Working with Claude Code: A Field Manual (blog.iannelson.uk)

A recent post outlines practical strategies for effectively utilizing Claude Code, emphasizing the shift from theoretical understanding to actionable workflows in agentic coding. The author shares key lessons learned from hands-on experience, particularly the importance of maintaining a small and disciplined context window, as well as the practice of clearing context frequently to enhance output quality. By delegating research tasks to subagents, users can focus on critical decision-making without the distraction of historical context clutter. Significantly, the author advocates for treating planning as an essential artifact, implementing structured GitHub issues as the starting point for discussions with Claude. This approach creates a clear distinction between planning, implementation, and review processes, allowing for better management of different phases of development. The suggested practices, such as using various models for each phase and employing version-controlled configuration files, aim to enhance coding efficiency and quality. Essentially, the post serves as a comprehensive manual for the AI/ML community, highlighting actionable techniques to elevate the coding process through effective use of AI agents.