If your agent can’t run tests locally and self-correct before you ever look at the code, you’re the bottleneck. Frontier developers invest heavily in making validation fast, local, and automatable so the agent can test in a tight loop.
The agent should have access to linters and unit tests, a browser to render and visually verify UI changes, local mocks of dependent services for integration tests, and the ability to bring up the full stack on a laptop for end-to-end testing. If you use a tool to validate your own work, the agent should be able to use it too, so it can self-validate without you copy-pasting output between tools.
Property-based tests are especially valuable here: they validate that the implementation matches your intent, and they catch edge cases the agent wouldn’t think to write explicit test cases for. The agent should be able to validate its own work, find failures, and fix them without you in the loop.
Without this investment, faster code generation just means more broken builds and more bugs. With it, overall quality in the codebase actually improves because the agent validates more thoroughly and more consistently than many humans bother to.
If your agent can’t run tests locally and self-correct before you ever look at the code, you’re the bottleneck. Frontier developers invest heavily in making validation fast, local, and automatable so the agent can test in a tight loop.
The agent should have access to linters and unit tests, a browser to render and visually verify UI changes, local mocks of dependent services for integration tests, and the ability to bring up the full stack on a laptop for end-to-end testing. If you use a tool to validate your own work, the agent should be able to use it too, so it can self-validate without you copy-pasting output between tools.
Property-based tests are especially valuable here: they validate that the implementation matches your intent, and they catch edge cases the agent wouldn’t think to write explicit test cases for. The agent should be able to validate its own work, find failures, and fix them without you in the loop.
Without this investment, faster code generation just means more broken builds and more bugs. With it, overall quality in the codebase actually improves because the agent validates more thoroughly and more consistently than many humans bother to.