I've built AI agents at work, but always by piecing things together — never with a formal grasp of the underlying concepts. I want to fix that: go deep, be systematic, and actually understand what I've been building. This series tracks that daily progress.
Today
Started Hugging Face's Agents Course, Unit 1: What are agents?
Core idea: an agent is a system where an LLM controls the workflow — deciding what actions to take, in what order, using tools it has access to — rather than following a fixed, hardcoded sequence. The general loop is:
- Think — reason about the current state and what to do next
- Act — call a tool (search, code execution, an API, etc.)
- Observe — take in the result and decide the next step
That loop repeats until the task is done. The spectrum of "agentic-ness" ranges from a plain LLM call with no control over flow, up to a fully autonomous agent that plans, acts, and self-corrects with minimal human input.
Simple framing, but a useful one — it gives me a vocabulary for things I was doing intuitively before.