Can I just tell it what to do in plain language?
Yes, and the distinction worth insisting on is between an assistant that answers and one that acts. Ben reaches every system in the product, and each of his capabilities is a real action inside a real account rather than a paragraph describing one. You say a sentence — in the app, out loud, or by text from your phone — and he does the work across as many systems as it takes, then tells you what he touched. Anything that would change state waits for your explicit confirmation first.
Acting, not answering
The common version of this is a chat window that produces a plan you then carry out yourself. The useful version closes that gap: the same sentence crosses the ads system, the CRM and the reporting layer without you opening three screens, and what comes back is a description of what was done rather than what could be.
It also works the other way round. He checks in on a schedule with reports and shift summaries, so the operation reaches you rather than waiting to be asked.
The confirmation boundary
Anything that would actually change state requires your explicit confirmation. That is not a limitation bolted on afterwards — it is the reason it is safe to give an assistant this much reach, and it is the same boundary the ads and outreach systems use.
What it does when it does not know
It says so, plainly, and tells you what it would need in order to know. That sounds minor until you have watched a machine invent a confident reason a pipeline dipped and then watched a month get spent on it.
What this does not claim
- It acts inside the accounts and sites you have connected. It does not act on the open web.
- Anything that spends money or speaks to a real person under your name is staged for approval first, with the cost and reach shown.
- It is not a chat assistant you prompt all day. The systems run on a schedule whether you open anything or not.