Can AI really build production software?
Yes: two of my tools are live on the internet with public source, and a full platform build with 52 automated checks sits in a public repository, all AI-built under direction. What AI cannot do is know what your business needs, notice its own confident mistakes, or prove its work: that is the human's job.
The longer answer
In practice.
What changes with AI
Volume and speed: the typing is no longer scarce. Internal tools that were a
quarter's project are now a matter of weeks, which moves whole
categories of software from "not worth it" to "obviously worth it".
What does not change
Something has to define right, check the result against reality, and take
responsibility. AI will build the wrong thing beautifully, describe tests that do
not exist, and say "done" as a vibe rather than a fact. I run it like a delivery
team: written specifications, adversarial review, proof against sources it does
not control, per AI development.
How to check any claim like mine
Ask for something inspectable without trust: live software, public source, a
build you can run. Mine is at
the reference build;
anyone who cannot show equivalent is asking you to buy on charm.
Same rabbit hole