Guardrails, approvals & trust
Here is the deal EnClaw offers your organisation: agents may do real work with real systems, because nothing consequential happens without a human in the loop. The guardrails aren't a brake bolted on afterwards — they're the reason the whole thing is allowed on the road.
The principle is simple. Looking is cheap; acting is gated. An agent can read your calendar, search the tracker or summarise a mailbox freely — that's the legwork you hired it for. But the moment it wants to change the world outside — send an email, create or update a ticket, post to a channel — the action pauses, and a human gets asked first.
What an approval feels like
You'll meet approvals as short, self-contained requests: who wants to do what, with the full draft or details attached — the actual email text, the exact ticket fields. They reach you wherever you are: in the web app, in Slack or Teams, or by email. You look, you decide. Approve and the action completes on the spot; decline and nothing happens. If nobody answers in time, the request expires and the action is not taken — the safe direction is always the default.
Two things are worth knowing early. First, an approval is not a formality — the draft you're shown is exactly what will happen, so the glance you give it is the quality gate. Second, some destinations are simply off the menu regardless of approvals: organisations keep whitelists of, say, addresses that email may be sent to at all. If an agent says it can't send to an outside address, that's not stubbornness, that's the fence line.
Who can do what
Underneath, every person in EnClaw has a role — most people simply use agents, while a smaller set can manage agents, curate skills, or administer the organisation — and every agent action is checked against layered rules that decide whether it runs freely, needs sign-off, or is refused outright. You don't need to hold the machinery in your head; its visible face is just those three outcomes: things that happen instantly, things that ask first, and things that politely refuse.
The paper trail
Every action an agent takes — every tool used, every approval given or declined, every message sent — lands in an audit trail your organisation's admins can inspect. For you this mostly matters as reassurance: "what did the agent do, and who said yes?" always has an exact answer. AI at work stops being an act of faith when it comes with receipts.
That's the trust story: bounded abilities, human sign-off, full accountability. What's left is the pleasant part — after an agent has worked, approved and finished, where do the results actually end up when you come looking tomorrow? The last chapter of this overview is a short tour of where your work lands.