NemoClaw and the Shrinking Firm

Every boardroom is asking the wrong question about AI right now.

Will claws take our jobs?

The job is not the unit of analysis. The firm is.

Corporations exist for a reason almost nobody states plainly. Coordinating humans through markets is expensive. Every time you need something done outside the firm you write a contract, negotiate terms, verify quality, chase delivery, and start over next time. Coase worked this out in 1937. The firm exists to make that cost a one-time hire rather than a recurring transaction. Bring someone inside and you stop paying that toll on every task they touch.

This is why firms grow. Transaction costs outside exceed coordination costs inside, up to a point. Past that point you get bureaucracy and politics and meetings about meetings, and the math quietly inverts. Most large organizations are somewhere past that point and don’t know it yet.

Claws break the math entirely.

OpenClaw is where this became observable rather than theoretical. The heartbeat architecture is the thing worth understanding. Every 30 minutes the claw wakes, reads its task list, checks whether anything needs doing, and if a client email has been sitting unanswered for four hours it drafts the reply, sends it, and logs it to memory before going back to sleep. The human set the goal three days ago. They haven’t thought about it since. That gap between setting a goal and every action required to pursue it is where most of management lives. The heartbeat runs inside that gap without anyone watching.

Everything before that was still assistance. Someone still had to initiate. The heartbeat removed the initiation requirement, which turns out to be most of what management actually is.

I keep thinking about how Chinese observers described it: one person with the output of a team. They weren’t talking about productivity. They were talking about something closer to a structural change in what a firm needs to be.

NemoClaw is what makes this usable inside a corporate security posture. OpenShell, sandboxed execution, policy-based guardrails, one command to install. The capability was already there. NemoClaw removed the last reason a cautious enterprise buyer could say not yet.

The first wave of builders around this mostly sold shovels. Setup services, hosting wrappers, deployment tooling. Helicone sold shovels around LLM observability. Koyeb sold shovels around inference hosting. Both got acquihired. First order thinking. It sees the gold rush and sells picks. It misses that the platform owner is watching the same thing, and when they build it natively the shovel seller’s customers evaporate overnight.

The second order question is not what businesses form around claws. It is which businesses become unnecessary because of them.

A firm with three people and claws can cover most of what a thirty-person firm covers. The thirty-person firm has overhead it cannot unwind quickly, compensation structures that require justification, managers whose job is partly to demonstrate the value of their team’s existence. The incentive system inside a large organization runs almost perfectly counter to what this moment requires.

So the disruption probably doesn’t come from inside. It comes from the founder who never hired the thirty people in the first place, who looks across at the incumbent one day and realizes the gap has already closed.

Whether the firm needs to exist at its current size to do what it does is the question most boards haven’t asked yet. Because asking it seriously would require answering it.