The GenAI Field Guide

Agents that hold up in production

Loops, tools, state, budgets and verification: the parts that decide whether an agent survives real traffic.

1. What is an agent loop?
2. Agent or deterministic workflow?
3. What is a tool schema?
4. What is idempotency?
5. What is harness engineering?
6. How do you set execution budgets?
7. How do you prevent infinite loops?
8. What is an agent state machine?
9. How do agents resume work?
10. How should tool results enter context?
11. Why do agent error rates compound?
12. What is indirect prompt injection?
13. What is tool result poisoning?
14. What is an action guardrail?
15. What is trajectory evaluation?
16. How do you trace an agent?
17. How do you know a task is done?