How Atested got built

Atested is built with AI. Not assisted by AI, not accelerated by AI. Built. Every file in the repository has AI involvement at some point in its history. This is the honest starting point and everything else follows from it.

AI-assisted development has real failure modes. Code that works the first time and falls apart on the second run. Code nobody can explain because nobody wrote it in the ordinary sense. Code that solves the demo and breaks the edge case. The developer community has words for this. Slop, vibe coding, the rest. They're not wrong. These failure modes are real and they matter.

The way through them is not to stop using AI. The way through them is discipline. A human decides what to build and what "built" means. The work gets structured into specific, bounded tasks with clear success criteria. An AI executes the task within those bounds. Another instance of AI reviews the result. A human verifies before anything ships. Every step leaves a record.

This is not a formality. It's the actual shape of the work. When Atested's development breaks, it breaks because one of those steps got skipped, and the failures are loud enough that we notice and the process tightens. The discipline isn't separate from the product. It's the philosophy that led to the product existing in the first place. Atested is the externalization of beliefs about verification, evidence, and responsibility that shaped how we worked before the product had a name.

The complementary model we use isn't the corporate one where AI does "creative work" and humans do "oversight." Both do both. The human provides direction, judgment, and taste. The AI provides breadth, speed, and consistent execution against clear specifications. The boundary between them moves depending on what's being built. On good days the collaboration produces work that neither could have produced alone. On bad days it produces slop, and the discipline catches it before it ships.

We can't convince the broader developer community that this works by saying it. Only time and proof will do that. What we can do is ship the product, show it working in real conditions, and let the people on the ground floor of agent development form their own judgment. They're the ones actually doing the work while the rest of the industry tut-tuts at them. Those people will know whether Atested is slop or whether the discipline holds up. They'll tell the rest of the community eventually. That's the only path we trust.