I handed AI agents almost the whole product. Except one part - and that part is the job

AI agents handled discovery, hypothesis generation, coding, and orchestration for a MERIDIAN prototype in just days.

The author kept final product judgment and review, preserving human control over what gets built.

A second agent audited the first one’s work and separated facts from guesses using a know / looks like / want it true framework.

The piece argues that AI is slashing software-building costs, but the real bottleneck is now taste, prioritization, and validation.