Last week we ended by recommending you have AI systems write the code once and keep it for any task you repeat and where accuracy is critical (as opposed to re-prompting a chatbot every time). But, if you're not a software developer, that raises the question of how exactly to do that. You're not going to hand-manage Python scripts out of a browser chat window. The tools built for this job are called agentic coding tools, and the two you likely hear about most are Claude Code and OpenAI's Codex. They are a different animal from the chat box you already know. This issue is about what agentic coding tools are, how they differ from browser-based AI systems, and when you would reach for each.
Welcome Back
The last three issues each described one part of a larger ecosystem. An agent is a loop pointed at a goal. “Reasoning” is that loop made to plan before it acts. A tool call is the moment the model’s text reaches out and runs a piece of deterministic software. Agentic coding tools are what you get when you assemble all three and point the resulting system at your files. They are the agent, doing the “planning”, and making the tool calls but in your real working environment instead of a sealed conversation in a browser. Knowing what that changes is what this week is about.
