AI says it's done but nothing works? Make it show the output
Ship ItChecked against the official source
Source checked 19 Oct 2026 · Published 23 Oct 2026 · Last reviewed 23 Oct 2026 · By Ctrl+Z Therapy (how we test)
The short answer
In a post on Anthropic's developer blog on October 15, 2026, the Claude Code team describes their first habit: give Claude something it can run, like tests, a build or a screenshot, and ask it to show the output instead of saying it's done. Boris Cherny, who created Claude Code, estimates it makes the final result two to three times better. Then a separate AI reviews the changes with fresh context.
What to do
- After each change, have your AI tool run something that proves it works: a command, the tests, or a build.
- For your app, that can be as simple as testing it in a browser or phone simulator.
- Then have a separate AI review the changes with fresh context.
- Ask your AI tool: "Don't tell me it's done. Run it, then show me the output or a screenshot that proves it works."
Source
Checked on 19 Oct 2026. Prices, limits and settings change, so check the source before you rely on them.