How coding agents use tools, memory, and repo context to make LLMs work better in practice

Sebastian Raschka's Blog offers insights, tutorials, and research updates on machine learning, deep learning, and artificial intelligence. Covering topics such as neural networks, data science, and Python programming, Sebastian Raschka's Blog provides resources for students, researchers, and practitioners in the field of AI. Developers can learn about  algorithms, research methodologies, and practical applications of machine learning through Raschka's blog posts and publications.

Sebastian Raschka

A deep dive into the six core components that make coding agents like Claude Code and Codex CLI work: live repo context, prompt shape and cache reuse, structured tool access, context bloat minimization, structured session memory, and delegation with bounded subagents. The post explains how the surrounding harness (not just the model itself) is what makes LLM-based coding tools feel significantly more capable than plain chat interfaces. A minimal Python implementation called Mini Coding Agent is provided as a reference.

Components of A Coding Agent

Claude Code, Codex CLI, and Other Coding Agents

On The Relationship Between LLMs, Reasoning Models, and Agents