Occasional articles on Ze: design notes, deep dives, and talk write-ups. For what shipped week by week, see the changelog.
Harnesses and models improve every few weeks, and the repositories they work in still explain nothing about themselves. Rails solved that once for people, and nothing equivalent has arrived for agents.
Read the articleAI coding needs more than a capable model. The repository must carry its architecture, relationships, tests and checks in a form the agent can discover and the project can enforce.
Read the articleZe uses AI to write code, but an RFC claim only counts after the standard, the tests, the public gap list and the commit process all agree.
Read the articleHow Ze reuses buffers, borrows wire data and limits copies so repeated routing work creates little garbage.
Read the articleHow Ze completes each peer's policy decision before encoding its UPDATE, then reuses an earlier encoding when the complete result matches.
Read the articleZe is an AI-written NOS. The useful question is whether the code is constrained, reviewed, tested and measured.
Read the article