The Illusion of Maturity
There is a pervasive myth in software engineering that massive, complex test suites are a sign of a mature engineering culture. Startups love to brag about their '100% test coverage' or how they have tens of thousands of automated tests running on every pull request.
But when you look closer at those codebases, a very different reality emerges. You find simple features—like a button that opens a modal—backed by 5-page test files filled with mocked data, spy functions, and complex setup and teardown routines.
Tests are not a substitute for good architecture. If it takes five pages of code to test a simple feature, the feature is built wrong.
The Test-Code Bloat Epidemic
Writing tests used to be about ensuring quality. But somewhere along the way, it became an administrative exercise. Engineers now spend hours writing 'defensive' test code just to satisfy arbitrary coverage metrics set by management.
This creates a massive amount of bloat. When you write thousands of lines of test code to verify hundreds of lines of implementation code, you are inherently tying your company's velocity to the maintenance of that test code. Your engineers are no longer building a product for your customers; they are building a product for your CI/CD pipeline.
The Velocity Tax
This bloat functions as a permanent tax on velocity. Every time the product team wants to iterate—say, changing a multi-step checkout flow into a single-page checkout—the engineering team has to pay the tax twice.
First, they pay the tax to rewrite the actual product code. Second, they pay the tax to rewrite the 5-page test file that was hardcoded to the old multi-step flow. This is why teams that brag about their massive test suites are often the slowest to ship new features. They have trapped themselves in a web of their own assertions.
Testing the Outcome
The fastest shipping teams in the world have realized a simple truth: you don't need to test the implementation. You only need to test the outcome.
Instead of writing a 5-page script that mocks the network, spies on the API, and asserts that a specific variable was updated in state, they use autonomous visual agents to simply look at the screen. Did the checkout succeed? Did the success message appear?
By shifting the burden of verification from human-written code to AI-driven visual intent, they eliminate the velocity tax entirely. They throw away the 5-page test files. They stop maintaining mocks. They just build the product, and let the AI verify that it works.

