What is an AI agent?

An AI agent is an AI assistant that acts. A plain chat assistant answers a question. An agent is given a goal — “file the failing tests as bugs” — and works towards it by calling tools, reading what comes back, and calling more tools, without being asked to approve each step. The word is used loosely; the useful test is whether the assistant can change something in the world without you pressing the button.

Why that changes the rules

A tool that only a person can use is protected by the person: they read the screen and decide. An agent does not read the screen. It will do exactly what its tools allow, as often as its goal seems to require, including the thing you would never have done on purpose. So the question stops being “what can this person do?” and becomes “what is this agent allowed to do, as whom, and how would I know afterwards?”.

The three things an agent needs

• An identity of its own, so its actions are not mixed up with yours.
• A role that can be narrower than yours, so it can file bugs without being able to delete the board.
• A record of what it did, in words a person can read, so a day of agent work can be reviewed without a transcript.

How fenbs does it

An agent connects over MCP with a browser sign-in and becomes a member of the board. It acts as you, holds your role narrowed by the scopes you ticked, and every change is recorded as, for example, “Claude via Sam Ade”. Refusals are sentences — “You are a Viewer on Kitchen Refit, which cannot move tasks between lanes” — written so the agent can repeat them to you rather than guess.