Glossary: 30 AI Agent Terms in Plain English
Bookmark this. Updated monthly.
| Term | Simple Definition |
|---|---|
| Agent | Software that perceives, reasons, acts toward goal |
| ReAct | Reason + Act loop - thought then action |
| Tool / Function Calling | Letting LLM call your API with JSON |
| RAG (Retrieval Augmented Generation) | Search your docs, add to prompt, then generate |
| Vector DB | Database that searches by meaning, not keyword |
| Embedding | Number version of text that captures meaning |
| Chunking | Breaking docs into small pieces for embedding |
| Memory (short/long) | What agent remembers now vs forever |
| Planning | Breaking goal into sub-tasks |
| Reflection | Agent critiques own work |
| CrewAI | Framework for multi-agent teams |
| LangChain | Library to connect LLMs + tools |
| AutoGen | Microsoft framework for conversational agents |
| OpenAI Assistants API | Managed agent API with threads/tools |
| Hallucination | When LLM makes up false info confidently |
| Guardrail | Rule to stop bad actions |
| Human-in-the-loop | Human approves before critical action |
| Prompt Injection | Malicious text that hijacks agent instructions |
| System Prompt | Hidden instructions defining agent role |
| Temperature | Randomness knob for LLM |
| Tokens | Billable pieces of text (~4 chars) |
| Context Window | How much text LLM can see at once |
| Observability | Logs, traces to debug agent |
| Eval | Test to measure if agent succeeded |
| Latency | Time from input to output |
| Cost per run | API + infra cost per task |
| Fine-tuning | Training model on your data |
| Zero-shot / Few-shot | Doing task without/with examples |
| Multi-agent | Several agents collaborating |
| Orchestration | Managing order of agents/tools |
Next: Start Here