Level 5: AI / LLM systems · L5.5 Agents, tool use & MCP
The agent loop (ReAct)
Trace reason, act, observe and its cost/error propagation.
12 min · Pro
What you take away
- The loop is three moves — think, act on one tool, observe the result — repeated until the model judges the task done or a limit you set stops it.
And 3 more inside the class, in Pro.
How the class is laid out
- The problem
- Three moves, over and over
- The first cost: errors compound
- The second cost: the transcript is re-sent every step
- Check your understanding