Skip to main content
Agent Picks measures which tool a coding agent picks for a request, such as adding authentication, payments or email to an app. DevTune runs the request in a fresh project and checks the packages and code the agent changes to record its choice. Open Agent Picks in your project’s sidebar. Agent Picks tests Claude Code and Codex. You can use it on its own, without setting up AI Search tracking.

​
Measure

Choose a covered category or describe the job an agent would be asked to do. Select the frameworks, agents and models to test. Each combination of project environment, request, agent and model is a test cell. The starter test gives you individual picks. Paid tests repeat each cell and show install rates with likely ranges. Overview summarizes the test; Matrix breaks it down by environment and request; Runs lets you inspect individual runs. Agent Picks reports Built its own and No provider chosen separately from named tools. A recorded choice shows what the agent added or used. It does not prove the resulting integration works.

​
Fix

Losing cells create Actions in the shared Actions queue. Each Action keeps the original measurement and the evidence behind it. Reasons recorded across several losing cells can create a shared reason Action. Open an Action to review its evidence and draft a fix. Review and approve the draft before publishing it. Suggestions name locked pages and propose supplementary content around them.

​
Prove

After marking a fix live, request a paid re-test from its Action. DevTune re-tests the affected cells against their original measurements. Results shows the before-and-after install rates, likely ranges and coverage. The result records what changed during the comparison. It does not establish that your fix caused the change. Start with Set up, then use Reading results and Losses, Actions and re-tests.