Over the past year coding agents have seen massive adoption and become flagship offerings from AI leaders like Anthropic, OpenAI, and Google. Watching these agents work can feel like magic. To fully explore their potential we need to demystify them. If you're considering contributing to an Open Source agent or even creating one of your own, this should get you off to the races!
The most common style of coding agent today is a chat loop between a user and an LLM that can perform useful software development tasks. To achieve that, there are just 3 tricks to understand beyond the LLM itself.
- How do we create an LLM chat loop?
- How can a chat loop use tools?
- What tools do we need?