10 min
Prompts, tools, and agents
How to get useful work from an assistant, and when you have actually built a workflow rather than a chat.
A prompt is a spec
Write the outcome, the constraints, and what “done” looks like. “Make this better” is not a spec. “Rewrite this email for a client who already knows the project, 120 words, no new promises” is.
Put source material in the prompt or attach it. Do not ask the model to remember a file you never gave it.
Tools are how chat becomes action
A tool is a function the product is allowed to call: search, a calendar, your repo, a database. MCP (Model Context Protocol) is one open way to attach those tools. The assistant is only as permitted as the tools you connected.
Agents are loops
An agent is a loop: plan, call a tool, read the result, continue. Loops fail in boring ways — infinite retries, the wrong file, a spend limit. Start with one tool and one success check you can see, not a swarm.
Related AI sites in the catalog
- Cursor — An AI-centered code editor. You stay in a project; the assistant can read and edit files with your permission.
- n8n — A workflow tool for connecting apps and, where configured, calling AI APIs as steps in an automation.
- Model Context Protocol — An open protocol for connecting assistants to tools and data sources. MCP is a spec, not a single product.