reading · 8 min
What Is an Agent?
Objective: understand what this track teaches, how the in-browser exercises work, and the one definition the rest of the curriculum builds on.
What this track covers
A job-ready path for software engineers becoming AI engineers who ship agentic systems. It is deliberately framework-agnostic: you learn the mechanics that stay true across SDKs and model versions. The full map:
- Foundations — the agent loop and the mock-LLM harness (you are here)
- Prompting & Control
- Tool Use
- Retrieval & RAG
- Memory & State
- Planning & Reasoning
- Multi-Agent Systems
- Evaluation & Testing
- Guardrails & Safety
- Cost, Latency & Reliability
- Observability & Debugging
- Production & Deployment
How the hands-on works
Interactive lessons run real Python in your browser — no install, no keys, no cost — so you can focus on mechanics. Each agent’s policy is a deterministic mock LLM you control: the same interface as a real model, fully reproducible. Calling real model APIs and frameworks is a production concern covered later; the patterns you learn here are identical either way.
What an agent is
An agent is a system that repeatedly perceives its environment, decides what to do, and acts — then observes the result and loops again. The decision step is where a model (here, a mock) acts as the policy.
Internalize this distinction: a plain LLM call is a single function
prompt -> text. An agent is a loop that calls a policy repeatedly,
feeds results back in, and stops on a goal or a budget. Tools, memory,
planning, multi-agent — every later topic is a refinement of that loop.
Best practice: treat the policy as a swappable dependency from day one. If your loop only knows “call
policy(observation)”, you can test it with a mock today and drop in a real model later with no rewrite.
Next: The Agent Loop — build this loop in Python and run it.