Summary
Cole Medin, an engineer with extensive experience building with coding agents both before and after the generative AI era, walks through 11 practical tips for making coding agents like Claude Code significantly more reliable. The video is explicitly designed to be actionable without requiring users to overhaul their existing workflows — each tip takes roughly one to two minutes to cover, and all advice applies regardless of which coding agent a developer uses.
Standouts include writing instructions for agents rather than humans — being specific about file paths, commands, and numbers rather than offering general guidance — and the concept of “rule drift,” where instruction files go stale as codebases evolve. Medin cites a study finding that one in four repositories with AI rule layers contain outdated references to deleted files or replaced databases. He recommends a “rules drift check” skill that audits a codebase for discrepancies between rules and actual structure. On the opposite side, he cautions against over-loading context: Anthropic officially recommends keeping global rule files under 200 lines, and recent research suggests too many rules now hurts agent performance more than it helps.
Other tips cover using hooks to automate agent responses to recurring events, structuring prompts to minimize the number of assumptions an agent must make, and leveraging Claude Code’s built-in features for context management. For teams actively building with Claude Code or other coding agents, this video offers a fast, dense collection of workflow improvements grounded in real practitioner experience.
📺 Source: Cole Medin · Published September 01, 2026
🏷️ Format: Tutorial Demo
