# AI - agents ## Source - https://lethain.com/what-can-agents-do/ from [[blog - Irrational Exuberance]] ## Keywords (topics and howto) - [[topic - ai]] - [[how to use ai well]] ## Related notes - this is the first zettle about AI ## Notes An agent's job is to coordinate calls to LLMs and tools. This creates a better context window. When added to the initial prompt, it improves the main model's answer. # An agent can call an LLM with a context window. If we don’t improve the context window, the agent will give the same response as a chatbot. # An agent can improve the context window by selecting and using tools. AI understands the concept of a **context window**. This refers to what context is provided with a request. This terminology is why we have shifted from prompt engineering to context engineering when calling LLMs. - A key role of an AI agent is to expand the context window of a request. This can be done by using external tools. - The agent asks an LLM which tools would best enhance the context window. - However, these tools are not called by the LLM. The calling software handles that, and only the returned values are sent to the LLM for a more complete context window. **LLMs do not call external tools; agents do.** # An agent follows a workflow (i.e. it is software) AI agents can follow a workflow of requests. > LLMs themselves absolutely cannot be trusted. Anytime you rely on an LLM to enforce something important, you will fail. [source](https://lethain.com/what-can-agents-do/) The workflow can be strict of based on stats. It is under the developer control.