The “one-shot app” narrative is getting out of hand.
AI has changed how software gets built. It helps me move faster. It helps with scaffolding, refactoring, tests, debugging, research, documentation, and exploring options.
But the idea that you can describe a serious product once, send twenty agents into the night, and wake up to a business is fantasy.
Maybe that works for demos. Maybe for internal tools. Maybe for small prototypes.
But software that real users rely on is a different game.
A real product is not just the happy path. It is years of edge cases, weird customer behavior, integrations, permissions, migrations, error handling, support, security, performance, UX decisions, and all the boring little details that never fit nicely into a prompt.
If you had to fully describe even a “simple” mature app, the prompt would look less like a prompt and more like a small book.
This is where I think some of the AI hype becomes misleading.
AI can absolutely multiply a capable builder. That is already happening. A good engineer with good tools can explore more options, fix more small things, write more tests, and move through implementation faster than before.
But AI does not remove the need for taste, product judgment, domain knowledge, architecture, QA, customer feedback, and maintenance.
The bottleneck is shifting.
Less time typing code.
More time deciding what should exist, how it should behave, and whether it is actually good enough for users.
That is still hard. Maybe harder than before.
Because now we can generate bad software faster than ever.