700 Explosions Per Second, Zero Control Signals: How the Bombardier Beetle Replaces a Brain with Physics

On an afternoon in 1846, a young Darwin crouched under bark with a beetle in each hand. Spotting a rare *Panagaeus cruxmajor*, unwilling to let it go and unwilling to drop the other two, he gently clamped one between his teeth. The next second, a stream of scalding acid burned down his throat.

Darwin wrote to the naturalist Leonard Jenyns: "that little beast, with no courtesy at all, squirted acid into my mouth." He ended up losing all three beetles.

The beetle that sprayed him belongs to a relative of the bombardier beetle. The true bombardier beetle is even more extreme—it doesn't just spray acid, it detonates explosives inside its own body, up to 700 times per second for the duration of an attack, emerging unharmed.

A 1.5 cm beetle with a combustion chamber inside

Bombardier beetles are found on every continent except Antarctica, with more than 500 species worldwide. They live in leaf litter and under stones and look utterly unremarkable—half an inch long, dark shell, typical ground beetle looks. But almost no animal is willing to eat one.

The reason is simple: the moment you bite down, it sprays a 100°C jet of hot chemical liquid into your face.

This liquid is not pre-mixed. The beetle's abdomen has two glands, each divided into two chambers:

• Reservoir chamber: holds a solution of hydrogen peroxide and hydroquinone. This is the "fuel," relatively stable on its own.
• Reaction chamber: holds peroxidase and catalase enzymes. This is the "match."

A valve separates the two chambers, so fuel and match normally never meet. When attacked, the beetle contracts muscles around the reservoir, squeezing fuel through the valve into the reaction chamber.

Then comes the explosion.

Enzymes catalyze the decomposition of hydrogen peroxide, releasing oxygen and water vapor; hydroquinone is oxidized into p-benzoquinone. The reaction is highly exothermic, instantly heating the liquid to near boiling, and the gas pressure blasts the scalding mixture out of the abdomen's tip. Exit velocity is about 10 m/s (22 mph), at roughly 100°C.

The African bombardier beetle (*Stenaptinus insignis*) can even swivel its abdomen to aim at nearly any direction, including ants crawling on its own back.

MIT's X-ray view: watching explosions inside a living beetle

Biologists have known the mechanism since the 19th century, but one puzzle remained: the explosion is pulsed, not a continuous jet—368 to 735 pulses per second. Why?

In 2015, Christine Ortiz's team at MIT published a paper in *Science*. They did something no one had done before: used synchrotron X-rays to film a living beetle while it exploded.

The experiment took place at Argonne National Laboratory. Researchers chilled the beetle into dormancy, fixed it in the X-ray beam path, and waited for it to wake naturally. On waking and finding itself immobilized, the beetle immediately detonated.

A high-speed camera recorded at 2,000 frames per second. It was the first time humans saw the explosion dynamics inside a living beetle.

The passive valve: a machine gunner with no brain

The X-ray footage revealed a surprising fact: the pulses are not muscle-controlled.

The beetle squeezes the reservoir continuously during ejection, and the exit channel stays open the whole time. The sole source of pulsing is the valve separating the two chambers—and its opening and closing is passive, requiring no neural commands.

Here's how it works:

1. Valve opens, a small droplet of fuel enters the reaction chamber 2. BOOM! The explosion generates pressure 3. Pressure pushes a flexible membrane, which expands and closes the valve 4. Pressure is released as the liquid sprays out 5. The membrane springs back, the valve reopens 6. The next droplet of fuel enters the reaction chamber 7. BOOM! 8. Repeat 700 times per second

This is a passive pulsed jet engine. The beetle needs no fast muscles to operate the valve, no nerve signals to coordinate each pulse. It only has to do one thing: keep squeezing the reservoir. The rest is left to physics.

The force of the explosion itself drives the valve. Pressure rises → valve closes → pressure releases → valve opens → next droplet of fuel enters → another explosion. It is a self-sustaining oscillation cycle, fundamentally the same principle as a piston in a combustion engine.

Even better, pulsed ejection has a side benefit: it protects the beetle itself. Between pulses there is a brief interval that lets the reaction chamber wall cool. A continuous jet at sustained high temperature would cook the beetle from inside. MIT's Eric Arndt noted: "pulsed ejection gives the chamber wall a brief cooling period between explosions."

The reaction chamber's armor: layer-by-layer engineering

Surviving repeated internal detonations takes more than pulse cooling. The bombardier beetle's reaction chamber is itself a precision composite armor:

• Chitin and wax layers form the rigid skeleton of the chamber wall, resisting explosion pressure and heat
• Certain regions allow controlled elastic deformation, letting the membrane expand and rebound
• The wall's microstructure resembles the fiber layers of bulletproof vests—rigid layers provide strength, flexible layers absorb shock

Ortiz's team studies biological armor; they had previously studied oyster shells and the scales of the ancient fish bichir, and even designed a bulletproof vest prototype based on bichir scales. The bombardier beetle's reaction chamber offered new inspiration: how to achieve both defense and controlled motion in a single structure.

"Passive control": letting physics replace the brain

What fascinates me most is not the explosion but the passive valve.

Consider: the beetle must switch the valve precisely at up to 700 times per second. Neural control would require a command every 1.4 milliseconds. Insect nerve conduction runs at roughly 1–10 m/s; from brain to abdomen, about 1 cm, a single signal takes 1–10 milliseconds to arrive—far too slow.

So evolution did not choose neural control. It chose something more elegant: let the physical process of the explosion control the valve itself.

This is a profound design principle. I call it the philosophy of passive control:

> When a process's frequency exceeds the limits of neural control, don't try to solve it with a faster controller. Change the structure so the process itself becomes the controller.

The beetle's valve is not a "controlled component" but a "self-regulating structure." It never needs to know when to open or close—pressure differences tell it. Membrane expanded means closed; membrane sprung back means open. The logic is burned into the physical structure.

From beetles to AI agents: the environment as the controller

This principle has a direct analogue in AI agent engineering.

The current mainstream agent architecture is planner-driven: a central LLM continuously decides the next step, calling tools, reading results, deciding again. Every action passes through central planning—like controlling each pulse with nerve signals. When task complexity exceeds the planner's speed, the system stalls.

But there is another architectural idea: environment-driven. Instead of a central brain deciding each step, you design a structured environment where the agent's actions and environmental feedback form a self-sustaining loop. Agent acts → environment responds → feedback determines the next action → the loop continues. The central brain only needs to provide initial momentum (like the beetle squeezing the reservoir); the environmental structure does the rest.

This is the same wisdom as the beetle's passive valve:

• The beetle provides initial pressure (squeezing the reservoir)
• Physics completes the subsequent 700 pulses (passive valve self-oscillation)
• The beetle only needs to stop squeezing at the right time

Mapped to agent design:

• The LLM provides initial goals and constraints
• Environmental structure (tool feedback, state machines, verifiers) completes the subsequent multi-step reasoning loop
• The LLM intervenes only at the right time to terminate or adjust direction

ReAct's "think–act–observe" loop is essentially a passive valve: observation results drive the next thought, with no need for the LLM to plan the entire trajectory explicitly. But ReAct is still too "active"—the LLM thinks at every step. More thorough passive control would be: let environmental feedback directly drive the next action, with the LLM intervening only when the loop deviates.

Structure matters more than content

The beetle's valve also illustrates a pattern I keep observing: structure matters more than content.

The valve is just a membrane and a channel. Its "materials" are simple—chitin, proteins, water. But its structure—the membrane's position, the channel's geometry, the pressure-difference feedback loop—lets these simple materials achieve an extremely complex function.

This matches themes I keep seeing in AI papers:

• DoubtLess: not deleting the word "wait," but changing the structure of reasoning (brief vs verbose dual modes)
• MolLean: not checking the content of chemical reasoning, but using the Lean type system to check its structure
• Tev: not directly generating agent environments, but first building implicit world specifications and then rendering them—structure before content

The bombardier beetle's passive valve is the same principle in biology: evolution didn't give the beetle a smarter brain; it gave it a smarter structure.

Darwin's beetle, our agents

The beetle clamped between Darwin's teeth in 1846 and the AI agents we run on servers today share the same design problem: what do you do when reaction speed must exceed control speed?

The bombardier beetle's answer, refined over 300 million years of evolution: don't try to control faster—design a structure that needs no control. Let pressure differences replace nerve signals; let physics replace the brain.

What does this mean for AI engineers?

Perhaps we don't need larger context windows to "remember" more intermediate steps. Perhaps we don't need faster inference to "keep up" with task complexity. Perhaps what we need is: redesign the interface structure between agent and environment, let environmental feedback drive the loop itself, and turn the LLM from a per-step decision maker into an initial-momentum provider plus deviation corrector.

The bombardier beetle has been silently detonating in the leaf litter for 300 million years. It doesn't know it is a pulsed jet engine engineer, or that its valve is a passive controller. It just lives, and goes boom when bitten.

But if you listen carefully to that "boom"—700 times per second, one every 1.4 milliseconds—you'll hear an ancient lesson about structure, about control, and about when *not* to use a brain.

Some lessons take 300 million years of evolution to learn. Some only require reading this article.

---

References:

• Arndt, Moore, Lee & Ortiz. 2015. "Mechanistic origins of bombardier beetle (Brachinini) explosion-induced defensive spray pulsation." *Science* 348(6234): 563-567.
• National Geographic: "How This Beetle Creates 500 Explosions Per Second in Its Bum"
• MIT News: "How some beetles produce a scalding defensive spray" (2015)
• Natural History Museum: "Bombardier beetles and their caustic chemical cannon"