Founders Can Now Build Before They Raise. The Accelerators Still Want Them Anyway.

Berkeley Haas is teaching MBA students a new rule of startup economics: do not raise money until you have to.

The course is AI Entrepreneurship, co-designed by Alex Zekoff, who co-founded Thoughtful AI and raised a growth investment from New Mountain Capital. His pitch to students is not about AI replacing jobs. It is about what changes when the cost of building drops toward zero.

The bottleneck used to be that to build products, you had to hire engineers, Zekoff told the class. Now you can build prototypes and test in the market and get a signal before you go raise and build product capital. (Haas Newsroom)

The course has the infrastructure to make that case seriously. More than 130 mentors from Nvidia, Google, OpenAI, and Anthropic support the program. By Demo Day on April 23, the numbers were concrete: 27 teams made it through from over 300 applications. Fifty-seven MBA and evening students enrolled, and in 13 weeks they built and shipped real products. (Haas Newsroom)

Twenty of the Demo Day teams committed to keep building. Six were accepted to accelerators including Berkeley SkyDeck. (Haas Newsroom)

What landed was not the metrics. It was what one student said afterward.

Daniel Humala, MBA 27, founded Courtship. He told the class: I do not feel like I need the money yet. In terms of building a product, AI can do this. I am not sure whether we actually need to fund it. (Haas Newsroom)

That sentence is worth sitting with.

The standard playbook for a decade has been: raise money, hire engineers, build product, iterate. The new playbook the course is selling looks like this: build with AI first, test in the market, get signal, then raise if you need to scale.

Zekoff is not the only person who has noticed this shift. Across the startup ecosystem, investors and founders have begun describing the same pattern. AI has dramatically reduced the cost of building a prototype and testing whether a product has market pull. The question is whether the rest of the startup infrastructure has fully registered what that means.

Accelerators built their model on the assumption that early-stage companies needed the infrastructure and mentorship they provided to get to first proof. Micro-VCs wrote small checks in exchange for early access to founders who needed those checks to pay engineers. The thesis made sense when the product was the hard part.

It is not always the hard part anymore.

This is not a clean narrative. The people who will build the most durable companies in this environment will still, in many cases, need significant capital to build distribution, hire talent, and construct competitive moats. AI changes the cost of one part of starting a company. It does not change the cost of winning a market.

It is also true that the VC ecosystem exerts real pressure on founders to raise, regardless of whether they technically need to. The network, the signaling, the competitive dynamics of a market where your competitor raised a large round are not solved by better prototyping tools.

And the students who will succeed here are not necessarily the ones with the best AI workflows. They are the ones who find the best problems to work on.

The Haas course sits inside the Berkeley LAUNCH accelerator, which has a track record of 250-plus startups that have raised more than $1.4 billion collectively. (Haas Newsroom) That institutional infrastructure matters in ways AI cannot replicate: peer networks, alumni connections, proximity to Sand Hill Road capital. The course teaches students to use all of it, including the AI.

Zekoff raised a growth investment from New Mountain Capital in April 2025. By May 2025, New Mountain merged Thoughtful AI into Smarter Technologies as part of a private-equity rollup, a structure that keeps the investor relationship intact rather than closing out the founder's stake.

What the course is actually arguing is that the unit economics of starting a company have changed in a specific, concrete way. The cost of building has dropped. The cost of distributing and competing has not. The new strategy is to use the cheap build phase to move as fast as possible before you encounter the expensive phase.

Whether this is right will be answered by the next decade of startup formation. The MBA students at Haas this spring are running the experiment. Their results are still being tallied.

The story behind this shift is not that venture capital is dead. It is that the rules written for an era when you had to hire people to build things are being quietly updated for an era when you do not. The students in Zekoff's class are learning the revised version first.