The Man Who Disrupted Entrepreneurship Education Says AI Just Disrupted Him

On the first day of this year’s Lean LaunchPad class at Stanford, something happened that Steve Blank had never seen in 16 years of teaching.

Every team walked in with a finished product.

Not a PowerPoint slide. Not a wireframe. Not a napkin sketch. A finished, working product – the kind that, in previous years, would have represented weeks of effort and served as evidence that a team actually understood what it was building and for whom.

“We looked at each other and went, ‘Holy…'” Blank says, pausing in a way that makes the ellipsis do a lot of work.

Blank – serial entrepreneur, adjunct professor at Stanford, co-founder of the Gordian Knot Center for National Security Innovation, and the man widely credited with launching the Lean startup movement – had spent the better part of two decades arguing that the business plan was the wrong tool for building new companies. Startups, he insisted, don’t execute known plans; they search for unknown ones. That insight upended entrepreneurship education, spawned the National Science Foundation’s I-Corps program, and propagated into more than 100 universities worldwide.

THE SHELL-SHOCK MOMENT

Now, talking with Poets&Quants from London where he is teaching his Wicked Problems class at Imperial College, Blank is saying something he didn’t expect to be saying quite so soon: the methodology he built to replace the business plan needs to be replaced itself. Or at least radically rebuilt.

“The syllabus is in desperate need of revamping,” he says. “Watch out for next year.”

The Lean LaunchPad class was designed around a specific sequence. Teams arrive with a hypothesis. They get out of the building. They talk to customers – 10 to 15 interviews a week, every week, for 10 weeks. They build minimum viable products to test their assumptions. They pivot when the evidence demands it.

The MVP – minimum viable product – was a load-bearing element of that structure. Building one took real effort, real technical skill, and real understanding of what problem you were trying to solve. It was evidence. It meant something.

Then came Claude Code. ChatGPT. Replit. Lovable.

“An MVP is now equivalent to a PowerPoint slide,” Blank says. “Of course I expect you to use Claude Code or OpenAI to generate a software product, or even a hardware bill of materials. Okay. That’s the opening bid. Now what are you going to do?”

The problem, as Blank and his teaching team quickly diagnosed it, was that students in 2026 could generate a polished, finished-looking product in hours – and then mistake that product for proof that they understood their customer. They were skipping discovery entirely, jumping straight to the validation phase, and arriving at conclusions that felt solid because the artifact in their hands looked real.

“They assumed that all they needed to do was jump to the validation phase and skip discovery,” Blank says. “And so they pivoted late, because they assumed that a polished product meant product/market fit.”

In his blog post summarizing this year’s class, he put it plainly: it wasn’t the AI that was hallucinating. It was the teams.

THE SYNTHETIC INTERVIEW EXPERIMENT

To test the boundaries of the problem, Blank and his teaching team ran an experiment. They had students conduct synthetic customer discovery interviews – AI-generated conversations with simulated users – and then compared the results to their actual human interviews.

Half the synthetic interviews, it turned out, were hallucinating. The AI was generating plausible-sounding customer feedback that had no grounding in reality, and the students had no way to know the difference until they put the two sets side by side.

It was, in a sense, a pedagogical demonstration of the core principle that Lean LaunchPad was built to teach: you cannot learn what customers actually need without actually talking to them. The building you have to get out of has just gotten more comfortable and more convincing.

“We were trying a lot of experiments,” Blank says. “But the biggest one was putting some rigor around, no, no, no – the fact that you could generate multiple MVPs now means that you should be extending the things you’re doing. You should be looking for design partners, not MVPs. You should be looking for early customers by generating multiple MVPs.”

The insight is almost counterintuitive: AI makes it possible to build more, faster, which means the bar for what constitutes meaningful evidence has to rise correspondingly. Spinning up a finished product in a day doesn’t validate a hypothesis. It just raises the stakes for the customer conversation that has to follow.

AGENT OUTCOME FIT

There is a deeper disruption underneath the pedagogical one, and Blank is candid about how much he is still working it out.

For as long as product/market fit has been the north star of Lean methodology – the point at which a startup can say it has found a real customer with a real problem and a real willingness to pay – the underlying model has assumed a human being on the other end of the transaction. A user who clicks, adopts, pays, churns or doesn’t.

Agentic AI breaks that model.

“Imagine a supply chain product where you have an agent monitoring the supply chain,” Blank says. “And then another agent responsible for ordering. Another agent looking at company policies about bidding. Another agent – whatever. You could build a complete replacement for classic supply chain enterprise software that doesn’t involve a human being, other than making sure it doesn’t put the company out of business.”

That is a different architecture entirely. And it requires a different vocabulary. Blank has started using the phrase “agent outcome fit” – deliberately echoing product/market fit – to describe the new north star: not whether a human user will adopt your product, but whether your agent produces the right outcome in the real world.

“In some cases, it’s no longer even product/market fit,” he says. “In some cases, it’s agent outcome fit.”

Two teams in this year’s Lean LaunchPad class actually built toward that vision, Blank says – designing systems in which agents, not users, were the primary actors. The enterprise customers they were pitching to weren’t always ready for it. The students could see where things were going. “It was pretty obvious to us and one of the student teams that that’s where the world was going to go,” he says.

THE REVAMP

Blank is spending the summer redesigning the syllabus, working through it with his teaching teams across all his classes. He is clear about the direction even if the final shape isn’t settled yet: AI tools are not going to be restricted or discouraged. They are going to be pushed harder.

“Don’t just do a better version of what you would have done last year,” he says. “Embrace these tools with alacrity.”

The core of getting out of the building, he insists, will survive. The need for ground truth – for actual human contact with the messy, unpredictable reality of customer needs – is not diminished by AI. If anything, the synthetic interview experiment suggests it is more important than ever, precisely because AI makes it so easy to generate convincing substitutes. When a team can spin up 50 websites or generate 30 apps in an afternoon, the temptation to skip the conversation that actually matters only grows.

But the artifacts around that core – the MVP, the syllabus structure, possibly even the language of product/market fit itself – are all in play.

THE IRONY IS NOT LOST ON HIM

Last October, the Strategic Management Society gave Blank what he describes with characteristic understatement as “the innovator of the century or something award or whatever.” Around the same time, the Journal of Management devoted an entire issue to his work – an institution that had spent decades as a bastion of exactly the kind of management orthodoxy Blank had spent his career disrupting.

“I call that the year hell broke loose,” he says, with a laugh. “Now that everybody, even the most conservative journal, adopts it, is the time that it’s going to change again.”

He tells a story about the scientist who developed the theory of continental drift – dismissed, ridiculed, and finally vindicated – who asked his Ph.D. advisor how to get people to accept a new idea. “Sometimes,” the advisor told him, “you need to wait for them to die.”

Blank spent years waiting for business schools to accept Lean. Now they have. And Lean is already moving again.

“There’s no one better to shoot it in its head than the guy who came up with it,” he says. “By shooting it in its head, hopefully it doesn’t all have to die. We could adapt it into a rapid AI methodology or whatever we want to call it. But it needs to change.”

He expects that process to take a year, maybe two. Not the quarter-century it took for Lean to become orthodoxy in the first place. That pace, he says, is itself evidence of how much has changed.

“The tools are getting exponentially better at a rate that just never existed before,” he says. “I think AI is just another major force multiplier – probably on the scale of writing. What happened to education when you could write rather than oral? I think that’s where we are.”

The next version of Lean – whatever it’s called, whatever it looks like – is being invented now, in real time, by Blank and his teaching teams at Stanford and Imperial College and wherever else this methodology has taken root. He is, as he puts it, at the bleeding edge of learning what this stuff is.

He seems to find that more energizing than alarming.

“I can’t wait,” he says. “That’s kind of fun. Not that I’ll get it right. But when I show people what it is we’re going to do, I’m sure we’ll get lots of comments and feedback, and it’ll act as a template – much like Lean did for everybody else making better ideas.”

Watch out, he says. Next year’s class is going to look very different.

DON’T MISS STEVE BLANK: THE CLASS THAT CHANGED HOW ENTREPRENEURSHIP IS TAUGHT and STEVE BLANK: LEAN MEETS WICKED PROBLEMS

© Copyright 2026 Poets & Quants. All rights reserved. This article may not be republished, rewritten or otherwise distributed without written permission. To reprint or license this article or any content from Poets & Quants, please submit your request HERE.