Everyone wants better coding agents. But what about improving our codebases first? Giving better agents to a messy engineering organization will just help you create mess faster. @tereza_tizkova from Factory made this point during her talk on software factories at @aiDotEngineer. She said, "When companies adopt AI for software engineering, they can either succeed big or fail big." The difference isn't necessarily which coding agent they're using... It's whether the environment around the agent is ready for it. Think about everything we usually complain about in engineering: • Poor documentation • Flaky tests • Inconsistent coding conventions • Developer environments nobody can reproduce • Years of technical debt Humans can work around a lot of this because we've accumulated tribal knowledge. But agents don't have that luxury. If you give them a messy codebase, they can generate even messier code at ridiculous speed. This is why @tereza_tizkova says Factory evaluates something they call "agent readiness." Before trying to automate more of the software lifecycle, they look at things like: • How reproducible the development environment is • Whether the codebase is properly documented • Whether good tests and linters exist. The takeaway: The companies that get the biggest productivity gains from coding agents might not be the ones with the fanciest agents. They might simply be the ones that did the boring engineering work first. AI doesn't remove the need for engineering hygiene. But it does make the consequence of having, or not having, it much bigger.