One hour to a prototype, a hundred to a product (external link)
Mac Budkowski built Cryptosaurus, a small app that turns your profile picture into a dinosaur. The prototype took an hour. Shipping it took about a hundred.
The value of this writeup is entirely in the itemisation, because "the last 10% takes 90% of the time" is a cliché until someone tells you exactly which hundred hours. His: design iterations on palette and layout; over two hundred prompt iterations to make image generation survive the range of real profile pictures; AWS plumbing, S3, Lambda, domains, deployment; access control, a multisig wallet and nonce handling for the contract; rate limiting and load testing; five hundred beta users; and then a launch-day scramble when concurrent transactions broke the thing.
Not one of those is the part an agent was good at.
His framing — that AI accelerates the first 90% and leaves the rest — matches what everyone reports, but I think the split is slightly wrong, or at least badly named. It is not that the last 10% is harder. It is that the first 90% is the part with a specification, and the remaining work is where the specification gets discovered: by hitting a profile picture that breaks the prompt, by watching a stranger misuse the interface, by two transactions landing at once. You cannot delegate discovering what you are building. You can only delegate building it once you know.
Which is why "I built this in an afternoon" and "I shipped this" remain different sentences, and why the gap between them is not closing as fast as the demo videos suggest.
Two things I would hold onto. The two hundred prompt iterations is the real number in this piece — an enormous amount of the work was iterating on the instructions, not the code, and nobody's productivity metric captures that. And the launch-day concurrency bug is exactly the sort of defect that no test suite you would have thought to write catches, which is the argument for Rootly's blast-radius approach rather than for more careful review.
An honest post about a silly app is worth three serious posts about transformation.