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 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.