AI agents: when to use skills vs RAG
When building AI agents, choosing between 'skills' and 'RAG' is a common source of confusion. The core distinction is simple: if your agent lacks information, add RAG; if it lacks the ability to act, add a skill. Most real-world systems end up combining both.
RAG (Retrieval-Augmented Generation) lets an AI look up information it doesn't already know — things like up-to-date product manuals, recent news, or internal company documents that weren't part of its original training. Skills, on the other hand, give an agent the ability to do specific tasks: run code, call an API, analyze a contract step by step, or follow a defined workflow. The agent knows how to act, not just what to say.
Picking the wrong tool leads to waste. If the real problem is missing information but you add a skill, the agent knows the steps but has nothing to work with. If the real problem is missing capability but you only add RAG, the agent has the facts but doesn't know what to do with them. In practice, the most efficient setup has a skill handle the overall process while RAG supplies fresh facts exactly when needed — avoiding unnecessary token usage and over-engineering.
Key points
- RAG gives the agent knowledge — it searches for up-to-date or proprietary information not in its training data
- Skills give the agent capability — the ability to execute code, call APIs, or follow multi-step workflows
- Diagnose the gap first: missing information → RAG; missing capability → skill; missing both → combine them
- The most effective pattern: a skill owns the workflow and calls RAG at the right moment to fetch current facts
- Mismatching the tool to the problem wastes tokens and adds complexity without solving anything
Quick term guide
- AI agents
- AI agents are AI tools that can carry out steps toward a goal, not just answer once.
- AI agent
- An AI program that can inspect information and suggest what to do next.
- agents
- AI helpers that follow your instructions and make changes for you.
- skills
- Extra built-in instructions that help the AI handle a specific kind of task.
- RAG (Retrieval-Augmented Generation)
- A technique where an AI searches an external knowledge base for relevant information before generating its answer
- retrieval
- The step where a system finds the most relevant text for a question.
- workflow
- A repeatable set of steps for getting a task done.
- tokens
- Tokens are small pieces of text that AI systems count when reading or writing.