TIL - Choosing Between Code, an LLM Call, and an AI Agent

A Go implementation compares plain LLM calls, chat with saved history, and tool-using AI agents for the same API orchestration task.

The piece argues that an agent is basically a loop where the model requests tools, rather than something fundamentally different from ordinary software.

It maps agent challenges to familiar distributed-systems problems like RPC dispatch, retries, idempotency, circuit breaking, and state persistence.

The takeaway: use an LLM or agent only when it adds real value; otherwise, conventional code is usually simpler and more reliable.