At some point, more thinking stops helping. It stops improving the answer and starts becoming a way to postpone the one thing that was always going to be a little wrong: acting on it.
Is there an actual answer for when that point arrives — not a vibe, a line you can compute? There's math for this: one line, from a field built to steer rockets and thermostats, repurposed here for whether to hit publish.
The Rule in One Line
There's a clean way to model this. Say your answer gets better the longer you think about it — call that q(t), quality after t units of thought — with the usual diminishing returns: the fifth hour helps less than the first did.
Meanwhile, the world doesn't wait for you. Every extra minute of thinking bleeds a little value off the opportunity to act; call that leak rate λ.
Multiply the two together — how good the answer is, times how much of the opportunity is still alive — and the product peaks somewhere in the middle, not at "think forever" and not at "act immediately." Work out exactly where, and the answer is one line:
q′/q = λ
Stop thinking when the percentage rate your answer is still improving drops to the percentage rate the world is moving on without you. Below that point, keep going. Past it, you're not improving the answer anymore — you're just delaying the verdict.
Thought and Action Were Never Two Things
This isn't only a productivity trick. That stopping rule falls out of a much older branch of math called optimal control, the field that works out the best way to steer a system over time: a rocket, a thermostat, a supply chain, and a life.
Optimal control has two classic ways of solving the same problem. One, from Pontryagin, works backward from the goal and carries a running valuation of every choice not yet made — a shadow price on wherever you currently stand. The other, from Bellman, computes a value function over every possible state, and the best action anywhere is simply: move in the direction that increases that value fastest.
Textbooks usually treat these as two different tools for two different situations. Along the best possible path, though, those two objects turn out to be the same object: the valuation you carry backward by deliberating and the direction you read off by acting are one vector, computed two different ways.
Thinking isn't a separate activity that eventually produces action as its output — done right, thinking already is the action-direction. You just haven't moved yet.
One Curve, Two Shadows
There's an older image for this. In 1908, Hermann Minkowski argued that space and time, treated separately, are "doomed to fade away into mere shadows" — that only their union, the curve an object traces through spacetime, is real. Space and time aren't two things added together; they're two projections of one curve, seen from different angles.
Thought and action are the same kind of pair. Not two activities competing for time on a shared clock, but two shadows cast by one curve — the actual shape of what you're doing, moment to moment. Time isn't the container that thought and action both draw down from; it's just how long the curve is.
Here's how much weight that last move is carrying. The physics and the newer math around it prove something real: there's no fixed, outside clock that thought and action are both spending down independently, and the whole idea of time as a container turns out to be less solid than it feels. That's a genuine result.
But going from "there's no container" to "thought and action are one process" is a step I'm taking, not one the math hands me for free. The dissolving of the container makes the bigger claim available. It doesn't make it true.
What I Actually Do With This
The practice version: I give myself a fixed, short window to think about something, then a rule that I have to act — produce one real, external rep, not another draft of a plan — before I'm allowed to think about it again. Whatever happens, I write down the outcome.
Not the idea. The outcome.
The clearest example this year is a pivot that's already on the record: the work moved off generating new ideas and onto compounding the ones already in flight. Technically q was still rising — each new idea a little better than the last — but barely, while everything already on the table sat there losing value at its own steady λ.
Not because the ideas got worse. Because the marginal one stopped being worth more than the next rep on an existing one.
The Part I Still Don't Know
One label in this piece is an overlay, not a theorem: calling the costate "thought" and the state "action." The vector identity itself is exact — λ(t) = ∇V(x,t) — but optimal control never says which of its two variables is which. Active inference is the one field where the science itself asserts perception and action are a single process, not two names laid over one. Everywhere else here, that identification is still a bridge, not a proof.
