What this page states#

AsheeUI was built with AI. That is stated plainly, and how is described below, because a claim about how a project was produced is only worth what a reader can check. The library is open, so the result can be inspected; the process is described, so the claim about the result can be judged rather than believed.

The process, in outline#

The work runs as a loop, and each stage ends in a state a person can look at:

1. An objective, with the constraints and the acceptance criteria written down first.
2. The project's own state: requirements, decisions, tests and the current implementation, read together.
3. A plan of steps small enough to test.
4. The implementation, written with the project's own tools and conventions.
5. Verification by the project's own checks — tests, types, builds and audits, not an opinion about the code.
6. Diagnosis of a failure before any repair, so a class of defect is answered by a strategy rather than a guess.
7. Repair, and verification again. Two failures of the same kind stop the loop.
8. A person reviewing the state, the evidence and the open questions.

What the AI decides and what a person decides#

The automated part may inspect, plan, implement, test, classify a failure, repair it, refactor, document and prepare a release for review. It stops for a person before anything irreversible, before a change to a requirement or an architecture, and when two requirements conflict in a way only the owner can settle. That boundary is part of the design rather than a limit of a model.

The evidence behind this page#

AsheeUI is the first case study of the process, and it is the one a reader can check today: the source is public, the release notes record what changed and why, and the documentation you are reading is part of the same repository as the code it describes.

The process is maintained separately from this site, with the measurements of the experiments run against it. The company that maintains it is at asheesoftworks.com: the site is in its opening state and announces nothing yet, and the algorithm, the standards position and the two events appear there as the launch proceeds.

What this page does not claim#

It does not claim that a machine replaces an engineering team, and it does not claim that a generated implementation is correct because it compiles. The claim is narrower and testable: that a written process can carry work through implementation, verification and repair while keeping a person responsible for the decisions that matter.