Dako's AI Builder Notes
System
Color theme
Changsha HubOffline
大可 Dako
大可 Dako大可 DakoAI Builder · Agent PractitionerTurning AI into working software. Practical notes on agent engineering, vibe coding, and commercial delivery.
Platform links coming soon
Tests make AI's output verifiableNo testsevery change hand-testedcan't delegate muchTestsa machine tellsyou if it's rightWith testsrun it after changes, green issafesafe to delegate more
Tests in place, delegation can grow.

Tests First, or at Least in Parallel

Lay tests along the critical paths, and let a machine answer the question 'will this break something' instead of you guessing. Tests let you delegate more.

In vibe coding, tests aren't a "nice-to-have after it ships"; they're the pass to delegate. Wherever tests are laid down, AI can be a little freer there.

After reading, you should be able to answer:

  • Why lay tests on the critical paths first?
  • How do tests move the "human as tester" job to a machine?
  • What's the difference between tests-first and tests-in-parallel?

Tests are the pass to delegate

  • Without tests, after every AI change you verify by hand-clicking
  • With tests, AI runs them after changes; green means you can relax
  • The more test coverage, the more independent things you can let AI do

Lay the critical paths first

You don't need everything at once; lay the most valuable first:

  • Core logic: the few blocks most likely to break and with the widest impact
  • Critical flows: the paths users hit over and over
  • Regression protection: where bugs happened before, write cases so they don't return copilot-cost

With those down, 90% of "broke something" gets caught by a machine in advance.

Tests-first vs in-parallel

  • Tests-first (TDD): write the test first (define "right"), then have AI write the implementation until green. Good for modules with clear logic; gives AI a clear target
  • In-parallel: implementation and tests written together. Good for exploratory features whose shape isn't settled

Beginners start with "in-parallel"; once comfortable, use "first" for core modules. The key is tests move in with the implementation—don't add them at the very end, which usually means you never will.

Let AI write the tests

  • Tests are exactly in AI's good 70%: it writes them even more steadily than implementations
  • After a module is implemented, have AI fill in its tests
  • You don't need to fully understand the tests, but should read "what this test verifies"

Tests + verification = trust loop

  • Tests green + layered verification pass = the evidence you relax on
  • Rebuild this loop every round, and trust grows with validation quality
  • Once validation is stable, keep enlarging the delegation

Lay tests on the critical paths, and AI isn't just "can write"—it's "you dare to have it write more." Finally, it's time to ship it: ship small, pad before you jump.

References

  1. How we make AI coding more cost-efficient (GitHub Copilot)