The builder
A founder with a consumer app in mind and no engineer to build it. This is external dev dependency in its purest form: you pay someone else to build, and you own the outcome without ever owning the capability. In 2024 the someone else was a vendor valued in the billions, with an AI product manager that would, its founder said, construct most of an app on its own, faster and cheaper than any human team.
The idea was the founder's and it was fine. He described what he wanted to a chatbot, got an estimate in five figures, and paid most of it up front.
What the AI made easy
The buying. That is what the AI made easy here: a brief taken by a chatbot, a price returned in minutes, a promise of most of the work done by software. What arrived over the following months were prototypes he described as very buggy and unrelated to his specification. Two or three months in he stopped, hired his own team, and eventually built something else.
He was one of the lucky ones, because he stopped early. In mid-2025 the vendor collapsed. Reporting and an internal investigation found that the AI was mostly a front: most of the work had been done by outsourced developers in two countries, revenue had been overstated to lenders by roughly three hundred percent, a creditor seized tens of millions, insolvency followed, more than a thousand people lost their jobs, and hundreds of clients were left asking for refunds that were refused or replaced with credit. Some projects were years past their estimate when the lights went out. Third-party services appeared to help stranded customers recover what code existed.
Every one of those clients had answered to someone, a co-founder, an investor, a first customer, for a product that now did not exist. The vendor answered for nothing.
The check
Every resource runs the same three rows. Founder judgement decides the first. Engineering judgement decides the third. Observation connects them.
| Row | Verdict | Evidence |
|---|---|---|
| Originated itShould this exist, in this form? | held | The app was his idea and worth attempting. Founder judgement about what was fine. The judgement about who builds it and what I hold is the one that was skipped, and that is founder judgement too. |
| Observed itWas anyone reading the work while the AI produced it? | surrendered | He never saw the middle, and could not. The vendor's design made the work invisible on purpose: a chatbot in front, seven hundred people behind, and a client who could only wait for the next prototype. |
| Answered for itWill it hold, and who stands behind it when it ships? | surrendered | Nothing in his name. No repository, no accounts, no code delivered as it was written. When the vendor went, so did the answer to every question about the product. |
Four flaws, each one the vendor's pitch made easy to skip:
- The invisible middle. Paying for an outcome with no view of the work is a bet on the vendor's honesty. Sometimes that bet loses on a scale that makes the news.
- Owning the outcome without owning the code. No repository in the founder's name, no environment he controlled, nothing delivered as it was built. The product existed only inside a company he did not control.
- The vendor's speed as the plan. "Six times faster" became the launch date, the fundraising story, and the reason not to ask what a delay meant.
- A refund as the only remedy. When the only thing you can ask for is your money back, you have already lost the thing you were buying, which was time.
What judgement would have decided
The founder could not have known the vendor's books. He did not need to. What an experienced builder holds no matter who builds:
- FounderDecide what you hold before you decide who builds. The repository, the hosting accounts, the data, the domain, the specification: all in your name from the first day, with the vendor working inside them. A builder who refuses that is telling you something.Costs: setup you did not plan to learn, and some vendors who walk away.
- FounderBuy visibility, not promises. Running software in your environment, every week, however small. Not a demo on their screen. The moment the weekly increment stops arriving, you know, and you know months before the collapse.Costs: your own time, every week, looking at unfinished things.
- EngineeringPut a reader on your side. Someone who can open the repository monthly and tell you whether what is there matches what you are paying for. A few hours from a fractional engineer is the cheapest insurance in this story.Costs: a few hours a month, and hearing things you would rather not.
- EngineeringEvery increment shippable; no big reveal. The product exists from week two, ugly and small, and grows. A vendor that needs six months before you see anything has chosen that for their benefit, not yours.Costs: a slower-looking start, and the discipline to ask for less per week.
- FounderKnow what the agents now let you hold yourself. In 2026 a founder can produce the first prototype alone, by directing an agent, and hire only for what he cannot hold. The founder in this story had that option by the time the vendor failed. The wall that comes next is its own post.Costs: your time on origination, and the prototype-to-production wall when you get there.
Where the fast path was right
Hiring someone to build was not the error. The middle can be someone else's; that is what the middle is for, and the founder in this story would have been right to outsource it to a vendor who delivered weekly into a repository he owned. The moves above do not bring the build in-house. They bring the ends in-house.
The trade is that the founder now spends time every week watching work he did not want to think about, and holds accounts he has to secure. Attention and a little admin, in exchange for never learning from a news story that the company holding your product is gone.
On the public record, June 2025, names removed. Written with AI assistance; the judgement is mine. — Clay