The Agentic Development Lifecycle - We wrote 29 rules for every check we built

Chapter 7 - Guides assert, sensors observe, and we bought the wrong one We wrote 29 rules for every check we built. That ratio says something about how we build AI agents. Birgitta Böckeler gives us a useful equation: Agent = Model + Harness You may not control the model. But you do control the harness around it. And that’s where things get interesting. There are two ways to influence an AI coding agent: 🧭 Guides tell the agent what to do before it acts. Examples: AGENTS.md, CLAUDE.md, coding conventions, and instructions. 📡 Sensors check what happened after it acts. Examples: linters, type checks, tests, and automated reviews. Both matter. But they don’t offer the same kind of protection. A guide says, “Please follow this rule.” A blocking sensor says, “This change cannot proceed until the rule is satisfied.” We examined 2,853 GitHub repositories using agentic coding tools. Context files appeared in 90.6% of repositories, while hooks appeared in just 0.9–3.2%, depending on the tool. In Claude Code repositories, the ratio was roughly 29 context-file repositories for every repository with hooks. Why the gap? Writing a rule in a Markdown file takes minutes. Building a check that runs automatically, blocks a bad edit, and has someone responsible for maintaining it takes much more effort. This isn’t simply a matter of developer discipline. It’s a cost gradient. But it points to an important engineering opportunity: move more checks into the agent’s edit loop. Don’t just tell the agent what good code looks like. Give it fast, deterministic feedback when the code falls short. Start small: • Run a syntax check. • Add a type check. • Run fast unit tests. • Block the edit when a critical check fails. One useful blocking check can be more valuable than another page of instructions. Try this: Pick one recurring mistake in your AI-assisted workflow. Can you turn it into a check that blocks the bad edit before it lands? 📖 This is the idea behind Chapter 7 of The Agentic Development Lifecycle: Guides and Sensors. Read the chapter: https://neam-lang.github.io/The-Agentic-Development-Lifecycle/chapters/07-harness-engineering-guides-and-sensors.html Question for developers: In your AI coding setup, what’s the ratio of written rules to automated checks? AIEngineering #AgenticAI #SoftwareEngineering #CodingAgents #DeveloperTools