Learning path · Agents & Orchestration · 60
AI Agents
LLM-driven systems that plan, use tools, and iterate toward goals—not just single-shot text completion.
Why it matters
- Unlocks workflows: research, ticket triage, data entry with supervision.
- Introduces loops where cost and failure modes multiply.
- Requires explicit stop conditions and human gates.
Key ideas
- Plan-act-observe loops
- Tool use
- State management
Agents wrap the LLM in a control loop: observe state, decide next action, call tools, update memory, repeat until done or blocked. Reliability comes from typed tool interfaces, max step budgets, and eval scenarios mirroring production messiness. Start with narrow domains—one API, one database—before open-ended "do anything" assistants. Require idempotent tools and explicit max-step limits in every agent PR; production incidents often start as unbounded retry loops. Ship only after eval gates pass on representative production failures.
Updated 2026-08-09 · Full learning path