Two important simultaneous important trends in how startups get customers: Make software agents want Use agents to make people want software

someone ran 76,948 live domain checks on AI-generated startup names. 45.7% had nothing free on any extension. saas names: 63% taken. consultancy: 64%. pet grooming: 19%. ideas are cheapest exactly where everyone is standing.

$10.3 trillion. that is the projected AI buildout through 2032, per brookings. larger relative to GDP than the railroads, electrification or the interstate highways. amount reserved for asking customers whether they want what gets built with it: rounding error. the largest untested assumption in economic history.

One honest customer conversation can save months of development.

everyone is sitting on one idea they never tested. not because testing is hard. because right now the idea is still good, and the moment you test it, it has to be true.

Eric Schmidt says UIs will disappear. Dylan Field says the market will prove him wrong. the fun part is neither of them gets a vote. the people who decide this never argue about it, they just keep clicking on whatever works.

an AI persona is the average of everything everyone has ever said online. what people say was already the worst startup signal we had. averaging it did not improve it.

Nobody can tell you whether to quit, because a flat month always has an innocent explanation and you get to pick it. A number you wrote down before you started is the only thing that takes the choice away.

The most expensive sentence in software is we were sure we knew what users needed. It never sounds expensive when someone says it. It sounds like confidence.

The senior engineer death spiral post is doing 1.7M views today, and the failure mode it describes has a founder version. You stop shipping and start having Opinions. You stop asking strangers and start asking your own judgment. Both feel like seniority. Both are avoidance with a better title.

Unpopular: shipping a prototype is now worth roughly nothing as evidence. One accelerator's intake this month: 613 applicants, 77% had already built something, 55% had not even incorporated. Building was never the proof. It was the expensive proxy for it, and it stopped being expensive.

michael moritz's stripe memo was under 4 pages. conviction compresses. if your pitch needs 40 slides to sound good, the slides aren't the problem.

miro sold for $1.36B after a $17.5B markup. the markup measured investor enthusiasm. the sale measured demand. founders run the same confusion at seed, just with smaller numbers.

harvey's new research chief on winning the app layer: it requires "working in close partnership with the professionals who live and breathe that work." the frontier stopped being model IQ. it's who actually knows their users.

a founder with $50M raised and a dead startup applies to your fund. more fundable or less fundable than a first-timer?

the consumer AI graveyard is full of do-anything bots. a travel agent you can judge in one booking. "anything" you can't judge at all. specific promises win because they're testable.

AI now writes the emails, runs the ads, edits the videos. it still can't tell you if anyone wanted the product. distribution got automated. demand didn't.

a founder doing $500k a year with no employees and a founder doing $5M a year with 40 employees and investors. who's actually further ahead?

'we love the team' from an investor is a no. 'i'd definitely use this' from a friend is also a no. both are politeness wearing a yes costume. paid is the only yes.

a big account called YC a ripoff today. 7% for $125k. the critics are pricing the check and missing the product. YC stopped selling money a long time ago, it sells proof you were picked, and a network you can't download. that's why this discourse never dies.