What is the difference between an AI assistant and an AI agent?

An assistant answers; an agent takes actions on your behalf, which makes permissions and approval the whole design problem rather than a detail.

The line is whether it changes anything outside the conversation.
Agents are useful in proportion to what they can reach, and risky for the same reason.
The useful question is not 'is it agentic' but 'what exactly can it do, and what stops it'.

The distinction, minus the marketing

An assistant reads, reasons and writes text back to you. Everything it produces is a suggestion you act on.

An agent does some of the acting. It calls tools, changes records, sends things. The output is not a suggestion, it is a change in the world, and that is a different risk profile entirely.

Most real products sit on a spectrum: reading widely, acting narrowly. That is usually the right shape.

The question worth asking a vendor

Not 'is it agentic', which everyone now says. Ask for the list of actions it can take, and what happens before each one runs.

A good answer is specific: this many actions, these are writes, here is the default for each, here is what an admin controls. A vague answer about safety and guardrails usually means the boundary is a prompt.

How Indexbrain draws the line

It reads widely: 237 read actions across 30 apps, allowed by default because reading is not the risky part.

It acts narrowly and visibly: 141 write actions, all off until an admin enables them, each one stopping for a human before it runs, with no auto-approve mode anywhere on any plan.

Related questions

Give your AI a company to work for

Connect your tools, and every agent you already use starts the day knowing how your company actually works.