingly rewards shrinking the problem until consequence is possible.
The answer is not to make university less intellectual. That would be the wrong lesson. We need more real thinking, not less. We need writing, history, theory, critique, abstraction, and deep reading.
But intellectual work has to be connected to consequence more often. Students should not only be asked, "What is your analysis?" They should also be asked, "What did you try? Who did you talk to? What changed? What did reality tell you? What did your first theory miss?"
A framework is useful if it sharpens action. A report is useful if it makes a decision easier. A deck is useful if it reduces uncertainty. Analysis is useful if it changes what happens next.
The strongest students in the AI era will not be the ones who can merely produce the cleanest explanation. They will be the ones who can move between explanation and intervention.
They will know when to expand and when to narrow. That may become one of the most valuable skills of all.
For students, the practical advice is simple: do not only build a portfolio of artifacts. Build a portfolio of changed states.
Show what you shipped. Show who replied. Show what failed. Show what improved. Show what you learned from reality. Show what you owned without someone else packaging the path for you.
For employers, the filter is also simple. Do not only ask whether someone sounds smart. Ask what they do when they lack context.
Do they expand into theory, or narrow into action? Do they ask for the whole map, or find a small path? Do they create dependency, or create evidence? Do they make the next step clearer, or make someone else responsible for it?
That moment reveals more than the artifact.
University is not obsolete. But the old signal is becoming less sufficient.
For a long time, polished demonstrations of understanding signaled promise. AI weakens that signal because it makes those demonstrations cheaper. At the same time, AI makes small teams more capable and pushes work toward speed, ownership, and evidence.
The market will increasingly reward people who can do something more specific: use understanding to create evidence.
That is the gap.
It is also the filter I care about when thinking about who I want to build Open WebUI with. Not who can produce the cleanest artifact in isolation, but who can turn ambiguity into shipped product, earned trust, and proof in the world.
University often proves that you can understand.
Work increasingly asks whether you can make something true.