We Don't Need AGI. We Need a Nervous System

"Can we have an Artifical Super Intelligence before 2030?"

The wrong question

Everyone is asking when AGI arrives. The better question is what happens if it never does.

My answer: almost nothing we need changes. The intelligence required to run most of the world already exists. What is missing is the body it should drive and the wiring that connects the two.

We have spent years trying to build one mind that contains everything. That is the wrong target. Intelligence is not a property of the model. It is a property of the model plus everything it is wired into.

Take a human brain out of its body, and it does nothing. No sight, no movement, no speech. On its own, it is as dumb as it could ever be.

Wire it to eyes, hands, legs, and lungs, and it drives all of them with ease. The intelligence did not come from the brain alone. It came from the brain being connected.

Neuroscience supports this more than most people expect. Much of the cortex is generic tissue. In rewiring experiments on ferrets, researchers routed visual signals into the part of the brain meant for hearing, and that region learned to process sight. Sensory substitution devices let blind people perceive their surroundings through touch on the tongue.

The brain does not arrive knowing what it controls. It learns its body from the signals it receives. Capability lives in the wiring.

Local brains and the global brain

The body does not run on one brain. It runs on many.

Your spinal cord contains circuits that generate the rhythm of walking on their own. Your hand pulls back from fire before your brain knows anything is hot. Balance, heartbeat, and breathing all run below conscious thought. Each of these is a local brain. It owns one variable, controls it well, and does nothing beyond it.

The global brain does something different. It does not walk, balance, or breathe. It fuses the local brains into relationships. It moves your leg and your hand in sync. It lets you walk and talk and carry a bag at the same time. It decides which local brain gets priority when two of them want the same resource.

Most importantly, the global brain never learns how each function works. It sends "walk," "faster," "turn left." The local brain already knows how. The knowledge of each function is encoded where the function lives.

The global brain never touches the mechanics. It talks to every part through the same socket, and each part keeps its own expertise.

This is why intelligence can be plug-and-socket. If every local brain carries its own expertise, the global brain does not need to know everything. It needs a standard way to plug into anything and drive it.

The socket carries intent, not internals. To work, it has to carry four things:

1. Intent in: what the global brain wants done, at the level of goals.
2. State out: what the local brain is doing right now, continuously.
3. Limits declared: what the local brain can and cannot do.
4. Escalation up: a way for the local brain to say "this is beyond my variable." Pain is exactly this signal.

The global brain still learns, but it learns something smaller. When you lift a heavy box, your legs brace before your arms take the load. Neither the legs nor the arms know that. The coupling between them belongs to the global brain. It learns relationships between parts, not the parts themselves. That knowledge transfers. Plug the same global brain into a new body and the couplings it already understands still apply.

This quietly redefines what "general" means. We do not need a general knower that contains the world. We need a general driver that can run any body it is plugged into. That is far more achievable, and far more useful.

Who is already building it, and why

This architecture is not hypothetical. Three groups are building toward it, each with a different shape and a different incentive.

Robot companies are building the full body, fused shut. The humanoid labs have landed on this exact hierarchy. A slow, large reasoning model plans. A fast, small policy moves the joints. Figure's Helix runs a 7-billion-parameter model at 7 to 9 updates per second above an 80-million-parameter motor policy at 200 (Decrypt). Helix 02 added a third network at a thousand updates per second for balance and posture (Atoms Frontier). Global brain, local brain, reflex. But the layers talk through private signals and are trained together. It is not a socket. The incentive is clear: the coupling is the moat. Open the socket and the body becomes a commodity.

Industrial automation already has its local brains. Factories run on controllers that speak OPC UA, a vendor-neutral industrial standard. The emerging stack puts OPC UA at the bottom for machine data, MCP in the middle for AI to reach tools, and A2A on top for agents to coordinate (ARC Advisory). At Hannover Messe 2026, Beckhoff showed MCP-based voice control of industrial robots, and gateway vendors shipped MCP support (ChatForest). This is the closest thing to true plug-and-socket today. The incentive: incumbents protect their installed base, so AI sits on top of their stack rather than replacing it.

Model labs are building the nervous system standards. In August 2026, Google donated A2A to the Linux Foundation's Agentic AI Foundation, the same home as MCP. MCP connects an agent down to tools. A2A connects agents (FlowHunt). The incentive: the labs sell the global brain. They want sockets open and bodies interchangeable, because cheap bodies make their brain more valuable.

The layer nobody owns yet

Look at the incentives together, and the gap is obvious. Brain sellers want open sockets and cheap bodies. Body builders want a fused, proprietary brain. Incumbents want sockets layered on their legacy systems.

Nobody owns the open nervous system for the physical world. The robot labs solve it privately, inside one body. The protocol foundations solve it for software. MCP is still call and response: the model asks, the tool answers. It has no live shared state, no timing, no reflexes, and no way for a tool to escalate on its own. Those are exactly the four things a real socket needs.

This matters most where the world is least wired. Most of the African economy is unsocketed: paper processes, informal ledgers, machines and services that expose no interface at all. We do not need to win the race to build the biggest brain. That race is expensive and already crowded. The brain will be available to anyone who can plug into it.

The real work is building the bodies and the nerves. Define the sockets. Encode the expertise locally. Let any global brain drive it. Whoever does that owns the position, whether AGI arrives or not.