Testing with Agents in a Minute · intermediate

Testing in a Minute: Test-first agent workflows

Test-first agent workflows make the agent write or confirm a failing check before implementing—so done means the check flipped green.

Last verified: 2026-09-06

In one sentence

A test-first agent workflow requires a failing, behavior-focused check before implementation begins.

Why it matters

If the agent implements first, you do not know whether new tests encode real requirements or merely mirror the code. Failing first anchors the contract.

How it works

  1. specify the behavior;
  2. add or identify a test that fails for the right reason;
  3. implement until that test passes;
  4. run the broader gate;
  5. review both product and test diffs.

Existing regression tests can serve as the “first” failing check for bugfixes. For new features, have the agent draft the test under tight review.

Example

Write a failing test for empty email -> 400.
Show me the failure output.
Only then implement the fix.
Do not change unrelated assertions.

Agentic coding use

Test-first is one of the highest-leverage patterns for trustworthy agent delivery. It converts vibes into a boolean.

Watch out

Agents may write tautological tests (expect(true)). Review assertions for user-visible behavior, not implementation trivia.

Related lessons