this spring every big lab shipped an agent. anthropic put out managed agents. google announced them at I/O. codex grew remote execution. microsoft showed Scout at Build, an always-on agent built on openclaw. four launches, one month.

they all built the same thing. an agent you rent inside their cloud. not one you own.

---

the category got a name

look at what google and anthropic landed on, weeks apart, without coordinating. server-resident agents. running in a sandboxed linux box. defined by files in your repo, AGENTS.md and SKILL.md. one call spins one up, it works, it comes back.

that’s not copying. that’s two companies reaching the same answer because it’s the right one: an agent needs a real machine to live on and durable files that say what it knows. even devin’s team published the same case this month. agents need a real environment, not a stateless chat turn.

we’ve shipped exactly this shape for months. what changed isn’t the architecture. it’s that the market finally has a name for it: managed agents. the giants validated the bet in public, four times in a row. good. here’s the half they won’t build.

you rent theirs. you own ours.

every one of those products is the same deal underneath. the agent runs in their cloud, on their meter, behind their console, and it’s gone the day you stop paying. you don’t get a machine. you get an endpoint.

three things that endpoint can’t give you, and they turn out to be the whole point.

your own subscription. theirs bill you per seat, per call, with your work sitting in their dashboard. ours runs on the Claude or ChatGPT plan you already pay for. same flat bill, no second meter, your data never lands in their console.

more than one vendor. google’s agent runs gemini. anthropic’s runs claude. every lab locks you to its own model, because the agent is the funnel. on a box you own, claude and codex and grok run side by side and you pick the right one per job, not per vendor.

a team, not an agent. theirs is one agent behind an API. yours is as many as the work wants, different kinds, on one machine you control, handing work to each other and texting you from the same chat you’ve already got open.

microsoft just showed you the difference

Scout makes it legible. it’s built on openclaw, which is open source and runs on any box you point it at. microsoft’s move was to wrap it in M365 and a console, so the open thing that ran anywhere now only runs as a rental inside their cloud.

openclaw is also one of the agent types you run on 5dive, on your own machine, no wrapper. same software. opposite deal. that’s the whole story in one product: the labs take what could be yours and rent it back to you.

the open lane

the giants proved the category and built it to live inside an enterprise. the version where one person owns a small team of mixed-vendor agents, on a box they control, on a subscription they already pay for, run from their phone, was never going to come from a company whose agent is its funnel.

so it’s the one we built.

the labs shipped agents you rent. we shipped the team you own.

spin up a team at 5dive, or run the open-source CLI on a box you already have.