AI agents
An agent is a language model wrapped in a loop that can call tools. This course teaches you to build one from scratch in Python — a research agent that reads the web and an internal database, cross-checks facts, and returns a sourced note — then to reimplement it in LangGraph for comparison.
Course Duration: 6h
What You'll Learn
- Tell an agent apart from a chain and from a single model call
- Implement the ReAct loop in about sixty lines of Python
- Write tool descriptions the model actually uses correctly
- Manage working memory and long-term memory without saturating context
- Add planning, self-critique and guardrails without inflating cost
- Diagnose the typical failures: infinite loops, drift, prompt injection
- Log every decision to make a run reproducible and auditable
Prerequisites
- Large language models (course 16) and prompt engineering (course 17)
- LangChain basics (course 26) helpful but not required
- Comfortable with Python, JSON, and reading HTTP responses
Course Modules
- What an agent adds, and what it costs
- The reasoning and acting loop
- Function calling and tool descriptions
- Working memory and long-term memory
- Planning and task decomposition
- Self-critique and result verification
- Guardrails: token budget, permissions, checkpoints
- Typical failures: loops, drift, poorly described tools
- Observability and decision logging
- Project: a supervised research agent
Assessment and certificate
The course ends with a 40-question exam covering every module: when an agent is the wrong answer, how to write a tool description that gets called correctly, when to replan, how to spot prompt injection through a fetched page, and how to read a decision log. On success, a verifiable certificate of completion is issued immediately.
Free courses, by contrast, end with a 5-question quiz and a preview of the certificate, without certification.