Who tests my app, and why does my AI write tests?
Short answer
A separate testing agent tests each part with small test programs when it can, then you try it; tests catch faults that come back.
Before you start
Words used here (1)
- agent
- An agent is your AI following one written procedure, its standing instructions, for one kind of work.
Deeper
Also asked:
Who tests my app: me, my AI or someone else? All three: Agent.Tester, a separate testing agent, tests each part; you try it on your computer and check it once it is live; and other people who try it once it is live test it too.
Three take part: an agent that tests, the agent that builds, and you. An agent is your AI following one written procedure, its standing instructions, for one kind of work.
Agent.Tester, the agent that tests what others built, tests each part. Agent.Coder, the agent that writes your app's code, starts it as a separate agent and gives it only the requirements written in your plan, where the code is, and which parts were built. It knows nothing of how they were built, so it forms its own view, and its job is to find problems, not to confirm that all is fine. It runs the tests that are there, writes tests for what has none, runs them all again, and saves a report with its verdict: PASS (all checks passed), PARTIAL (only some of the checks could be run) or FAIL. A verdict that is only in a reply does not count. If Agent.Coder cannot start a separate tester, it tests the part itself and marks the result self-QA, which may stand in for a PASS from Agent.Tester but is not equal to one. Agent.Scaling, the agent that checks what was built, checks it too. You cannot choose Agent.Tester or Agent.Scaling yourself.
You test it as well. When the work is built and tested, your AI starts it on your computer and tells you what to try. Once it is live, Kai-Zen, the planning agent, asks you to check it live, and fixes what you send back until you approve. Other people who try your app once it is live test it too; the answer How do I get feedback from the first people who try my app? shows how.
What a test is. A test is a small program your AI writes beside your app's code. It uses one part of the app, such as a sum or a button, and checks that the result is the one expected. Some tests check one small piece at a time; others open the whole app in a browser and use it the way a person would. Tests also try what goes wrong: a page that is not found, access that is refused, a lost connection, or wrong input. Each test that checks something your plan asks for carries a note naming that part of the plan.
Why your AI writes them. A test can be run again in moments. Agent.Tester runs the tests that are already there again for each new part, so a fault that comes back is found before you see it; the answer Does my AI check that a new part did not break the old parts? tells more. Agent.Tester never gives a PASS while a test of the part fails, or while something the part must do has no test. For a change to a screen, Agent.Coder, if it has a browser tool, also opens the screen in a real browser at phone size, takes a screenshot and counts the errors the browser reports; if it has none, it says so. A passing test run is not that proof.
Tests prove what they test, on the computer they ran on. A pass on the builder's own computer does not count as working in production, the live system people use. The answer How do I know my app works? tells more.
You do not need to read or write tests yourself. They are kept with your app's code.
This answer is written for a computer screen.
Source: Standards read from GilbPlatform on 13 November 2026: Agent.Tester (Role: You did not build this code ... Your only job is to find problems.; Task: saved report, A verdict that is only in your reply does not count.), Rule_Code_qa-testing (RuleQatestRole, RuleQatestIsolation: spec, package path, increment scope only; RuleQatestVerdictRules: Always write tests for gaps, "Never give a PASS verdict if any current-increment test is failing or any in-scope spec criterion has no test"), RuleQatestPartialIsLawful, Proc_v_i_c_QATesting (Run, Find, Write, Tests), Rule_Code_testing (RuleTestTools: unit tests and end-to-end browser tests; RuleTestWhatGetsTested: error states 404, 403, network failures, bad input; RuleTestFileLocation: tests live alongside the code; RuleTestSpecLink: spec reference comment; RuleTestE2e: from the user's perspective), Agent.Coder (AfterIncrement: self-QA, browser proof; Solutions), Kai-Zen.Cycle (Check, Human), Kai-Zen.HumanCheck. Live help pages, sentences shared: How do I build an app with an AI coder? (You cannot choose Agent.Tester or Agent.Scaling yourself); What counts as proof that something works?; Can I trust what the AI suggests and builds?
Checked against GilbApp d529620.