kodwai Reviews

kodwai Description

Anyone can get an AI agent to turn tests green on a toy problem. kodwai measures the part that is harder to fake: how well you direct it. Developers pick a real engineering challenge (a CRDT text buffer, an event-sourced ledger, a zero-downtime migration), pull it down with npx @kodwai/cli@latest challenge <slug>, and solve it locally with Claude Code, Cursor or Codex. No browser sandbox. When they run npx @kodwai/cli@latest submit, kodwai reads the session (prompts, commits, test runs) and returns a 0 to 100 score across three axes: Direction (how you steer, verify and decompose), Outcome (what shipped, replayed and stress-tested) and Lift (how far you beat a solo AI). Each signal cites its evidence, and the score comes with a confidence interval instead of false precision. Developers get 3 free scored submissions, then connect their own Anthropic key for unlimited runs. Teams can use the hiring track to run private real-world interviews.