Stage 7: Tools, agents and memory
Letting models act safely, with bounded cost and a clear stop point.
Summary, rules and context: AI Engineering Guide, Stage 7.
Learn
Curriculum lessons, in reading order.
Apply
Field-manual scenarios: implementation steps, an example, trade-offs, production notes, a checklist and metrics.
1-agents/
- Agents → Sandboxing (code execution and shell tools)
- Agents → Multi-agent communication (MCP, A2A, shared blackboard)
- Agents → Cost & step budgets
- Agents → Single agent loop
- Agents → Task decomposition ("define down")
- Agents → Orchestrator / subagent
- Agents → Fan-out / fan-in
- Agents → Loop-back / reflection (critic, verifier, replanning)
- Agents → Planning styles (ReAct, plan-then-execute, search)
- Agents → Tool design
- Agents → Human-in-the-loop (approval gates and resume)
- Agents → State & durability (checkpointing and idempotency)
2-memory/
- Memory → Types (short-term, working, episodic, semantic, procedural)
- Memory → What to write (the extraction step)
- Memory → When to retrieve (relevance + recency, with a cap)
- Memory → How to forget (TTL, decay, deletion, user controls)
- Memory → Conflict resolution (supersede, don't accumulate)
- Memory → Security (memory as an injection surface)
Previous: Stage 6: Retrieval-augmented generation · Next: Stage 8: Choosing the lever