# Tests Come for Free Now, and Most of Them Are Pointless > Source: > Published: 2026-10-06 10:09:18+00:00 ### How to Make Videos with Claude Code Claude Code can't generate video directly. It isn't a video model like Veo or Sora. It writes text and code. You can… For years everybody agreed that tests make software better, and almost nobody wrote them. In the first ten years or so of my career I probably didn't write a single test. I always felt a bit guilty about it. Now AI coding agents write tests for everything they touch, so tests come for free. And now that I finally have all of them, I can see that most are pointless. They need constant updates, and they slow everything down. I was never alone with this. For the last few years I have worked as a freelancer on client projects. Almost none of my clients had a good test suite. Most had no tests at all. The ones that did usually had a big CI setup with GitHub Actions that was only half done. Developers skipped the tests to get any work done, and the tests ran on one person's machine and failed on the next. On one project the suite ran in a few minutes on my client's MacBook. On my brand new, very expensive ThinkPad it took about two hours. With AI these days everything is a lot easier. The AI fixes any broken test setup, and it adds tests to everything without even being asked. Codex with the newer GPT models is the most eager, and every feature comes with a pile of new test files. One of my own Laravel codebases ended up with about 6,000 tests in almost 1,300 files. A full run took more than five hours. AI agents love to create tests that e.g. open a Vue file and checked that a certain line of code is still in it. Or compare the translation files key by key. Or running database migrations backwards and forwards. These things are almost never worth testing. The parts that do break are hard to test anyway. Calls to an outside API depend on a service you don't control, and a scraper depends on websites you haven't seen yet. So the tests pile up around the easy code, while the risky code is as uncertain as before. Every test is also code someone has to keep up to date. You change a feature, forget the matching test, the suite fails, and you go and update the test. It's a chore, and only worth it for things that really make sense to test. A green suite also doesn't mean the app works. A coding agent finishes the task I gave it, reports that all tests pass, and I open the app and find problems immediately. What works much better for me is letting the AI use the app. I give it tools to open the app, log in and click through a feature the way a user would. It writes small scripts as it goes and changes them when something looks off. These scripts don't run automatically, and I keep them outside the codebase as they are not meant to be a "suite" that you "run", but rather a set of tools that the AI agent can adapt as it goes. So in my Laravel codebase I just deleted almost all the tests. The app now has 63 left, down from thousands. They cover the things that would really hurt if they broke, like what team sees which data, credits and payments stuff, sending email and deleting data. The whole suite runs in about 30 seconds. All my agents now follow a new rule that says the full suite must finish in under a minute, and a new test is only allowed for good reasons. My agents used to spend hours waiting on thousands of tests. Now they spend that time clicking through the app, which is where they find the actual bugs. Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request. Take a look at vroni.com