@rollins.io

What they said

1. Definition: software engineering is releasing production-quality code to the real world reliably and safely, with the ability to recover from disaster.
2. Start with tests: every good professional team has a CI/CD pipeline running in the background.
3. Unit tests: in a vibe-coding world every line should be covered by a unit test. It is easy, free, and a failed unit test costs almost nothing.
4. Not enough: cover every functional piece with multiple kinds of tests at vibe-coding speed.
5. Unit test = one function, input to output. Integration tests chain multiple units. End-to-end tests give one input at one end of the whole chain (front end, API, business logic, database) and check the output is right.
6. Most teams never built E2E tests because they were too hard; with AI it is easy: the agent clicks buttons and tests every path and every part of the UI, especially for risky features.
7. Performance tests: nothing is worse than clicking and waiting 5 seconds; set rigorous, aggressive performance tests (frame shows a checkout going 3.9 s then 5.0 s).
8. Conclusion: focus first on CI/CD; once right you move faster and with more confidence. Creator replies in comments add: they run 25,000 to 50,000 tests, GitHub runners cost over $1,000/month across three projects and they may leave them; they test DB changes by copying production DB into a new environment on PRs; "we don't read or write code anymore"; "last 20% takes 80% of the time".

Try this

• Anyone building with AI code: Put tests and a CI/CD pipeline around AI-written code before you scale.
• Unit tests: Cover each function. They are cheap and a failure costs almost nothing.
• Integration and end-to-end tests: Chain pieces together and test the whole path from the front end to the database. An AI agent can click through the UI and test each path.
• Performance tests: Set a limit, such as no page waiting more than a few seconds, and fail the build if it is exceeded.
• Releases and review: Make sure you can roll back, and test database changes on a copy of production data. Coverage does not prove correctness, so review risky features by hand.
• Count the cost: Tests and CI runners are not free. The creator's own reply cited over $1,000 a month in runners.

Links

• CI/CD pipeline (GitHub Actions/runners mentioned in replies), Playwright (commenter suggestion; creator asked if people use headless Playwright or agents with browsers).
• Claude Code/agents as the coding agent (commenters).
• No links in caption (empty).