After 1000 Hours with Claude Code: 5 Habits to Stop Doing Now
tomcrawshaw01 · x · 2026-08-05
The author shares 5 habits they stopped doing after spending 1000 hours with Claude Code. The core takeaway is that improving AI coding efficiency is less about learning new technical tricks and more about stopping unproductive behaviors.
- Stop running one session until it dies: The longer a session goes, the worse the model performs, and compacting doesn't fix it. End it early, create a handoff.md for context, and start a fresh session.
- Stop saying 'make it better': 'Better' isn't an instruction. Define exactly what 'better' means before letting Claude touch anything.
- Stop reading every line it wrote: Stop watching the AI work in real-time. Instead, evaluate the final product directly.
- Stop letting the same mistake happen twice: If you correct a behavior, immediately update the corresponding skill so it doesn't repeat the mistake a week later.
- Stop diving straight in: Don't just start issuing commands. Ensure there is an approved plan before any building begins.
More from coding & agent
- Agent Search Workflow for Obsidian: Semantic Meets Keyword — dSebastien · 2026-08-05
- NeurIPS 2026 Call for Papers: Focus on Responsible Meta-Agents — RobertTLange · 2026-08-05
- Spring Evals: Benchmarking LLMs on Real-World Spring Boot 4 Code with Hidden Tests — therealdanvega · 2026-08-05
- Navigating SAP's Agent Architecture: MCP for Tool Execution, A2A for Orchestration — SillyDot3305 · 2026-08-05
- YC's Multiplayer Agent Harness: Great Decoupled Architecture, But It Kills KV Cache — Instance_Not_Found · 2026-08-05
- The 'oh-my' Configuration Trend in Open-Source Coding Agents — HarveenChadha · 2026-08-05