Running a company as a fleet of stateful agents
Domain-based AI roles instead of Scrum. How Temporal coordinates long-lived agents across billing, operations and support — and what that does to an org chart.
Most “AI in the company” stories are about a chatbot bolted onto an existing process. Ours is the opposite: the process is agents. cocharge’s commercial operation — billing, host onboarding, customer support, monitoring — runs as a fleet of long-lived, stateful agents, coordinated by Temporal, organised around domains rather than tickets.
Why durable execution matters for agents
An agent that does anything real needs to survive restarts, retries, rate limits and human-in-the-loop pauses. A request/response loop can’t. Temporal gives us durable workflows: the agent’s state and history are persisted, so a step that calls a model, waits three days for a host to reply, then resumes, is just normal control flow.
val onboarding = workflow {
val host = awaitHostSignup()
val docs = agent.collectAndVerify(host) // may pause for human review
val contract = agent.draftContract(host, docs)
awaitSignature(contract, timeout = 14.days) // durable wait
activate(host)
}Domains, not standups
We don’t run Scrum. We run domain-based AI roles — an Operations Analyst, a Customer Operations Manager, an AI Product Engineer — each a bounded context with its own knowledge, tools and authority. It maps cleanly onto DDD: the agent owns a domain, speaks through well-defined interfaces, and escalates across boundaries explicitly.
The org chart became an architecture diagram. The architecture diagram became the org chart.
What it costs you
This is not free. You trade the legibility of “a person did this” for “a system did this,” and you have to rebuild that legibility deliberately — through audit trails, evals, and the discipline that every agent action is traceable to a workflow step and a knowledge page. Done well, you get something a human team can’t match: an operation that runs continuously, improves by editing, and scales without hiring.