Synthesis, narrators and the loudest story

Nobody planned jazz. African polyrhythm and European harmonic theory had spent over a century on the same continent without producing it, the ingredients orbiting each other across plantations, parlors, and church halls without ever quite colliding.

What finally changed the equation was a city. New Orleans in the late 1800s offered a particular kind of friction: a port with no stable center, a post-Civil War social order coming apart at the seams, a geography that pressed cultures into each other whether they wanted the contact or not.

From that pressure came something neither tradition could have generated alone, that needed both to exist, and that, crucially, neither could claim afterward as its own.

I keep coming back to that last part. Jazz got to be itself before anyone could name it on its behalf, defining itself through being played in rooms where the audience was as new to the music as the players were.

There was no global apparatus standing by to explain what this sound was supposed to mean, who it belonged to, or what it might eventually be worth in dollars or cultural authority.

That kind of breathing room, the gap between when something exists and when someone else gets to tell you what it is, has become genuinely scarce, and I think it's worth dwelling on as we watch the next round of emergence happen in real time.

How Synthesis Becomes Visible

Stuart Kauffman called the edge of what can exist at any given moment the adjacent possible, meaning not what is technically conceivable in the abstract, but what is actually reachable given the tools, concepts, and conditions currently in play.

Jazz wasn't in the adjacent possible of either tradition working in isolation. It required the intersection, and the intersection had to be lived in long enough for the music to find its own shape, its own vocabulary, its own internal logic.

The same dynamic governs technology, where the gap between what's reachable and what gets reached tends to be filled by something nobody planned and nobody can quite explain after the fact.

When large language models hit production a few years ago, the adjacent possible for most builders looked like chatbots, summarization, smarter search; variations on the model-as-oracle, where you ask and it answers and that's roughly the end of the interaction.

A model that reads context, executes code, observes the output, and iterates autonomously toward a goal was already within technical reach. The APIs and the inference infrastructure were both sitting right there waiting, but what was missing was the collective frame that would make people want it, and frames take time to assemble because one tool has to be used long enough for the shape of the next tool to become legible.

The adjacent possible is not only where new combinations emerge. Sometimes it is where existing ingredients discover a more stable arrangement of themselves, revealing capabilities that were technically present all along but inaccessible until the path between states became visible.

Miles Davis didn't invent modal jazz by being smarter than everyone before him. He arrived at the moment when bebop had been played long enough that its edges were mapped, and the next move could finally be heard rather than just theorized into existence.

The phrase "played long enough", though, is doing more work than it appears. It assumes an audience that gets to listen, which is the variable most stories about innovation quietly skip over.

The Hidden Variable

There were agentic AI products before the ones that won, and instant payment networks before Brazil's PIX, and social graphs before Facebook. The same pattern shows up almost anywhere you look closely.

The ideas existed, often years ahead of the dominant adoption wave, built by people who understood the technology more clearly than the people who eventually got rich from it.

They were simply early, which in practice turns out to be indistinguishable from being wrong, and the distinction only becomes legible in retrospect, by which point it no longer matters much to anyone who was building at the time.

Bill Gross analyzed hundreds of startups and concluded that timing was the single strongest predictor of outcome, above team quality, above the underlying idea, above funding levels.

That finding raises a question that should unsettle anyone trying to build something genuinely new: how do you actually know whether you're early or wrong?

You mostly don't, because timing isn't really a property of the market in the way the conventional discussion treats it; it's a property of who currently has the microphone, since a product becomes early or on-time depending on whether someone with reach decides to call it the future.

The pointing matters more than the thing being pointed at, which is uncomfortable enough that most builders prefer not to think about it directly.

When the Narrator Is Public

Brazil's PIX didn't follow the sigmoid adoption curve that consultants love to graph onto everything. Launched in November 2020, it processed more transactions than credit cards within two years, and by 2024 it was moving annual volume equivalent to roughly two-and-a-half times Brazilian GDP.

The Banco Central mandated participation from every institution above a minimum size, standardized the protocol, and removed interchange fees, which together eliminated most of the variables that usually make organic adoption slow and uncertain.

The whole thing was closer to the government deciding on a Monday morning that every radio station would now play jazz than to jazz spreading on its own through smoky clubs for a decade.

What Brazil did, underneath the technical mandate, was something stranger and more interesting than building a payments system. The Banco Central decided what payments would mean in this decade, and the banks, the fintechs, and the foreign platforms all had to fit themselves into that story rather than the other way around.

For a moment, the state was the narrator of its own digital infrastructure, which is rare enough in our era that the fact it can still happen feels worth noting. In most places, across most decades, the narrators are private, and they tend to have very different interests in what gets called the future of anything.

The Honest Version

I want to be careful here, because the easy version of this argument slides into a conspiratorial register that I don't actually believe and don't want to write.

The platforms that own modern distribution do produce genuinely useful things, often at a scale and reliability that would be hard to replicate any other way.

Hype cycles, for all the damage they do, sometimes leave real infrastructure behind. The dotcom bust laid the fiber that almost everything since has quietly run on, and even the crypto winter produced some primitives that turned out to matter once the noise died down.

The story I'm trying to tell here is more boring than malicious-plot would have it. Narrative capture is a structural feature of having centralized distribution, and the people running that distribution would be acting irrationally if they didn't use it.

Within that frame, though, the metaverse wasn't really synthesis, and Web3 wasn't really synthesis. They were ideas with possible legs, accelerated past their actual readiness by an apparatus capable of making almost any concept feel inevitable for as long as the apparatus stayed pointed in its direction.

Billions of dollars moved on the strength of stories told by the same actors who stood to profit if you believed them, and when the hype eventually receded, the underlying technology was roughly where it had been before, still interesting, still partial, still unfinished.

The capital, attention, and talent that had moved during the boom mostly didn't move back afterward, which is the part of the cycle that tends to get under-discussed.

The move has a certain elegance to it: drown the real synthesis in a fake one, ride the hype cycle until it collapses, and keep whatever infrastructure got built in the meantime.

The dominant capability of the platforms turns out to be neither technical nor financial nor even regulatory, but narrative. Within that capability, hype functions as the actual moat that everything else defends, and fear of missing out is the operational tool that keeps the moat filled.

Anyone who isn't the narrator ends up as a character in someone else's story, no matter how original the thing they're actually building happens to be.

Too Fast for the Dance Floor

Bebop musicians used to play at tempos deliberately too fast for dancers, a conscious refusal to let the music become background, a way of making the art inaccessible on purpose in order to protect what it was.

There's a kind of grace to that move that I find moving every time I think about it. The idea that you can carve out space for a thing by making it unwelcoming to the wrong audience, that formal difficulty itself can be a kind of armor.

I think about it often, and I think about why the equivalent move doesn't really work anymore.

The platforms don't need to absorb your work as background music. They don't need to dilute it, or co-opt it, or even acknowledge that it exists. They simply tell a louder story about something else, and your work becomes invisible without ever having been silenced.

There's no tempo fast enough to escape that, no formal move clever enough to refuse it, because the refusal isn't being heard in the first place; it's happening in a room nobody is currently looking at.

This is the part of the contemporary situation that I find genuinely difficult to think clearly about, and that I haven't seen good answers to in most of the writing on it, including my own.

What's Orbiting Now

Three things are circling each other at the moment, close enough to feel like proximity but not yet close enough to produce the collision that makes everyone say of course in the same slightly embarrassed tone.

Open-weight models are becoming cheap enough that inference is shifting from a privilege you rent to a commodity you can run, which changes the economics of who can credibly build at the frontier.

Protocol-level identity, the boring, slow work of letting agents and humans authenticate to each other without routing through a platform first, is finally getting serious attention from people who aren't trying to sell you a token along with it.

And the loops are getting longer. Agents that used to fall apart after a few steps are starting to chain work together over hours instead of seconds, which means the unit of digital labor is sliding from the individual prompt toward the longer-horizon goal.

Picture what happens when those three click together properly. You'd have an agent that you actually own, running on inference you don't rent from a hyperscaler, authenticating into the systems where you have business on your behalf without an intermediary platform deciding in advance what it's allowed to see or do.

That's the New Orleans of this particular moment, or at least the version of it I find most plausible. It isn't a product yet, and the people closest to it can't quite describe it without sounding either grandiose or paranoid, which in my experience is usually a sign that something is genuinely close to happening.

Whether the synthesis happens isn't really the open question, since synthesis tends to find its New Orleans eventually. The conditions that produce it are messier and more distributed than any single actor can fully suppress.

The open question is who gets to narrate it once it arrives.

If the storytellers are the same actors who own the distribution, the synthesis shows up wearing their colors and producing rent for them, regardless of who actually built it. We've watched that playbook run cleanly twice in five years with no particular reason to assume it won't run a third time.

A different outcome would require a different narrator: an open protocol legible enough on its own terms that the platforms have to fit themselves into its story, or a public institution willing to do for some piece of digital infrastructure what the Banco Central did for payments, or a movement with enough organic gravity to write its own history without asking permission.

Every great synthesis in history has looked, from inside, like chaos, and from outside afterward, like inevitability.

The distance between those two views is where everything interesting happens. Possibilities exist before they are recognized, often for years, sometimes for decades, waiting for the path that makes them visible. The narrator arrives afterward, deciding which version of the story you remember and which possibilities were supposedly inevitable all along.

That position is still unfilled at the moment, which means it's still available. I'd gently suggest that builders who are quietly betting on someone else stepping into it on their behalf are making the same mistake every previous generation of builders has made, which is assuming the music gets to name itself.