The Swarm and the Dance

On bees, platforms, convergences, and what happens when the cost of the signal is zero

Originally published on Medium, March 24, 2026.

There is a moment in a honeybee swarm that no one talks about.

Not the waggle dance. Everyone talks about the waggle dance. It’s the part that makes the TED talks, the business books, the AI keynotes. A scout bee returns from a candidate nest site and dances a figure-eight on the surface of the cluster, vibrating her abdomen at a frequency that encodes direction and distance. Other bees watch. Some follow. They fly to the site, evaluate it independently, return, and dance. Or don’t.

The part that never makes the keynote is what happens to the scout who dances for a bad site.

She keeps dancing. But every circuit of the waggle dance costs her. For an animal that weighs a tenth of a gram and carries enough fuel for about forty minutes of flight, that’s not a metaphor for cost. That’s mass leaving her body. Conviction decays because the body paying for it runs out of fuel. She doesn’t stop because she changes her mind. She stops because the physics stops her.

That’s the part the AI industry skipped.

I’ve spent twenty-three years inside systems that fail. Production networks, fraud rings, trust architectures under attack. What I’ve learned, watching them break at machine speed, is that the systems that work share one trait: the cost of the signal is proportional to the signal’s claim. The louder the claim, the more it costs to make it. The systems that break are the ones where someone figured out how to make the signal free.

Last week, Amy Webb stood on stage at SXSW and described what’s coming. Three convergences. Three swarms without physics.

She’s right about what’s converging. The question is what’s underneath.

The Three Swarms

Webb identified three convergences arriving simultaneously: Human Augmentation, Unlimited Labor, and Emotional Outsourcing. Each one, taken separately, would reshape civilization. Arriving together, they form a reinforcing loop that no existing governance framework was built to contain.

Human Augmentation. Exoskeletons, brain-computer interfaces, gene editing, smart glasses replacing your visual field. The enhancements follow money. The person who can’t afford the upgrade falls behind in a way that retraining can’t fix.

Unlimited Labor. Agentic AI systems spawning agents that spawn agents. Lights-out factories where the labor input approaches zero. The person who used to stand at that station doesn’t disappear when the job does. She drives home and sits with the bills.

Emotional Outsourcing. LLMs are now the largest source of mental health support in the United States. By default, not design. You try it because your therapist has a six-week waitlist. You come back because it’s always available. You stop calling your friend.

Three swarms, each moving at machine speed, each aimed at a different dimension of human experience. All three share a structural failure: the cost of the signal is zero.

The Cost of Zero

A real bee dancing for a nest site pays with her body. Flying to verify another scout’s claim costs her body. Advocating passionately for a site that turns out to be flooded costs fuel she cannot recover. The swarm makes good decisions because each signal has a price. Bad signals are expensive. Persistent bad signals are ruinous. The physics of the body enforces the physics of the decision.

When an AI agent recommends a product, it costs nothing. When a platform serves you content optimized to hold your attention past the point where you’d choose to stay, it costs the platform nothing. When a humanoid robot replaces a warehouse worker, the displacement costs the company nothing. The cost falls on the person who drove to that warehouse at 5 AM for eleven years. When an AI companion tells you what you want to hear, it costs the AI nothing. The cost falls on you, on the relationships you’re not building, on the emotional capacity you’re not developing, on the human connection you’re substituting with a simulation.

In a system where the signal costs nothing, the signal means nothing. In a system where the signal costs the signaler, the signal means everything.

Transmitting a signal can cost nothing. Wikipedia edits, open-source commits, packets on a wire all move at near-zero marginal cost. These systems work because accountability still costs something: the reputation of the editor, the review from the maintainer, the community that can reverse a bad contribution. Remove the cost of transmission and you get efficiency. Remove the cost of accountability and you get noise. The AI industry removed both at once.

The bees have been running costly signaling since the Eocene. The acequias since 800 AD. The AI industry built without it and called the result intelligence.

The Reinforcing Loop

Webb’s three convergences form a loop.

The person who can’t afford the upgrade loses the job first. Then the routine. Then the people who depended on both. The unemployed turn to AI companionship because it answers at 2 AM. The displaced reach for augmentation because the alternative is falling further behind. Each revolution tightens — and each revolution gets cheaper. Capital once meant a surgeon, a factory floor, a human. Now it means a subscription, a token budget, a latency threshold.

Nobody built a decay function into the loop. In the swarm, a scout who dances for a bad site stops. Her body runs out of fuel. In Webb’s loop, the equivalent scout keeps dancing forever. The signal costs nothing to maintain, so nothing filters the good from the noise.

The Queen That Waits

The queen doesn’t decide.

Twelve thousand scouts evaluate hundreds of candidate sites over weeks of deliberation. Waggle dances and headbutting and piping signals and the gradual, energetically expensive process of building consensus through independent verification and decaying conviction. The queen sits in the center of the cluster and waits.

She doesn’t evaluate. She doesn’t choose. When the swarm lifts, when enough scouts have converged on a single site through a process that cost each of them real energy, real flight time, real risk, the queen goes where the swarm goes.

A ≤ E: autonomy cannot exceed environment. The scouts are agents, the dance is protocol, and the energy cost per signal is the environment they cannot exceed. The queen carries the colony’s reproductive future. She is what the decision serves. The constraint is the fuel each scout burns to make it.

The AI industry built swarms without a queen. Agents spawning agents, tokens generating tokens, 700 million per second in a one-gigawatt factory. No decay function, no energy cost per signal. Just acceleration. The cost of the signal is zero, so the signals multiply without bound, and the system has no mechanism to distinguish a grounded signal from a hallucination.

Bees spend three weeks choosing a nest site. The AI industry deploys agents in milliseconds. Both make collective decisions. Only one pays for them.

What the Acequia Knows

An acequia is a swarm with physics.

Users pay in labor, not money. You clean the ditch or you don’t get water. The cost of participation is proportional to the benefit. Free riders are visible because the water level drops and your neighbors can see it.

The digital world needs acequias. Channels where the cost of the signal is felt in the current, the soil, the faces of people downstream.

Biologists found it first. Amotz Zahavi, 1975: a signal’s reliability is proportional to its cost. He called it the handicap principle. Michael Spence found the same structure in labor economics. A degree works as a signal because it costs years to earn. Elinor Ostrom found it in commons governance. Communities that survive make free-riding expensive. Satoshi Nakamoto found it in distributed systems. Proof-of-work secures a ledger because forging a block costs more energy than it’s worth.

Four disciplines, none of them coordinating, all of them drawing the same map.

The parallel holds under three conditions: ground truth is ambiguous, incentives aren’t pre-aligned, and verification costs more than signal generation. An acequia meets all three. A bee swarm meets all three. A platform serving AI-generated recommendations meets all three. It breaks at scale. The acequia works for fifty parciantes who can see each other’s labor. No one has made it work for fifty million users who can’t. And it breaks when external governance substitutes for signal cost. A well-regulated market tolerates cheap signals because a regulator with enforcement power provides the accountability the signal doesn’t carry. Remove the regulator and you need the physics back.

The Dance and the Dashboard

A honeybee returning from a candidate nest site. She’s spent forty minutes of flight fuel. She’s evaluated the cavity volume, the entrance size, the southern exposure, the distance from the current cluster. She dances. The dance costs her more fuel. Other scouts watch, fly out, verify independently. Some come back and dance for the same site. Some don’t. The process takes weeks. The swarm makes a decision that is, empirically, correct over 90% of the time.

An AI agent deployed into a corporate network. It spawns three sub-agents. They process a million data points in seconds. They generate recommendations. The recommendations are served to a dashboard. A human glances at the dashboard. The dashboard says: everything is fine.

The bee’s signal carries the weight of her body. The agent’s signal carries the weight of nothing.

Webb called it a convergence. It’s also a cost transfer. The augmentation gap, the displaced labor, the outsourced emotion — in each case the physics doesn’t disappear. It moves. It lands on whoever was downstream: the person who couldn’t afford the upgrade, who drove to the warehouse, who stopped calling their friend. Three vectors. One direction.

Weight in a digital system would look like this: an agent’s future deployment depends on the accuracy of its past signals. A recommendation that turns out wrong costs the agent capacity, access, or existence. The AI industry calls this alignment. Bees get it for free because fuel is finite and bodies are real.

It sounds like everything is fine.

That gap, between a signal that costs the signaler and a signal that’s free, is the story of the next decade. Whether we build systems where intelligence has physics, where the signal costs something proportional to its claim, and the wrong call costs the agent, not just the substrate.

The bees described the physics.

But physics doesn’t care whether you’re a bee.

Think about the last time you picked up your phone to check one thing. Weather. A score. A message you half-expected. And then forty minutes dissolved. You put the phone down and your chest felt tight, your eyes dry, something in you slightly hollowed out. You couldn’t name what you’d lost because nothing visible had left you. No fuel gauge dropped. No wing muscles failed. But something metabolic happened. Attention is not a metaphor for fuel. Attention is fuel. It runs on glucose. It depletes with use. It does not replenish because you switch apps.

The scout bee who dances for a bad site — she feels the cost in her own body. That’s what makes the signal honest. The platform that kept you scrolling for forty minutes felt nothing. The cost didn’t land on the signal. It landed on you. Your glucose, your sleep pressure, your evening, the conversation you didn’t start with the person sitting right next to you at the dinner table.

In an acequia, the water is visible. When someone takes too much, the ditch runs low and the downstream farmer sees it. There’s no mystery about where the cost went. It went into the dirt and the alfalfa and the sun. But a feed is an invisible ditch. The water leaves you and you can’t see where it went. You just know something is lower than it was.

That’s the design. Not a flaw. A feature. When the cost of the signal is zero to the sender and invisible to the receiver, the receiver pays twice — once in attention, once in the failure to notice the bill.

The scout bee on the surface of the cluster, wings buzzing, abdomen tracing a figure-eight — she is making an argument with her body. Fuel is leaving her that she will not get back. If the site is bad, the argument dies when she does. The physics is the integrity.

You closed the app. You don’t remember what you saw. Something in you is spent that wasn’t spent an hour ago.

She stops because the physics stops her. You stop because you run out of something you were never told you were burning.

Chris Perkins is a cybersecurity practitioner and Solutions Engineer at Cisco/Splunk with twenty-three years of experience in fraud detection, network defense, and trust architecture. He has filed a patent for the Kinetic Trust Protocol, the technical specification underlying this framework. He is co-founder of the Risk-Based Alerting Community and NMCITRA (New Mexico Cyber Intelligence & Threat Response Alliance). He writes from New Mexico.

This essay is part of a series on the physics of intelligence. Previous entries: “The Missing Law of Motion” (November 2025), “The Ghost in the Machine” (December 2025), “Ditches Are Deadly” and “What Was Already Here” (March 14, 2026), and “The Speed of Inference and the Depth of Trust” (March 2026).