There is a dial at the Federal Reserve called the countercyclical capital buffer.1 Its job is simple. When the financial system gets frothy, credit ballooning, asset prices detached from anything, the Board turns the dial up, and the largest banks are forced to hold more capital against their loans.
It leans against the boom. It’s the closest thing American finance has to a governor: a mechanism whose entire purpose is to slow the wheel when the wheel spins too fast. It was created in 2016. It can be set anywhere from 0% to 2.5%.2
It has been at zero every single day since.3
Under 12 CFR 217.11(b)(2)(v)(A), an increase takes effect twelve months after announcement. So a raise cannot happen quietly; there would be a public announcement a year in advance. None is findable.
Not zero through the calm years and raised in the wild ones. Zero. Through the 2021 everything-bubble, through the SPAC mania, through crypto’s first collapse, through commercial real estate’s slow-motion unwind. And this is not an oversight; it is the stated policy. The Fed’s own framework says the buffer should be zero “most of the time.” Thirteen other countries have engaged theirs. Hong Kong, Norway, and Sweden have run theirs at the maximum. The United States has never once turned the dial off its stop.
The brake exists. It has never been touched.4
I want to be careful here, because the easy read is regulatory capture, corruption, cowardice and that read is too cheap. Something more interesting is going on, and it deserves seeing it for what it is.
The buffer requires a vote.
That’s it. That’s the whole flaw. And it is the same flaw in every mechanism we have ever built to restrain a compounding system.
Why “checks and balances” is the wrong metaphor
My last piece argued that flywheels, self-reinforcing loops where each rotation makes the next one cheaper, are now the only durable form of competitive advantage, and that the West has largely stopped building them. The most common objection I got back was reasonable and, I think, wrong:
Fine. But surely two flywheels can hold each other in check.
They can’t. And the reason is mechanical, not political.
A flywheel is positive feedback. The defining property of positive feedback is that it amplifies deviation. So when you point two of them at each other, any advantage, a lucky quarter, a rounding error, a single good hire, compounds into a larger advantage, which compounds again. Symmetry is not an equilibrium. It’s a ridge you’re standing on, with a machine underneath whose only function is to convert a lean into a fall.
Stability requires negative feedback. Two positive loops in opposition is a knife’s edge with a sharpener attached.
Look at what actually happened every time we tried it:
- Labor versus capital. For about three decades this genuinely looked like two wheels in balance, union density feeding political power feeding labor law feeding density, running against the capital loop. Then one wheel got a lead. Union density peaked at 34.2% in 1945 and held above 30% from the mid-1940s through the late 1950s5. In 2025 it was 10.0% overall, and 5.9% in the private sector, the lowest in a century6. It did not stabilize. It tipped, and then it kept tipping, because that is what positive feedback does the moment the counterweight slips.
- The United States versus the Soviet Union. Two compounding systems in direct opposition. One of them collapsed. And what we called “stability” in the interim cost trillions and held the species hostage to a launch code. Balance is not the same thing as good.
- Coke and Pepsi. Boeing and Airbus. The two parties. These persist, but not by checking each other. They persist because two wheels in opposition eventually discover that colluding is cheaper than competing, and that the most efficient move available is to stop pushing against each other and start pushing against the referee, together. Duopolies write ballot-access laws. Incumbents draft the regulations that raise the cost of entry. Opposition converges on capture.
And when opposed loops neither tip nor collude, they oscillate, predator and prey, boom and bust. That isn’t a check. That’s a crash on a schedule.
Two flywheels do not hold each other in check. They race, they merge, or they thrash. What you actually need is a different object entirely, and we have known exactly what it looks like since 1788.
What a governor actually is
James Watt’s steam engine had a problem: give it more steam and it accelerates, and acceleration gives it more capacity to accelerate. A flywheel, in the literal sense, the original sense, the one every business writer since has been borrowing without paying.7
The fix was the centrifugal governor. Two iron balls on hinged arms, spun by the engine’s own output. Faster the engine turns, wider the balls fly, and the arms pull the throttle closed. Slow it down and the balls drop and the throttle opens again.
Sit with the elegance of that, because it is the entire argument of this essay:
The governor is powered by the thing it restrains. The faster the wheel spins, the harder the governor bites.
You cannot defund it, because its power source is the engine. You cannot capture it, because it has no discretion to capture. You cannot lobby it, because there is nobody to lobby. It is a linkage, iron, hinge, and physics. It does not care how important the engine has become.
James Clerk Maxwell wrote the mathematics of this in 1868, in a paper called On Governors, and in doing so more or less founded control theory.8 And Maxwell found something that matters here too: a badly tuned governor doesn’t just fail, it hunts. It overcorrects, oscillates, and can tear the engine apart. So this is not a free lunch, and I’m not going to pretend it is. Automatic control is hard. It is just not impossible, which is the state our institutions are currently in.
We built them all backwards
Now hold Watt’s governor up against every restraint we’ve built for economic flywheels.
- Antitrust is funded by congressional appropriation. Its enforcement depends on the priorities of appointees. It can be starved, redirected, or simply declined.
- Environmental review depends on standing, on permits, on an agency’s willingness to litigate, and, as we saw in June, on whether the Justice Department decides to file on the company’s side.
- The countercyclical capital buffer requires a Board vote, once a year, and the Board has voted zero, every year, since the day it was created.
Every one of these draws its power from a source separate from the wheel it’s meant to restrain. Which means the wheel can buy the power source. And here is the part that should genuinely frighten you; it is not that they can be bought. It’s when.
A wheel’s ability to purchase its own governor scales with the wheel’s speed. The faster it spins, the more surplus it throws off, the more lobbyists and lawyers and national-security arguments it can field, and the more indispensable it becomes to the people who’d have to stop it.
Our governors get weaker exactly as the wheel gets faster. Watt’s gets stronger.
We didn’t build a bad governor. We built the inverse of a governor and hung the same word on it.
The three properties
If you wanted to build a real one, it would need three things, and I’d argue it needs all three or none of them count.
- Coupled. Its strength must be a function of the magnitude of the thing it restrains. Not of a risk assessment. Not of a vote. Of the quantity itself.
- Automatic. No discretion. No annual determination. No enforcement budget that can be zeroed out in a reconciliation bill. The moment a human has to decide to engage it, you have reintroduced the surface the wheel buys.
- Non-negotiable. Changing it must cost more than complying with it. Otherwise you haven’t built a linkage, you’ve built a price.
By that standard, almost nothing we have qualifies. But one thing does, and it’s instructive that it’s the most trivial thing we own:
Stock market circuit breakers. The S&P falls 7%, trading halts. Falls 13%, halts again. Falls 20%, the day is over. Nobody votes. No commissioner exercises judgment. No one can lobby the threshold in the middle of a crash, because the threshold is already tripped by the time anyone picks up a phone. It is coupled, automatic, and non-negotiable, it is a genuine linkage, and it works9.
We have exactly one Watt governor, and we pointed it at the stock ticker.
Progressive taxation was supposed to be another. It has exactly the right shape, automatic, coupled to the quantity, no discretion required at the margin. The top marginal rate sat at or above 90% from 1944 to 1963. Today it’s 37%. Case closed, you’d think: the governor was disengaged.10
Except the real story is worse, and it comes from a hostile witness. The Tax Foundation, no friend of high marginal rates, points out that in the 1950s, when the statutory rate was 91%, the top 1% actually paid about 42% of their income in all taxes combined. By 2014 that figure was 36.4%. The statutory rate collapsed by 54 points. The effective rate moved by about six.
The governor was never engaged.
Even at 91%, the base was so riddled with shelters and exclusions that the linkage never actually gripped the crankshaft. It looked like a Watt governor from a distance. It was always a suggestion, and the suggestion was drafted, then and now, with extensive help from the engine.
Statutory coupling is not mechanical coupling. Watt’s governor is bolted to the shaft with iron. Ours is written on paper, by lawyers, for a fee.
Which brings us to the thing being built right now
I’ve spent this essay on solar and banks and steam engines, and you can feel where it’s going. We are, at this moment, constructing the fastest flywheel in human history. And every governor being proposed for it has the identical defect.
Voluntary commitments… discretionary, and revocable by the party who made them.
- Safety frameworks: internal, unenforceable, defunded on the schedule of whoever holds the budget.
- Regulatory bodies: appropriated, appointed, and staffable with people the wheel approves of.
Every single one draws its power from outside the system it’s meant to restrain, which means every single one gets weaker precisely as the thing gets more important, more capable, and more indispensable to national and economic security.
We watched this exact sequence run in solar. We watched it run in banking. We watched it run in the CHIPS and IRA programs, where the **pro-**flywheel institutions got dismantled in three years, proving, if nothing else, that we’re perfectly capable of turning a dial when we want to.
So here is the only question I think is worth asking about AI governance, and I have not yet heard anyone answer it:
Can you name a single proposed AI governor that gets stronger as capability increases?
Not one that scales with capability in the sense of “we’ll write tougher rules later.” One that is powered by the thing it restrains. Where the throttle closes because the engine sped up, with no committee in the loop, and no budget line to cut, and nobody to call.
If the answer is no, then we haven’t built a governor. We’ve built a brake pedal, and handed the keys to the wheel.
- The wheel does not need permission to spin. That’s what makes it a wheel.
- The governor needs permission to bite. That’s what makes it a suggestion.
Until we fix that asymmetry, every restraint we design is a courtesy, extended by the fastest-moving thing in the system, to the people it is about to run over, for exactly as long as it feels like extending it.
We have known the answer since 1788. It is two iron balls on a hinge, spinning on the engine’s own shaft.
We just never built one for anything that mattered.