Isabelle Bousquette

This is a very overhyped, overused term, and I don't think the industry has a common definition.

But at its base, an agent is an AI that can do something for you.

It becomes agentic when you say agent.

make this restaurant reservation for me, book this doctor's appointment for me, book this flight for me, buy this dress for me.

Anything where it's going out into the world and executing some behavior on your behalf, that's the point at which it becomes an agent.

We're seeing two kind of big categories emerge so far.

One is in the coding space.

There are a million companies that offer this from the Cloud Code and OpenAI Codex of the world to Cursor, Replit, Lovable.

Everyone has an AI coding agent these days.

Engineers are using this a lot.

The other area we've started to see agents take off is in customer service.

It kind of grows out of the traditional phone tree technology that was not really AI-enabled at all.

But now it can be.

If a customer calls with a question about where their order is, an AI agent can handle that call and give status updates or replace a missing loyalty card or things like that.

We know companies are spending a lot of money to use them.

If we take engineering, this is the example a lot of people use because it's just the most mature use case by far.

So every time a coder uses an agent, it costs money.

It costs what they call tokens.

And some companies give their engineers a certain allotted number of tokens.

And some companies are more flexible, saying have as many tokens as you want.