EN
Foundations
Tool calling, planning vs reactive, and the limits youβll hit in prod.
- What an AI Agent Is Made Ofβ ββWhat inside allows it to do this?
- How an Agent Uses Tools (Basics)β ββIt works with text only.
- How to Restrict Tool Accessβ β βTool calls are where agents break production: schema drift, retries, side effects, and that same 'oops' admin token. Here's how to survive.
- How an agent decides what to do next (Planning vs Reactive)β β βIt is not given an instruction. It is not told the action order.
- Why LLM Agents Can Fail: Technical Limits Explainedβ β βUnderstand why LLM agents fail: hallucinations, context limits, tool errors, and the engineering guardrails that make agent behavior reliable.
- Agent Memory: What It Remembers and Whyβ β βWithout this, every new action would be like a first attempt. Sometimes endlessly.
- What an agent is allowed to do (and what it is not)β β βNot all actions are safe
- When an agent should stop (and who decides)β β βBecause its job is to complete work. Not decide when enough is enough.
- When an agent is ready for real workβ β βYour agent works on a laptop. In production it can loop, spam tools, leak data, and burn money (depending on the use case). This is the checklist we wish we had before release.