AI coding agents highlight the flaws in development environments by exposing their brittleness and inconsistencies. While many teams believe they have solid practices in place, the reality is that much of their workflow operates on informal traditions. For instance, developers often rely on ad-hoc solutions like deleting `node_modules` when builds fail or asking colleagues for environment variables. These gaps become glaringly apparent when agents enter the scene, as they lack the human intuition needed to navigate ambiguity. Instead of simply generating code, agents reveal the weaknesses in existing systems, turning previously overlooked issues into significant pain points. The article emphasizes that many engineering practices are aspirational rather than robust. Teams may have testing and CI processes, but these often lack the reliability needed for agents to function effectively. Flaky tests, inconsistent deployment procedures, and poorly enforced standards create a chaotic environment. This chaos amplifies when agents, which operate without the benefit of human judgment, encounter these shortcomings. They can either get stuck or produce incorrect outputs, leading to costly inefficiencies. To address these challenges, the author suggests teams adopt a gardening mindset toward their development ecosystems. This involves standardizing processes, making success and failure signals clear, and minimizing hidden complexities. Key recommendations include creating a single command to run and test the code, enforcing version control, and documenting workflows in a way that is machine-readable. By building a more structured and reliable environment, teams can unlock the true potential of AI agents, transforming them from simple tools into powerful allies in development.