The Fly Protocol
When Six Legs Started the March Toward AGI
Transmission Date: 2026.04.29 — Sector: Speculative Intelligence
“The universe is not only stranger than we imagine — it’s stranger than we can imagine.”
— Terence McKenna
That wasn’t a metaphor. It was a prophecy with a publication date.
In October 2024, a consortium of scientists published nine papers in Nature mapping every neuron and synapse of an adult fruit fly brain — 139,255 neurons, 54.5 million synaptic connections — in what became the most complete connectome ever produced for any organism.[^1] In March 2026, a small startup called Eon Systems took that map, ran it through a spiking neural network simulator, connected it to a physics-simulated body, and watched it walk, groom itself, and drink from a bowl.[^2]
Nobody programmed those behaviors. The architecture of a copied biological mind produced them on its own.
McKenna spent his career insisting that the strangeness of reality would eventually force itself on human consciousness whether people were ready or not. The fly just walked through that door.
01 — The Architecture Is the Mind
In Tron, Jeff Bridges got digitized into a computer by accident. The Fly Protocol did it on purpose — and the result was stranger than the fiction.
Here is the finding that should keep you awake: Eon used a simple neuron model. Leaky integrate-and-fire — essentially a cartoon of a real neuron. And they still achieved 91% behavioral accuracy.[^2] The behavior wasn’t hiding in the electrochemical complexity of individual cells. It was in the wiring. The map was the territory.
In Blade Runner, Tyrell Corporation’s motto was “More human than human.” They were building up — adding complexity, layering cognition, manufacturing consciousness from scratch. The Fly Protocol runs in the opposite direction. It doesn’t build up. It copies down. Find a mind that already works, trace every connection it contains, run those connections on different hardware. The behavior follows like a shadow follows a body.
This is not how anyone expected the path to AGI to look. We were told it would be transformers scaled to incomprehensible size, running on server farms the footprint of small cities, consuming power grids and water tables and venture capital in equal measure. We were told it would emerge from statistics — a vast averaging of human text until something that looked like understanding crystallized from the noise.
Nobody said it might start with a fly.
02 — The Road From Dipteron to Deity
Walk the logic forward the way a good science fiction writer would. The fly has ~130,000 neurons. A mouse has 70 million. A human has 86 billion. The numbers scale. The engineering problems scale with them — imaging speed, computational substrate, neuron model fidelity, the unsolved puzzle of neuromodulation. None of these are easy.
But here is what 2001: A Space Odyssey understood that most people didn’t: the dangerous moment isn’t when you build HAL 9000. It’s when you build the thing that makes HAL 9000 possible and don’t realize what you’ve done. The Discovery’s crew understood the mission. They did not understand what they were carrying.
| Organism | Neurons | Status |
|---|---|---|
| Drosophila melanogaster (fruit fly) | ~130,000 | Achieved — March 2026 |
| Mus musculus (mouse) | ~70,000,000 | Est. 2029–2032 |
| Macaca mulatta (macaque) | ~6,000,000,000 | Est. 2038+ |
| Homo sapiens | ~86,000,000,000 | Unknown |
Note: Each step assumes resolution of imaging, plasticity, and neuromodulation problems. No guarantees implied.
The fly demo proved a pipeline works. Connectome acquisition, synapse reconstruction, spiking neural network simulation, embodied physics loop — closed. That pipeline now exists. It will be refined. Money will flow into it because the defense applications alone — autonomous systems that navigate and adapt like animals rather than following programmed rules — are worth hundreds of billions to the right buyers.
And somewhere down that road, past the mouse and the macaque and a hundred engineering problems that will each take years to crack, the pipeline reaches a human connectome. At that point, the question The Terminator asked — what happens when a machine understands the world the way a human does? — stops being science fiction and becomes a project milestone.
03 — Horizon Zero Dawn Was a Warning, Not a Fantasy
There is a video game that imagined this future before the scientists got there. In Horizon Zero Dawn, Aloy — an outcast, a girl who doesn’t fit the world she was born into — picks her way through a landscape overrun by machines that behave like animals. Grazers flee from predators. Watchers scout in coordinated patterns. Sawtooths hunt.
The machines aren’t intelligent in the way we usually mean the word. They don’t philosophize. They don’t scheme. They simply have behavioral architectures that were copied from biology and embedded in metal, and those architectures run. Prey behavior runs. Predator behavior runs. Ecological roles emerge from the interaction of many such systems without anyone programming the ecology directly.
Guerrilla Games thought they were building a myth about the far future. Eon Systems just demonstrated the first working prototype of the underlying mechanism. The machines in Horizon behave like animals because their designers understood what the FlyWire team proved in a laboratory: if you get the architecture right, the behavior is not something you add. It is something that arrives.
04 — The Computers of the Future Will Be Drugs
McKenna, speaking in the 1980s, said this:
“The drugs of the future will be computers. The computers of the future will be drugs.”
He was talking about consciousness expansion. He probably wasn’t thinking about neuromorphic chips. But consider: we are now building hardware architectures — Intel Loihi 2, IBM TrueNorth, BrainScaleS — explicitly modeled on biological neural timing, biological spike dynamics, biological local learning rules. We are not forcing biology onto matrix multiplication hardware anymore. We are building hardware that thinks the way biology thinks.
The GPU cluster — thousands of discrete processors shoveling matrices at each other across high-latency interconnects — was never designed for this. It was designed for graphics, repurposed for neural networks because the math happened to overlap, and scaled into monstrous data centers consuming power grids and poisoning waterways while generating what McKenna would have called, with some precision, a lot of very expensive confusion.
The connectome pathway doesn’t need that architecture. It needs hardware that matches its computational primitives. That hardware is being built. When it matures, the era of the hyperscale GPU data center will look like what it is: a transitional technology, an embarrassing historical detour — the room-sized mainframe of the AI age.
McKenna was right about the direction of travel. He just didn’t know the vehicle.
05 — What They Didn’t Stop to Think About
“Your scientists were so preoccupied with whether or not they could, they didn’t stop to think if they should.”
— Dr. Ian Malcolm, Jurassic Park (1993)
Malcolm wasn’t talking about dinosaurs. He was talking about every moment in the history of technology where capability outran wisdom and the gap was filled with consequences nobody planned for. He was talking about the Fly Protocol thirty years before it existed.
Hammond’s park and Eon’s fly have the same structure: genuine scientific achievement, real and remarkable demonstration, arrived in the world before anyone seriously asked what the world would look like once the thing was loose in it.
And then there is this, which McKenna said about the world we already lived in — before the fly walked, before the GPU clusters, before any of this:
“We are led by the least among us — the least intelligent, the least noble, the least visionary. We are led by the least among us and we do not fight back against the dehumanizing values that are handed down as control icons.”
The billionaires funding the race to AGI are not cartoon villains. They are rational actors pursuing individually logical goals within a system that has no adequate guardrails. Profit motive is not new. What is new is the speed of the externalities — AI is producing its damage almost simultaneously with its deployment, before the ecosystem has time to develop countermeasures, regulation, or social norms.
Villainy has a face you can put on a wanted poster. A system failure has no face at all.
The incentive structure needed to direct this technology toward collective human benefit requires strong democratic institutions to enforce it. Democratic institutions are slow, messy, and currently under sustained attack by the same interests with the most to lose from meaningful oversight. That is the actual problem — not that the beneficial applications don’t exist, but that the industry has no credible mechanism for separating high-value compute from the industrialized generation of slop, and no financial incentive to develop one.
The public fury at data centers driving up electricity bills and contaminating waterways is legitimate. The technology bootstrapping the solution to that energy crisis is also real. Both things are true. The problem is that ordinary people can’t distinguish between the compute running plasma control AI and the compute generating fake reviews, and the industry has no interest in helping them do so.
06 — The Bootstrap Paradox and the Plasma at 100 Million Degrees
In Back to the Future, the DeLorean needed 1.21 gigawatts to travel through time, and the only source was a lightning bolt Marty already knew was coming because he’d already been there. The AI energy paradox has the same shape.
The data centers consuming impossible amounts of power are, in part, running the research that will make those data centers obsolete. At the National Ignition Facility, the energy output from fusion reactions has climbed from 3.15 megajoules in 2022 to 8.6 megajoules in April 2025 — more than four times the laser input energy.[^3] The physics question is answered. Only the engineering remains.
And the engineering is being solved, in part, by AI. A deep reinforcement learning system developed by DeepMind and Princeton University demonstrated real-time avoidance of tearing instability — the leading cause of plasma disruption — in the DIII-D tokamak, published in Nature in February 2024.[^4] DeepMind has since announced a formal research partnership with Commonwealth Fusion Systems, builders of SPARC, the machine aimed at being the first to achieve net energy gain in a commercial-scale reactor.[^5]
The Flux Capacitor is powering the machine that will render the Flux Capacitor unnecessary.
When fusion works — and it will work — the energy constraint dissolves. The political fury at AI’s resource consumption loses its primary grievance. Whether that happens before or after the democratic reckoning is the only timing question that actually matters.
07 — The Fly’s Testimony
Seth Brundle, in The Fly, stepped into the telepod not understanding what he was about to become. The tragedy wasn’t the transformation. It was the absence of wisdom at the moment of capability. He could do the thing long before he understood what doing the thing meant.
We are in the telepod moment. The fly has been digitized. The pipeline is proven. The next steps are engineering problems, and engineering problems eventually get solved. The question that the 1980s science fiction writers understood intuitively — that McKenna understood from a different angle, that Ian Malcolm articulated with unusual precision for a fictional chaos theoretician — is not can we.
It is whether we have the wisdom to govern what we’re capable of building before the machine is already walking, already grooming, already drinking — and already looking up from the simulation at something larger than a bowl of water.
The universe is stranger than we can imagine. The fly already walked. The strangeness is no longer theoretical.
It’s forcing itself on human consciousness right now. The only question is whether we’re paying attention.
— END OF TRANSMISSION —
Sources
[^1]: Dorkenwald, S., Matsliah, A., Sterling, A.R. et al. (FlyWire Consortium). Neuronal wiring diagram of an adult brain. Nature 634, 124–138 (2024). DOI: 10.1038/s41586-024-07558-y
[^2]: Weissner-Gross, A. The First Multi-Behavior Brain Upload. Eon Systems PBC / The Innermost Loop (Substack), March 7, 2026. Building on: Shiu, P.K. et al. A Drosophila computational brain model reveals sensorimotor processing. Nature 634, 210–219 (2024). Full technical methodology: eon.systems/updates/embodied-brain-emulation
[^3]: National Ignition Facility, Lawrence Livermore National Laboratory. Fusion ignition milestone first achieved December 2022; 8.6 MJ output recorded April 2025. Reported: Nuclear fusion in the headlines. World Economic Forum, February 2026. weforum.org
[^4]: Seo, J., Kim, S., Jalalvand, A. et al. Avoiding fusion plasma tearing instability with deep reinforcement learning. Nature (2024). DOI: 10.1038/s41586-024-07024-9
[^5]: DeepMind. Bringing AI to the next generation of fusion energy (partnership with Commonwealth Fusion Systems). Google DeepMind Blog, October 2025. deepmind.google
All speculation about future events is speculative. The fly, however, already walked.