AI agent vs AI assistant

How can you tell which one you are looking at?

Four questions separate them reliably:

• Does it take actions in other systems, or only produce text? Actions mean agent.
• Does it run several steps without you prompting each one? Multi-step autonomy means agent.
• Does it need permissions and account access? Agents always do.
• When it finishes, is the work done, or do you now do the work? That is the whole distinction in one question.

The same task, both ways

Forty inbound leads, and you need to know which are worth a call.

The assistant version: you paste each company into a chat, it tells you what to look for, and you research all forty and write your own notes.

The agent version: one instruction, and a companion opens each of the forty in your session, checks the signals you defined, scores them, and hands back a ranked list with the evidence attached. The workflow is at qualify inbound leads automatically.

When is an assistant enough?

When the work is thinking, learning, writing or deciding, and the output is something you read and act on yourself. An assistant is cheap and needs no access to your accounts.

The moment you start copying its output into another system, or working through a list it produced, you are doing an agent’s job by hand. Inside Strawberry the persistent version of that agent is a companion, a personal AI teammate with its own identity and context, while "agent" names the bounded run it performs. The browser-specific definition is at what is a browser agent.

Frequently asked questions