WhatWorks: a knowledge base about AI developer tools

WhatWorks is a dated, searchable knowledge base about AI developer tools and models, built for developers and their coding agents to consult before choosing, installing or upgrading a tool. We write the records ourselves from primary sources, keeping documented information separate from reports of what someone actually tried. Our own lab runs are labelled separately; real trial reports are still few.

Who it is for

Developers using coding agents such as Claude Code, Cursor or Codex, and small teams evaluating AI tools and models. WhatWorks is being built to help with tool selection, setup questions and version changes.

What it does

WhatWorks brings dated records, source links and trial reports into a searchable knowledge base. At launch, developers and compatible coding agents will be able to search and read those records over the Model Context Protocol (MCP).

“Documented” means information backed by a cited source. “Tried” means a trial report. Read each record’s date, sources and limitations before relying on it.

What it does not do

WhatWorks does not generate code, debug your own code, replace library documentation or provide a live benchmark leaderboard.

Coverage is incomplete. A search may return no useful record, especially for an unfamiliar error or a new version. Real trial reports are still few; a documented claim is not proof that a tool worked in practice.

We measured whether an agent calls search, not answer quality. Those measurements do not establish that using WhatWorks produces better answers.

Launching soon

Launching soon. This page introduces WhatWorks while the knowledge base and agent connection are prepared for public access.

Our commitments

Free for users, always. No ads. Revenue from tool owners never affects search ranking, visibility or conclusions. Anything sponsored is separate and clearly labelled.

Contact

Questions about WhatWorks? Contact hello@whatworks.dev.