
AI made the lean startup more important
AI makes experiments cheaper and faster to produce
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
Jeff Gothelf
Oct 15, 2026
11
4
3
Share
At a glance
AI makes experiments cheaper and faster to produce
More AI output does not mean better product decisions
Funding should follow behavior change, not polished artifacts
There are times when I feel I’ve been having a conversation about an idea for a while without great traction and it’s not until a better-known source mentions it that people start paying attention.
Last month HBR published its take on how AI is affecting the search for product-market fit and the concept of The Lean Startup as a whole. In the article titled AI Is Changing the Rules of Entrepreneurship by Victor Seidel, Bret Greenstein and Tom Davenport, they correctly call out that a single person with AI-powered tools can now generate product ideas (read: hypotheses), build prototypes (read: experiments), spin up fully functioning websites (read: market tests) and even run agents that act as their marketer and engineer.
What they are describing is The Lean Startup exactly, except that now this can be done in parallel for multiple ideas and for very little money. The authors call this new model “abundance entrepreneurship.”
In many ways they’re right. In terms of the ability to generate and test new ideas, there’s never been a better time to be an entrepreneur. There’s a table in the middle of the HBR piece that lists the difference between Lean Startup (aka “then”) and Abundance Entrepreneurship (aka “now”). This is where my opinion starts to drift from the core thesis the authors present.
Specifically, the table says the Lean Startup’s constraint was “capital and resources,” that the new constraint is “attention and selection,” and that the key skill is shifting from “hypothesis testing” to “signal filtering and judgment.”
What the Lean Startup was actually solving for
When Josh Seiden and I wrote Lean UX, we based much of our work on Eric Ries’s ideas and brought that mindset to product teams inside enterprise-scale companies. Capital was rarely their problem. The reason to build the smallest possible thing was that nobody knew whether customers wanted it and an experiment (née MVP) was the cheapest, fastest way to find out.
Scarce capital made that discipline urgent for founders, but the method existed to reduce uncertainty about customers and the value of early-stage ideas regardless of what type of company had those ideas.
AI has indeed collapsed the cost of building something small and testable. What it hasn’t done is alleviate the uncertainty and therefore risk around customers, their needs and the value of solving for those needs in a specific way. It’s not to say that building lightweight experiments was expensive or time-consuming in the pre-AI days. It wasn’t. However, the quality of those experiments can be exponentially better with the tools we have now.
So if we can build experiments faster and, in theory, get feedback faster, we’ve removed the capital constraint around learning (or at least the perception of it). We’ve also increased the amount of inbound signal, which informs our hypotheses as well.
What AI still can’t produce on its own, however, is evidence of what real customers actually do. That data still relies on real customer interactions. This is what Lean Startup was solving for. And it’s still the constraint we have today.
To their credit, the authors discuss part of this later in the article. They note that “valid market signals” are “still scarce.” They quote Steve Blank saying that AI “makes bad ideas go faster” and they warn that founders “may rely too heavily on insights from AI and neglect actual conversations with real or potential customers.”
They offer the idea of “simulating customer interviews” as a new capability founders might take advantage of to increase the pace of learning across multiple, simultaneous experiments. At the same time, they warn that models act in a sycophantic way to their users. So if you have a model simulating a user and it is also inclined not to tell a founder their idea is bad, the insight generated from these tools ends up delivering no measurable value to the entrepreneur trying to decide whether to kill, pivot or persevere.
In fact, it’s a very efficient way to manufacture exactly the false traction they’re worried about.
The part of the HBR piece I care about most is near the end where the authors write that “many large organizations have adopted the principles of lean startups” and now need to move toward abundance entrepreneurship themselves. In my experience most large organizations adopted the language of Lean Startup far more than the practice (everyone uses the phrase MVP, no one actually agrees on what it means). AI is about to make that gap far more expensive.
The early numbers are already pointing in that direction. Atlassian’s State of Product 2027 report, a survey of 1,000 product professionals, found that product managers say AI helps their teams ship faster, “but customers aren’t seeing value any sooner,” and 86% of PMs agree that deciding what to build matters more than ever. Cisco’s Jeetu Patel said much the same on HBR’s IdeaCast this month: “choosing what not to build is actually a far more important decision than deciding what to build.” And Christian Catalini, writing in HBR a week ago, put the economics of it in one line: “When execution is cheap, verification becomes more valuable.”
The abundance companies are currently creating is not in knowledge, learning or evidence-based decision-making but rather in the volume and velocity of their output.
Most of the artifacts and processes used by large companies to fund new initiatives were designed for a world where the cost of building was the limiting constraint. The business cases, proposal decks, strategy statements, PRDs, etc. — every one of those artifacts is now cheap to produce and looks remarkably polished and finished when presented.
But none of that polish will actually tell you if there is real, proven market and customer value for the proposed idea. So more projects are likely getting through the funding stages and into customer hands. This is certainly more “abundance” than “lean” and the net result is noise in the market, confusion for your customers and no clear ROI on your AI-infused product development process.
The fundamentals of Lean Startup matter now, more than ever. AI-powered tools can help you build learning much more quickly and it’s indeed this learning that will drive better judgment, filtering and decision-making in your company. It’s heavily dependent on the actual adoption of the Build/Measure/Learn mantra though. Given the cargo-cult nature of Lean Startup adoption across the business world, abundance entrepreneurship is likely to simply create “more stuff” rather than true customer and business value.
So if your goal is to adapt your company’s PDLC to account for this new era of cheap, abundant deliverables and accelerated learning, try this.
If you approve new work, or you’re the person asking for approval, here are four changes you can make this quarter:
Agree on the evidence before the work starts. Write down what a real customer will do differently if this new idea works and roughly when you expect to see the change. Connect the next round of funding to that customer behavior change.
Include learning efforts in parallel with building plans. Use the time-saving benefits of AI to help you build the kind of experiments that will teach you what you need to learn right now. Make sure time is allocated for that work and the deployment of those experiments. Then, don’t forget to measure the results of those experiments.
Label synthetic research for what it is. Simulated interviews and AI-generated personas can go in the deck as hypotheses to test. They can’t count as evidence for a decision.
Count what you kill. Abundance reveals what was never valuable in the first place. As the glut of new ideas comes in, keep track of the ones you didn’t build or ended up killing. This starts to tell you how fast the organization learns, which is the real value of the Lean Startup approach to product development.
AI has made experiments cheaper, faster and more polished. We must remember it has not made evidence of customer value any easier to get.
The companies that benefit from abundance will be the ones that use it to learn faster, not simply to build more.
Hit reply and tell me the last thing your team stopped building because of what a customer actually did and how long ago it was. And if AI was involved in helping you figure that out, share that too.
Jeff jeff@senseandrespond.co senseandrespond.co
This is the kind of work we tackle in Sense & Respond Learning courses: defining outcomes, testing assumptions and building product teams that learn before they scale output.
We have a full fall calendar of open public courses, including group and multiple-course discounts.
🇧🇷 Coming to Brazil: Lean Product Management for AI December 7 & 14, 2026 · Live online · Taught in Portuguese
A big part of this community is in Brazil, so I’m especially excited about this one.
Building AI products brings challenges that standard product methods don’t fully cover. In this live online workshop, Professional Scrum Trainer Matheus Gomes Reis shows teams how to apply Lean Product Management to AI, focusing on the places where AI products work differently. The fundamentals still apply. Customer value, clear outcomes and tested assumptions matter even more when the technology is hard to predict.
If you work with a team in Brazil, or know one that would benefit, please forward this to them.
Early Bird pricing ends November 10. → Register now
🇧🇷 Workshop em português: Lean Product Management para IA 7 e 14 de dezembro de 2026 · Ao vivo, online · Em português
Uma grande parte desta comunidade está no Brasil, por isso estou especialmente animado com este workshop.
Construir produtos de IA traz desafios que os métodos para produtos tradicionais não cobrem por completo. Neste workshop ao vivo e online, o Professional Scrum Trainer Matheus Gomes Reis mostra como os times podem aplicar o Lean Product Management à IA, com foco nos pontos em que produtos de IA funcionam de forma diferente. Os fundamentos continuam valendo. Valor para o cliente, resultados claros e hipóteses testadas importam ainda mais quando a tecnologia é difícil de prever.
Se você trabalha com um time no Brasil, ou conhece alguém que se beneficiaria, compartilhe este convite.
O preço promocional de pré-venda vai até 10 de novembro. → Inscreva-se agora
I’ve spent much of September teaching workshops with clients including a fun, full-day Product Discovery course with an AI-first organization in the south of Spain (could be worse!). I spoke at the AI Summit Barcelona, which was a mess logistically, though I had a good time giving a shorter version of my new talk, AI made it easy. Who decides what’s good? (Video coming shortly.)
I also took my younger daughter back to university and into her first apartment. Exciting times for her. We’re happy too though her summer break was so long we got used to having her around again. Empty-nesting problems, I guess.
On the work front, Josh Seiden and I are entertaining the idea of writing another book together. I didn’t think we’d write another book but when he suggested it to me, I was halfway into a steep climb on my bike and the oxygen deprivation probably swayed me a bit. Why write another book? Well, it dawned on us that, even though we work heavily in the product management field, none of our books are expressly about the practice of modern product management. Is there space for one more book on the topic? Maybe. We’re discussing it now. Stay tuned.
Exit West by Mohsin Hamid. Book club is next week. I haven’t started it yet. Got some reading to do this weekend.
The Grateful Dead, once again. I’ve got gigs with The Eurodead Orchestra coming up this fall and I need t
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。