Agent = LLM + Tools in a Loop

Agent = LLM + Tools in a loop. It’s not that complicated.

If you’re a software engineer and haven’t tried to build one yet, it’s a fun exercise to make your own.

Try to ask Claude Code or Codex to list their tools and document them. Then choose an open source harness (pi for example) and ask them to reimplement the spec following the sdk provided by it.

You’ll see how easy it is to get to a good quality level, and from there you can extend for the agent to match your workflow, not the other way around.