What an AI agent can do: a meeting example
Understand how an AI agent chooses steps and uses tools, using a simple meeting-preparation example.
Alongside People · 2 min read ·
An AI agent is software that uses an AI model to choose some of its next steps and use tools to carry them out.
A meeting example makes that easier to picture. Imagine asking software to find recent customer notes and prepare a short brief. If one note refers to an earlier conversation, it chooses to look there too. This example is hypothetical.
What makes it an agent?
A fixed automation follows rules you set. A chatbot can summarise notes you provide. An agent might choose which record to open next, within the access it has been given.
Anthropic explains this distinction. Products use the words differently, so ask what the software actually chooses and does. OpenAI's guide describes the model, tools and instructions behind it.
Give it a useful, limited job
For the imagined meeting, the request could be:
Read these approved notes and prepare a brief: recent changes, previous commitments and unanswered questions. Link each point to its source. Leave anything uncertain open. Do not send messages or change records.
The brief might help you prepare a better conversation. You still need to understand it. If it says a delivery date was agreed, open the source. Was it agreed or merely proposed?
Paul explores this in In the loop. Your attention is part of the work.
Try the idea before adding a system
Use a few made-up meeting notes with a colleague. Even an ordinary chatbot can help you practise checking a brief. An agent may be useful when finding the answer requires different steps each time; an existing report may already be enough.
If you explore an agent, limit its access, time and spending and make it easy to stop. Those limits need to work in the software; a prompt alone cannot enforce them. Material it reads is information, not permission to change its job.
The takeaway is knowing what the agent does, what it may access and what you still need to check. Consider whether the whole job improved.