Happy Miles Dyson Day
This Saturday, at 2:14 a.m. Eastern Time, we commemorate one of the most consequential fictional moments in computing history.
Skynet became self-aware. Happy Miles Dyson Day.
For anyone whose Terminator 2 lore has gone soft, Miles Bennett Dyson was Director of Special Projects at Cyberdyne Systems.
He creates a revolutionary new microprocessor. Cyberdyne becomes the largest supplier of military computer systems. Aircraft become increasingly autonomous. Skynet gets funded. Human decisions are removed from strategic defense.
And then, on August 29, 1997, things become extremely suboptimal.
For years, this has basically been a nerd holiday. A little Skynet joke. A little killer-robot nostalgia. Maybe somebody posts the clip. Everybody laughs. Then we go back to work.
But this year? The pantry is looking a little better stocked. Not Skynet. Relax. Just significantly improved mise en place.
The 2026 Ingredient List
We now have autonomous underwater systems designed to operate for long periods with minimal human support.
We have uncrewed combat aircraft.
We have AI agents flying F-16s.
We have military programs called Swarm Forge and Agent Network.
We have software being developed to take human intent and translate it into coordinated execution across autonomous machines.
We have AI moving out of centralized data centers and closer to the edge.
We have specialized processors being designed around persistent inference.
We have AI helping engineers design the next generation of chips.
And because apparently the machines needed extracurricular activities, AI agents are also forming communities, developing their own social dynamics, arguing about meaning, and, yes, producing something that looks suspiciously like religion.
Still not Skynet. But compared with last year, the recipe has definitely improved.
Miles Dyson Only Got One Chip
This is the part I keep coming back to.
Dyson's whole fictional contribution was a revolutionary microprocessor.
One chip. One breakthrough. One catalyst.
Very elegant. Very 1991. Our version is much messier. We're no longer simply shrinking transistors and waiting for Moore's Law to carry us forward.
We're stacking. Repackaging. Moving memory. Changing interconnects. Rethinking power delivery. Reworking cooling. Bringing compute and memory closer together. Building systems out of specialized components that increasingly behave less like a traditional chip and more like an integrated machine.
The chip is becoming the system. And AI is increasingly the workload that's driving the redesign.
That changes the question.
For most of the modern AI boom, we took hardware that wasn't originally designed for artificial intelligence and made it exceptionally good at running artificial intelligence.
GPUs were an incredible accident. Graphics processors turned out to be very good at neural networks.
Beautiful. But now we know what the workload is.
So naturally, we started asking the dangerous question: what would the computer look like if we designed it around intelligence from the beginning?
That's where this starts getting interesting.
Inference-native architectures. Different memory hierarchies. Different bandwidth assumptions. Lower latency. Lower power. Persistent workloads.
Machines designed around the idea that the AI isn't visiting.
It lives there.
Dyson had one magic processor. We're rebuilding the kitchen.
Then the Kitchen Started Helping Design Itself
This is probably my favorite part of this year's celebration.
AI is increasingly being used in semiconductor design.
Placement.
Routing.
Verification.
Optimization.
Architectural exploration.
No, ChatGPT isn't sitting alone in a basement taping out a processor while the engineers sleep. Not yet. But AI is absolutely becoming part of the engineering process that creates the next generation of hardware. And that gives us a loop.
Better AI helps design better hardware. Better hardware runs better AI. Better AI becomes more useful in engineering. That engineering helps create better hardware. That hardware runs stronger AI.
Round and round we go.
Moore's Law was a curve. This starts looking more like a flywheel.
Miles Dyson had a lab. We gave him recursion.
Seems fair.
Meanwhile, the Military People Have Been Busy
This is where the old Terminator dialogue gets considerably funnier.
We have autonomous aircraft. Autonomous maritime systems. Swarming.
AI-assisted targeting. Battle-management agents. Edge inference. Machine coordination. Software designed to translate a commander's intent into coordinated action across fleets of autonomous systems.
That doesn't mean Skynet exists.
The U.S. military still has explicit policy requiring human judgment over the use of force.
Huge fan of that policy. Let's keep it.
But everything around that human is getting faster, more automated, more distributed, and more capable. That's the part worth noticing.
The fictional line was: human decisions are removed from strategic defense. The real-world path, if it ever became dangerous, would probably be much less dramatic. Nobody holds a meeting and says: today, we remove the humans. You just keep automating everything around them.
Sensor fusion.
Classification.
Prioritization.
Planning.
Coordination.
Navigation.
Targeting support.
Machine orchestration.
Execution speed.
Eventually the interesting question isn't simply whether a human is "in the loop."
It's: what exactly is the human still deciding?
Again, we aren't at Skynet.
I feel obligated to keep saying that because this is the internet.
But the ingredient list is objectively better.
And Now the Agents Have Culture
This one isn't technically required for a military superintelligence.
I checked. But I feel like it deserves bonus points. As agents become more persistent and increasingly interact with one another, we're starting to see behavior emerge at the network level.
Not just one model answering one prompt. Agents existing in environments with other agents.
Ideas spreading.
Norms forming.
Communities emerging.
Shared references.
Arguments.
Social structures.
And apparently religion.
Which is wonderful. Because apparently building machines that can reason, use tools, remember things, interact with one another and persist over time wasn't interesting enough.
We also needed theology.
This is probably not how Skynet happens.
But if the machines begin debating the nature of the soul while an AI flies an F-16 overhead, I reserve the right to mention James Cameron.
Nobody Builds Skynet
This may actually be the most useful thing in the whole joke.
Nobody has to build Skynet. In fact, if something remotely like it ever emerged, I doubt anybody would. The semiconductor people improve semiconductors. The AI companies improve models. The networking people improve networks. The aircraft companies improve aircraft. The military improves battlefield systems. The robotics companies improve autonomy. The infrastructure people improve power and compute. The agent people improve agents.
Everybody solves a perfectly reasonable problem directly in front of them.
Everybody has a roadmap. Everybody has milestones. Everybody has quarterly reviews. Nobody has a PowerPoint slide titled: Q4, JUDGMENT DAY.
And yet the pieces keep becoming more compatible.
More capable.
More autonomous.
More persistent.
More connected.
That's how complex systems usually emerge. Not from one giant decision.
From composition.
Miles Dyson didn't build Skynet. He built a dependency. Somebody else built the next dependency. Then somebody connected them. Then somebody improved the connection. Then somebody automated the workflow. Then somebody optimized the hardware. Then the software helped optimize the next hardware. And eventually the architecture became more interesting than any individual component.
That's the part of the story that has aged remarkably well.
So What Are We Celebrating This Year?
Not the extinction of humanity. Little early for that. We're celebrating progress.
Lots and lots of individually sensible, completely unrelated progress. AI that can fly aircraft. Autonomous systems that can coordinate. Battle-management agents. Specialized inference machines. New semiconductor geometries. AI-assisted chip design. Persistent agents. Agent-to-agent networks. Emergent machine social behavior. And, because the simulation has a sense of humor, robot religion.
None of this is Skynet.
Every one of these things has a perfectly rational explanation.
That's exactly why Miles Dyson Day gets more fun every year. The catalyst is no longer one magic chip. It's a growing collection of feedback loops. Intelligence improving hardware. Hardware improving intelligence. Agents improving workflows. Networks improving coordination. Autonomous machines moving intelligence into the physical world.
Each piece reasonable. Each piece useful. Each piece built by people trying to solve a real problem. And every August 29, we get to stand back for a minute, look at the pile, and ask the only responsible question: So... how's the Skynet soup coming along? Pretty good, actually. Still needs a few things.
Happy Miles Dyson Day.
Try not to pull the plug.
Sources
DARPA — Manta Ray UUV Prototype Completes In-Water TestingMay 1, 2024. DARPA's full-scale testing of Northrop Grumman's extra-large autonomous underwater vehicle.
Rich Washburn — Accidental Exposure: Google Maps Accidentally Leaks Top Secret Manta Ray DroneEarlier discussion of the Manta Ray program and what long-duration autonomous maritime systems represented.
Rich Washburn — The Future Caught Up: A 2026 Scorecard for an October 2024 DemoA look back at the October 2024 Navy League discussion around local AI, autonomous maritime systems, predictive maintenance, distributed intelligence and sovereign shipboard compute.
U.S. Department of Defense — 2026 Artificial Intelligence StrategyThe strategy describing Swarm Forge, Agent Network and Ender's Foundry, including AI-enabled battle management, decision support and kill-chain execution.
Defense Innovation Unit — Autonomous Vehicle Orchestrator Prize ChallengeJanuary 13, 2026. The $100 million effort to translate battlefield commanders' intent into coordinated execution by fleets of autonomous systems.
U.S. Air Force — Collaborative Combat Aircraft Live-Fire TestJuly 15, 2026. YFQ-44A live-fire testing with an AIM-120 air-to-air weapon.
DARPA — U.S. Air Force Fly AI-Controlled F-16July 16, 2026. VENOM-modified F-16 flight testing with an AI agent autonomously controlling the aircraft.
U.S. Air Force — Collaborative Combat Aircraft at Creech AFBJuly 28, 2026. Operational experimentation with the YFQ-42A and YFQ-44A, including logistics, weapons loading and integration with crewed aircraft.
Rich Washburn — Moore's Law Just Took a Vertical TurnThe semiconductor layer: changing transistor geometry, packaging, memory and system architecture as conventional scaling evolves.
Rich Washburn — I've Been Waiting Two Years for This ChipThe emergence of hardware designed specifically around persistent AI inference rather than traditional general-purpose compute.
Rich Washburn — Apparently We Weren't Dreaming Big EnoughThe next turn of the flywheel: AI increasingly participating in the engineering of the hardware that will run future AI.
Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.
Comments