Kill It Before the Market Does
Description
A sequel to A Million New Founders, No Net. Part of the answer is on the founder's side: because AI made building nearly free, building stopped being evidence of merit, and we sink cost into software nobody wants. The disciplines that matter become cost awareness, lean validation (test demand before building), default pessimism (pre-mortems), and actively seeking brutal adversarial criticism as a cheap stress test — manufacturing the credible 'no' the sycophantic AI will never volunteer.
Sources & further reading (7)
- CB Insights: Why startups fail — top reasonshttps://www.cbinsights.com/research/report/startup-failure-reasons-top/
- BLS Business Employment Dynamics (establishment survival)https://www.bls.gov/bdm/
- IEEE Spectrum: Why Software Fails / Trillions spent on failing projectshttps://spectrum.ieee.org/why-software-fails
- RAND: The Root Causes of Failure for AI Projects (Ryseff et al., 2024)https://www.rand.org/pubs/research_reports/RRA2680-1.html
- Altar.io: The lean customer-feedback playbook for foundershttps://altar.io/validating-your-startup-idea-lean-customer-feedback-playbook/
- Pre-mortem (Gary Klein; prospective hindsight)https://en.wikipedia.org/wiki/Pre-mortem
- Science: Sycophantic AI decreases prosocial intentions and promotes dependencehttps://www.science.org/doi/10.1126/science.aec8352
Script
Cold open
Last time: the weekend founder, the AI cofounder that called the idea brilliant, and the one voice missing — a credible NO. We said the machine is built never to give it. So tonight, the uncomfortable fix. What if you went and got that NO yourself — on purpose, brutally — before you wrote a single line of code?
Frame
Here's what changed under our feet. AI made building so cheap that building stopped being proof of anything. A working prototype used to mean a team believed in it for a year. Now it means someone had a free weekend. And that's a trap: when the prototype costs nothing, you build BEFORE you check whether anyone wants it — and then you defend the thing you already built. The discipline we need isn't more building. It's the willingness to kill the idea first.
If building got easy, what's the real test a prototype never passes?
Remember the number: the top reason startups die isn't running out of money — it's building something nobody needed. Forty-two percent. Building was never the test. About eighty percent of businesses survive year one, but only half reach year five — and most of that attrition is products meeting a market that just shrugs. The hard part was always demand. And demand is the one thing a prototype cannot prove.
Does AI fail more often — or just let us start more doomed things?
Now watch what cheap building does — and what it does NOT do. It doesn't make projects fail more OFTEN. Software has failed at a brutal clip for decades; the Standish Group still calls roughly one in five projects an outright write-off, and AI fares even worse — RAND finds more than eighty percent of AI projects fail, about double the rate of ordinary IT. But the rate was never the story. The DENOMINATOR is. When a build costs a free weekend instead of a funded year, we start far more doomed things — the same grim failure rate, times a vastly bigger pile, is a much taller mountain of sunk cost. US firms already waste tens of billions a year on software nobody wanted; cheap building just pours more onto the heap. And notice — this isn't really an AI problem, it's a venture problem; it just bites hardest wherever AI has made starting nearly free.
How do you test the water without building the boat?
So flip the order. Before you build, test the water — cheaply. A landing page that asks for the sale. A pre-order before the product exists. Ten real conversations with the people who'd actually pay. This is the boring core of lean validation, and it aims straight at that forty-two percent. It's cheap to test an idea and ruinously expensive to be wrong about one. A prototype proves you CAN build it; a pre-sale proves someone WANTS it. Only one of those is scarce.
How do you manufacture the pessimism the AI won't give you?
Then go further — invite the pain. Run a pre-mortem: assume it's eighteen months later and the thing failed spectacularly, and write down exactly why. Imagining the failure as already real makes you far better at naming its causes. Then take the idea to the harshest critic you can find and ask them to destroy it — not soften it, destroy it. Because if it can't survive an hour of brutal, adversarial criticism, it will not survive a year in a market that owes you nothing.
Turn
And here's the turn. Adversarial, brutal criticism isn't cruelty — in this new world it's the cheapest, fastest stress test you will ever get, and it's the exact thing your AI cofounder is built never to give you. So manufacture it. Stop asking the machine 'is this brilliant?' and start ordering it: steelman the case AGAINST this; who already does it better; who actually wants it; why is it dead. Cost awareness, lean tests, a default of pessimism — that isn't timidity. When building is free, the only scarce discipline left is the courage to kill your own idea before the market spends a year doing it for you.
Closer
Back to the weekend founder. Same Friday, same AI, same free weekend. But this time the first prompt isn't 'build me this.' It's 'tear this apart — tell me why it fails, and who's already done it.' Maybe the idea survives, sharper. Maybe it dies in an afternoon… and saves a year. The net we were missing wasn't only out in the world. Some of it is the NO you were too hopeful to ask for.