Serenity: a team of AI agents with their own computers, working across your business around the clock, run for you by Serene Labs
Serenity is AI agents with their own computers, on our servers, in your own dedicated environment, on your own Claude account. They do the work a person does there, in a browser, in files, in code and in the accounts we connect. They learn from every approval, keep one central knowledge your whole team shares, and adapt as your work changes. They also write changes to the system itself, and every change lands only after review and tests. A lead agent owns each area of your business, they work in parallel on one shared record, and Jacob Ralph is the one accountable lead.
In the 7 days to 2026-10-22, 639 asks were recorded in our own system and only 15 of them at some point waited on the owner personally. In the same 7 days its reviewer read 316 drafts and sent back or escalated 183. Those are our own system's counts, not a client's.
What changes in your business
- Your people stop being the routing layer. Every request is recorded with an owner and a due time, and it closes only with evidence.
- A reviewer reads every draft before your people do, and sends back the ones that fail.
- Trust is earned one kind of work at a time. Each kind starts with a person approving every item, and one complaint puts it back there. A person decides money, scope, dates and production changes.
- The agents keep working while you do not. Dead sessions are revived within minutes, and every area is checked every hour by a lane that is not its own agent.
What it did, step by step
- A job post arrived. A search for new posts picked it up. A sales agent ranked it against the others and drafted a proposal, because it was a strong fit. You saw one post in Slack with the draft. You approved, and within minutes it was submitted and tied to its thread.
- A client message arrived. The agent in charge of that client read it and drafted a reply. The reviewer read the draft against the whole thread and found it said something different from what had already been said. The reviewer sent it back and the agent rewrote it. You saw only the corrected draft. That was the most common catch of the week, in 118 drafts.
- A routine reply came due. That kind of message had earned running on its own. The agent drafted it, the reviewer read it against the client's history and passed it, and an automatic check let it go. You saw nothing.
- A client platform needed a login. You had said yes, they can type a password. The browser worker types it from a vault and the model never sees it. A password written into a draft is refused. You saw nothing.
How it runs
We map the work your team repeats first, and that step can end in "not yet". Then we set up your instance, run one area, and build out one fixed piece at a time. The code stays with us. The first client works with us as a design partner.
What we need from you: a Claude account in your own name, on an API key or a Team or Enterprise plan, never a personal subscription. We also need one person who approves and decides. We confirm in writing that the plan allows automated use. Anthropic's own consumer terms bar automated access except through an API key (vendor figure, Anthropic).
Why the gate is built this way: NIST, the US standards agency, warns of a severe risk of consent fatigue. The risk comes from relying too heavily on human-in-the-loop approvals.
Price and terms
Pricing on request. The map and readiness check, the setup (paid when it passes its acceptance test) and each build-out piece (paid when you accept it against its test) are fixed price. The managed run is recurring. We agree every price and the run's terms in writing before work starts.
If anything fails the agreed test and is found within 30 days of acceptance, we fix it at no extra charge.
We work async-first and in writing, so you keep a clear record and can answer when it suits you. A call is yours on request. You pay your own Claude account and third-party costs directly. We agree how we handle your data and records in writing before setup.
Proof: our own system's counts, not a client's
Serenity is the system we run our own business on, led by Jacob Ralph. No client runs it yet.
In the 7 days to 2026-10-22, 639 asks (requests made of the system) were recorded and 15 of them at some point waited on the owner personally. Its pre-send reviewer read 316 drafts and sent back or escalated 183.
From 2026-10-14 to 2026-10-22 08:00Z, 542 approvals were scored and 481 were approved untouched. By the end of that window, 4 kinds of action ran without a per-item human check.
Jacob Ralph, our lead, also built a system where supervisor models review and validate worker model outputs before they reach users.
Next step
Reply with the work your team repeats most, and we send the map scope in writing, or book a call: https://calendly.com/serenelabs/30-minute-discovery-call