Discussion
1. Give it a device
Maestro's MCP server lets the agent tap, assert and explore on a simulator.
2. Typed tools, not a debug menu
"Log in as test user" or "reset cart" get a name, a schema and a timeout. Callstack's Appduct does this without shipping a debug menu in production.
Hidden cost: seed data and auth. Agents are flaky because fixtures are unstable, not because the UI is complex.
Keep agents on exploratory work. The critical-path suite stays deterministic and human-owned.
Build the plumbing and the model barely matters.
Install Fedizen
Get the full app experience
3. Leave a file, not a transcript
Exploration ends as a Maestro flow in .maestro/. The agent writes the test.
4. CI owns it
That flow runs on every PR like any other test.