Why testing got so much more important with the rise of AI agents

Today we're talking about the thing everyone loves to skip and nobody regrets writing: tests. Except this time there's a twist. Because the moment AI agents started writing our code for us, testing quietly went from "nice to have" to "the thing holding the whole operation together."

I'll make a bold claim and then back it up: tests matter more now that AI writes the code, not less. By the end of this post you'll know why, you'll know the difference between functional and unit tests, you'll hear how a 60-hour client estimate turned into a one-day build, and you'll walk away with the exact set of rules I hand my AI agent so it never "forgets" to test again.

All examples are in Pest, because it's the default in modern Laravel and life's too short for boilerplate. Let's go.

So why did testing suddenly get more important?

Here's the uncomfortable truth about AI agents: they're fast, they're confident, and they have absolutely no idea whether the code they just wrote actually works. It looks right. It's plausible. But plausible and correct are two very different things, and an agent can't tell them apart. And here's the kicker: no matter how sharp your input is, no matter how well you've tuned your agents, every now and then a case just slips through. Something gets forgotten. It happens to the best prompts and the cleanest setups, which is exactly why you can't lean on the agent alone.

When you write code by hand, you're the first reviewer, you read every line as you type it. And let me be clear: you should always review what your agent produces, no exceptions. Reading the output and understanding it is non-negotiable. But here's the thing, an agent fires off 200 lines across five files in ten seconds, and reviewing that volume at that speed, every single time, is genuinely hard. Even a sharp reviewer misses something when the code keeps coming. So review stays mandatory, but it needs a backup that's automated and objective, something that catches what slips past your eyes. That backup is your test suite.

And here's the part that really flips the switch: tests are how the agent fixes its own mistakes. Give an agent the ability to run your tests and the whole game changes. It writes code, runs the suite, sees what's red, and patches it, all on its own, before any of it lands on your desk. No tests? The agent is coding blindfolded, and so are you. Good tests? The agent has a target to hit and an alarm when it misses. The better your tests, the better it self-corrects.

Think about where the bottleneck went. Writing code used to be the slow, expensive part, so that's what we optimized for years. Agents made writing code almost free. The expensive part now is being confident the code is actually correct, and tests are how you automate that confidence. In an AI workflow your tests aren't overhead bolted onto the real work. They are the real work. They're your leverage.