Agents
Switch on an agent and it works through your records in the background: the quote nobody answered, the shipment that went quiet, the booking that's half built, the invoice past due. It reports what it finds, drafts the next step for you to approve, or — if you allow it — does it. You decide which, one action at a time.
Grouped as Settings → Agents groups them. Open one to see what it looks for, what it does, and when it stops.
Can act marks an agent that can do part of its work on its own, on Pro and above. Every other agent reports or drafts on every plan. The agents guide has each one in full.
Write how your team handles something in Settings → SOPs, on Growth and above. The assistant and voice find the SOP and follow it when you ask. Run an SOP as its own agent and it starts from a trigger — an inbox email, a booked shipment, an accepted quote, or a daily sweep — reads the record, and drafts what the SOP says to do for a person to approve. An SOP agent never changes a record or sends anything itself, and it counts as one of your agents.
Not per agent — per action. An agent can draft the customer email while it only reports everything else.
Observe
It finds the work and reports it. It writes nothing.
Draft
It prepares the email or the change, and a person approves it in one click. Emails wait in your review queue with the agent's name on them.
Act
It does it on its own, within its limits. Pro and above, only when you turn it on, and only for the actions the agent allows — sending a quote follow-up, for one.
These are checked by the software before every step, not asked of the model.
Agents spend your plan's included AI actions — there is no separate add-on. A run that only checks your records is free. A run that drafts or acts costs at least one action, and anything past your included actions is $0.30 per action, shown as its own line on your invoice.
Start with an agent at observe. It shows you the stalled work in your own records before it drafts a single email. Raise it when you trust what you see.