“Without structure, AI makes code worse.” - Tereza Tížková
AI is a supercharger, not a magic wand. It doesn't magically fix all your problems and make your code better. It just amplifies what's already there. If your codebase is an unstructured, sloppy, brittle mess, adding agents will just help make more of that. Additionally, and hopefully unsurprisingly, how you implement AI coding practices on your teams affects the outcome powerfully. Much like an actual supercharger, it works much better if you incorporate AI coding practices in a structured, measured, and deliberate way, versus just bolting it on and firing up the engine.
Luckily, there's a rubric for ensuring your codebase is agent-ready. Here are the pillars the Factory team enumerated as indicators of a well-structured codebase: style and validation, a build system, testing, documentation, a dev environment, observability, security, and task discovery. If you're feeling comfortable with how your codebase scores in all these areas, then it might be ready for some agents to work within it.
Luckily, this is a net win no matter what: As it turns out, all of those things on that agent-readiness list are also the things that make for a really solid, maintainable, scalable, and stable software product for humans to work on, too. Making improvements in these areas ends up being zero-cost versus if you weren't planning on adopting some agents. This has been a theme so far at the World's Fair: It seems like we're working our way out of the initial hype cycle. It's no longer good enough to burn tokens and ship the product as fast as possible. We're engineers. We're learning to use these tools like we've learned to use all of our other tools: to build good things well.
Check out Tereza's talk on the AI Engineer World's Fair 2026 livestream for much more detail and references.
Top comments (0)