Does my AI check that a new part did not break the old parts?
Short answer
Yes, where tests exist: all the old tests run again with the new part's, and failures are reported; the testing agent adds tests for gaps.
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
Checking that the old parts still work after a change is often called regression testing. It is part of how your app is built and tested.
The tests stay. Every test your AI writes, a small program that checks one part of your app, is kept with the app's code. Agent.Tester, the agent that tests what others built, starts only when all the tests that are already there pass. It runs them again for the new part, writes tests for anything the new part must do that has none, and runs everything once more. It never gives a PASS while a test of the part fails. Tests that open the whole app in a browser, if your app has them, count the same: a failure among them is a FAIL. It also runs the whole set of tests on the changed code and on the main line of the code, one file at a time, and compares the results. If Agent.Coder, the agent that writes your app's code, cannot start a separate tester, it runs the test suite itself and marks the result self-QA, which may stand in for a PASS from Agent.Tester but is not equal to one. An agent is your AI following one written procedure, its standing instructions, for one kind of work.
Its job is to find problems, so it also tries what the builder may not have thought of, such as wrong input or a lost connection.
Parts kept apart. Agent.Scaling, the agent that checks what was built, checks that each part has one job and deals with other parts only through their agreed interface, the fixed way parts talk to each other, and that nothing the app already has was built again. That keeps a change in one part from reaching into another.
Goals measured again. When a change alters what an earlier part rests on, Agent.Coder, the agent that writes your app's code, measures it again, or lowers its Status, where it stands now, and says why. In each round Kai-Zen, the planning agent, has each goal the step aims at measured, a goal here being a goal for how well your app must do something, and its report names each goal that was not measured again.
What tests cannot do. Tests check only what someone thought to test, so an old part that has no test, or a fault in a gap between tests, can still slip through. That is why you also try the app yourself and check it again once it is live. If something that worked before is broken, tell your AI; the answer How do I tell my AI about a fault so it can fix it? shows how.
This answer is written for a computer screen.
Source: Standards read from GilbPlatform on 13 November 2026: Proc_v_i_c_QATesting (ProcQatEnExisting: All existing tests are passing; Run, Find, Write, Tests), Rule_Code_qa-testing (RuleQatestAdversarial; RuleQatestVerdictRules: Never give a PASS verdict if any current-increment test is failing, E2E test failures count toward PASS/FAIL the same as unit test failures; RuleQatestWholeSuite: "Run the whole suite one file per process, on the branch head and on main, and compare per file."), Rule_Code_testing (RuleTestFileLocation, RuleTestMustPass, RuleTestProbeArtefactNotDiff), Agent.Coder (Solutions: "When an increment changes what a built spec rests on, measure it again or lower its Status, and give the reason"), Kai-Zen.Cycle (Measure; Update: A Value not measured this run keeps its Status, and the report names it.), Agent.Scaling as cited by the live page How do I build an app with an AI coder? (one job, agreed interface, nothing built again).
Checked against GilbApp d529620.