Innovation when you can't pick winners

Under fire, and out of necessity, Ukraine built an innovation engine that outpaces ordinary military procurement by a wide margin. The interesting part for the rest of us is not the drones. It is the allocation mechanism underneath them, which is a clean, brutal answer to a question every market and every company keeps getting wrong: who should decide what gets built, and who should pick the winners. The mechanism transfers. The thing that powers it mostly does not, and being honest about that gap is the whole point.

Two platforms, one idea

Two systems carry the story. Delta is an open, standardised common operating platform: a live map of the war into which any drone, sensor, or shooter can plug. Brave1 is the part that matters here, the platform many describe as an Amazon for drones. It works on two layers.

The first layer is a directed one. When the government identifies a hard problem, say, Russia fielding jet-powered Shahed drones too fast for current interceptors, it posts a target on Brave1 and funds a contest: build something that exceeds 450 km/h, here is a grant to try. Startups race. That is top-down, problem-first, and it seeds the field with candidates.

The second layer is a market. Frontline units earn a digital currency, e-points, by showing proof of results, and they spend those points ordering the exact kit they want from a public catalog of hundreds of suppliers, from a few people with a 3D printer to firms with thousands of employees. Roughly half of orders now flow through e-points; the rest is centrally allocated. Producers see what units actually buy and iterate against it in weeks. The people closest to the problem pick the winners, and their choices are paid for in a currency they earned by being effective.

Procurement asks a planner what the troops need. Brave1 asks the troops, and lets them pay the answer in points they earned by surviving.

Why it beats procurement

Ukraine learned the failure mode of the alternative the hard way. Early in the war it depended on foreign systems, and the limitation became obvious fast: backers sent what was politically convenient, not what the front needed, after months of debate, and some promised systems never arrived at all. That is procurement's structural flaw in concentrated form. The decision sits with a planner far from the work, optimising for the planner's constraints, on a slow clock.

Brave1 inverts who decides. The demand signal comes from the user, not the planner, so what does not work is filtered out almost immediately instead of surviving for years on a contract. This is the same instinct as the real decision is upstream: the loud question (which drone) is downstream of a quieter one (who gets to choose), and moving the choice to the front line answers the loud question continuously, for free.

It is also a known idea in civilian dress. Eric von Hippel spent decades showing that innovation disproportionately comes from lead users, the people at the sharp end of a need, not from central R&D. Brave1 is that observation turned into infrastructure, with a budget attached.

You cannot pick the winner in advance

Underneath the politics and the slow clock there is a deeper reason procurement fails: the thing it is trying to do may simply be impossible. Stephen Wolfram recently ran the clean version of the experiment. Instead of studying the strategies people happen to submit, which is what Robert Axelrod's famous prisoner's-dilemma tournament did, Wolfram enumerated all possible strategies as small programs and ran every competition between them.

The result is computational irreducibility: which program wins is not something you can read off the rules. There is no formula and no theorem to shortcut it, and in his words, "there's basically no choice but to run them and see what happens." The winners are not even reliably the simple ones or the clever ones; it is fine-grained detail that decides, the kind no specification captures.

Lift that out of automata and it describes any competitive design space, drones and products and strategies alike. You cannot reliably predict the winner by analysis, because winning is a property of the thing running in its real environment, not of its description. A central planner is trying to compute, on paper and in advance, an answer the world will only yield by being run.

You cannot pick the winner by reading the spec. You find it by running the trajectory. So the leverage was never better picking; it is cheaper running.

That flips the whole objective. If the winner cannot be predicted, stop trying to predict it, and make the running cheap instead:

• Lower the cost of fielding a candidate
• Shorten the loop between trying and learning
• Let outcomes select, not analysis

Brave1 is an apparatus for exactly that, a way to run an enormous number of real competitions at low friction and read the winners off the results. Procurement optimises the prediction; Brave1 optimises the search, and only one of those is computable. Tellingly, in Wolfram's full enumeration the celebrated tit-for-tat ranks low and a less forgiving machine wins outright. Even our canonical best strategy turned out to be an artifact of trusting the programs people described rather than running the whole space.

Why capital buys what already won

This is also the cleanest explanation of a familiar pattern: why funds and corporations with effectively unlimited capital still cannot reliably execute a winning strategy, and why they so often end up buying what has already won. Money cannot compute an answer that is only available by running. Capital does not shorten the trajectory. At most it buys more trials in parallel, which is real leverage, but only when the running is cheap and the selection honest. A good venture portfolio is exactly that: many small bets, most of them failing, a few paying for all the rest. It works because it is the marketplace, not because the investor can pick.

What does not work is spending the money to pick: the big strategic bet, the synergy deck, the central plan funded all the way to certainty. That is procurement with a larger budget, and irreducibility says the budget cannot buy the prediction. So when an organisation can neither run cheaply nor pick reliably, the rational fallback is to buy the trajectory someone else already ran: acquire the proven company, pay up in the late round, buy back your own stock, track the index. Purchasing the resolved winner is the one move that never required predicting it.

Infinite money cannot buy a winning strategy. It can buy more trials, or it can buy the trajectory someone else already ran, at the price certainty commands.

Which is the catch, in two parts. A run trajectory is priced: by the time the uncertainty has resolved, the valuation reflects it, so you pay full freight for certainty and capture little of the return that came from bearing the irreducibility in the first place. And the winner you buy is usually fused to the cheap-running environment that produced it. Drop it into a corporate procurement machine and you often kill the very loop that made it win, which is the failure the rest of this piece is about.

The part you can't copy

Before borrowing any of this, name the thing that makes it work, because it is the thing you almost certainly lack. The forcing function is an existential war. The feedback is honest and instant in a way no peacetime metric ever is: a design works or people die, this week, and everyone knows it. Urgency strips out the politics, aligns every incentive, and makes the signal impossible to fake. The e-points themselves are earned by destroying the enemy, a metric as grim as it is unambiguous.

The forcing function was an existential war. That is the part that does not transfer, and pretending it does is how you end up with a suggestion box nobody uses.

Two cautions follow from that:

• The currency distorts. An outcome currency drifts toward what is measurable: tie points to kills and you underfund the boring, slow, or hard-to-score capability, which is precisely why Ukraine kept half of allocation central. The market half finds what works now; the directed half funds what will matter later. It runs both on purpose, and reading "markets beat planners" out of this story is the wrong lesson.
• The signal was free. Any peacetime version of this lives or dies on whether you can manufacture an honest, fast feedback loop, which is the real work, and the part the war handed Ukraine for nothing. This is the metric trap from why data-driven firms converge, met from the other side.

What actually transfers

Strip away the war and a portable design remains: put the choice with the people doing the work, denominate their buying power in outcomes, seed the field with directed contests for the hard problems, and run it all on an open platform anyone can plug into. Here is the same skeleton in three settings.

For a market, the lessons are concrete:

• Stop stockpiling artifacts in a category that turns over in months. The Turkish Bayraktars went from war-winning to obsolete almost immediately, and no amount of inventory would have helped.
• Stay in the race to stay current. The durable asset is the loop: the German firm Quantum Systems keeps current not by planning but by supplying the front and being forced to iterate at frontline speed.
• Do not depend on a single supplier who ships what is convenient to them. The antidote is the sovereignty thread from how to fund the rent-free alternative: build enough of your own capability that an election abroad cannot switch you off.

You cannot stockpile a drone that is obsolete in three months. The durable weapon is the loop, not the artifact.

The same skeleton inside a company

For a company, the translation is sharper still, because most large organisations run the procurement model internally without noticing. A central group picks the one approved database, the one messaging stack, the one approved vendor, and hands it down to teams far from the work, on a slow clock, optimising for the central group's constraints. It is the foreign-aid failure mode with an org chart. The Brave1 inversion:

• Outcome-linked budget. Give the teams that ship discretionary budget tied to outcomes, and let them choose their own tools and platform providers.
• Directed contests. Run internal competitions for the hard problems leadership actually cares about, and fund the winners.
• Open interfaces. Build the internal platform so any team or outside vendor can plug in, the way Delta does, rather than mandating a stack that teams route around.

Usage becomes the vote. The internal tool teams genuinely "buy" with their points survives; the one that was mandated but unused dies, which is the cure for the platform nobody adopted. It is also the way out of the bind in AI as layoffs, not opportunity: when an organisation cannot measure the value it might create, distributing the bet and pricing it by outcome beats trying to score opportunity from the centre.

The catch is the same one. Inside a company you have to build the honest feedback signal yourself, because nothing is trying to kill you this week. Revenue, retention, and incident load are the closest peacetime analogs, and they are slower and noisier than a destroyed target. Get that signal wrong, or game it, and the marketplace just rewards whoever is best at scoring points, which is the failure the directed half exists to catch.

The signal is the engine

The drones are not the lesson. The lesson is an inversion of who decides: procurement and central architecture both hand the choice to a planner far from the work, and Ukraine handed it to the people at the sharp end, tied their buying power to results, and let an open platform turn that into a supplier race. Underneath the inversion is a fact about what is knowable: the winner cannot be computed in advance, only run, so the goal is never better judgment, it is cheaper and faster trials.

You can adopt the loop without the war. But only if you can find the one thing the war supplied for free, a feedback signal honest and fast enough that nobody can argue with it. That signal, not the marketplace, is the engine. The marketplace is just what you build once you have it.

A note on tone: this is a live war, and the engine described here was, in the video's own words, bought hard by the blood of Ukraine's people. Drawing an allocation lesson from it is not the same as admiring the circumstance that forced it. The lesson is worth learning precisely so the rest of us can build honest feedback loops without that price.