I really like Simon Willison’s new series on agentic engineering patterns. If you’re using coding agents day-to-day, this feels like required reading:

One post in particular stood out to me: First Run the Tests.

Automated tests are no longer optional when working with coding agents.

This feels exactly right, and it’s fortunately very easy to implement.

Any time I start a new session with an agent against an existing project, I now begin with:

First run the tests.

He describes several benefits of this approach: it tells the agent a test suite exists and forces it to learn how to run it, reveals project scale and complexity, and puts the agent into a testing mindset, which makes it more likely to add or update tests during implementation.

Small prompt, big leverage.

<
Previous Post
Something Big is Happening
>
Next Post
AI Reading List: April 2026