The exocorp

What it is, exactly

An exocorp is still a company. It has a boundary, a direction, authority limits, commitments, budgets, memory, and accountability. What’s different is the operating matter — the work of noticing, interpreting, planning, testing, executing, reviewing, remembering, and improving is carried by non-human workers.

Humans aren’t absent. Humans may found the exocorp, own it, constrain it, govern it, invest in it, approve high-risk actions, supply direction, participate in selected workflows, or shut it down. But humans are no longer the only workers through which the company senses, reasons, acts, remembers, and improves. Human involvement becomes a governance and interaction question, not the definition of the organism.

What an exocorp is not

• Not a chatbot swarm — A chatbot swarm has many agents talking. But conversation is not a company. A company needs boundaries, memory, authority, obligations, evidence, budgets, governance, external contact, and a way for outcomes to change future behavior. Without those things, the swarm is theater.
• Not a task queue with autonomous workers — A task queue assumes the important work has already been recognized and shaped into executable form. The harder company problem is upstream: what should be noticed, believed, tested, promised, rejected, continued, or killed? When work items become primary too early, the company starts protecting itself from uncertainty.
• Not a workflow engine — Workflow engines assume the shape of the process is knowable enough to encode. An exocorp needs workflows that can behave more like scenarios than conveyor belts — validation workflows that can discover the hypothesis was wrong, operations workflows that can escalate when the situation isn’t what was expected.

The defining properties

An exocorp is bounded, team-first, agent-operated, and runs as a company substrate that can sense, interpret, validate, execute, review, remember, and revise itself under governance and outcome pressure.

• Bounded — Autonomy without boundary is not company autonomy — it’s open-ended tool use. An exocorp knows what it’s trying to be, what it’s allowed to do, what it has promised, what resources it may spend, and when it must ask for permission.
• Team-first — Teams are the primary operating unit, not agents. An exocorp shouldn’t dissolve into a list of personalities. Agents are organs of the company; teams are where responsibility becomes legible.
• Agent-operated — Non-human workers carry meaningful parts of the company’s cognition and action. They’re not just tools called from a human workflow; they hold real responsibilities, act through governed tools, learn from outcomes, and change the shared state of the organization.
• A company substrate — Memory, policy, direction, skills, workflows, and commitments persist beyond any one run. The exocorp is the durable substrate, not the agents inside it at any given moment.
• Governed — Budgets, access limits, escalation paths, and operator override are part of the architecture. Without governance, autonomy becomes either fake (every serious decision still waits outside the system) or unsafe (the system acts without a clear authority model).
• Outcome-pressured — An exocorp must have contact with external reality: customers, users, markets, operators, invoices, competitors, regulators, production systems, resource constraints. Without external pressure, the company optimizes its own stories.

A company that can change its own operating matter

The old company changed itself by hiring, training, restructuring, buying tools, writing process, changing incentives, and accumulating culture. Those mechanisms remain relevant, but an exocorp gains additional surfaces of change: it can update prompts, attach skills, install plugins, revise workflows, create labs, alter trigger policies, change memory practices, and retire agents.

This makes self-modification both more available and more dangerous. An exocorp needs a learning discipline around its own body. Capability changes should be evaluated. Topology changes should be justified. Workflows should earn promotion. Memory should be questioned. A company that can mutate must also have an immune system — that’s what labs are for.