# License to Coordinate *No Treaty. No Mandate. No Permission Asked.* By [Pioneering Spirit](https://pioneeringspirit.xyz) ยท 2026-07-31 --- Every so often someone from Silicon Valley discovers cities. It goes about how you would expect. A brilliant person, genuinely brilliant, who has built systems of real consequence, arrives at the observation that municipal permitting is slow, or that water data is hard to get during a drought, or that the same address is spelled four different ways across three different databases. All true. All things I have spent a decade of my professional life on. And then comes the proposal, delivered with total confidence, that is either something we tried in 2013 and watched die for reasons that had nothing to do with the technology, or something that would require rewriting a state constitution the speaker has not read. I do not say this with contempt. The energy is real and I would rather have it than not. But there is a specific failure mode where technical brilliance gets mistaken for domain knowledge, and the tell is always the same: the proposal treats the accumulated institutional sediment as noise rather than as data. Somebody built that sediment. Usually for a reason. Sometimes a bad reason, but a reason you have to understand before you can dissolve it. Chesterton's fence and all that really isn't a book learning lesson so much as a parable that points to experience. So when [Eli Dourado went after Silicon Valley's unworldliness](https://marginalrevolution.com/marginalrevolution/2026/07/the-wisdom-of-eli-dourado.html) about international institutions this week, I recognized the shape of the argument immediately. Ten thousand percent yes. * * * Dourado's specific complaint is worth taking seriously because he has credentials most of the people he is arguing with do not: he has advised the State Department, sat through multiple treaty negotiations, and served on technical committees for UN agencies. His account of that world is not flattering. Treaty processes, in his telling, are less about the stated object of the agreement than about leverage, with participants routinely using multilateral negotiations to hobble rival industries, launder domestic illiberalism through international legitimacy, or steer a technology toward authoritarian defaults. The damage is bounded mostly by the fact that no major country will actually bind itself to anything it does not already want, and inconvenient treaties get quietly reinterpreted into irrelevance. Then the line that does the work: > "International efforts," therefore, is not a reassuring answer to the governance problem; it is the name of another enormous, unresolved governance problem. That is insightful, and it is exactly the kind of thing you only learn by being in the room. It is the international-relations version of me watching a founder pitch a slick 311 app to a city council, confident that better bug reporting will finally fix the potholes, with no idea that the road resurfacing schedule is currently frozen because two agencies are in active litigation over who is liable for a water main that cracked the pavement in the first place. The app is not even wrong. It is just answering a question nobody in that room was actually asking. But here is where I want to push. Dourado is describing one particular institutional form, the intergovernmental treaty process, with real authority. What he is not describing is the full space of things planetary coordination could be. * * * Theodore Roosevelt said he had always believed that wise progressivism and wise conservatism go hand in hand. He meant it domestically, trust-busting married to institutional continuity, but the formulation travels. A wise conservatism about how international institutions currently work is exactly what Dourado is offering. Take it. Believe it. The treaty apparatus is captured, slow, and structurally incapable of binding anyone who does not already want to be bound. That is not cynicism, that is a field report. A wise progressivism about how planetary institutions might work is the other half, and the two are not in tension. Roosevelt's insight was that reform worth doing conserves what functions and replaces what does not. Dourado has handed us an unusually credible account of what does not. The question is what goes in its place. Here the deeper history is more useful than the recent history. Jonathan Blake and Nils Gilman make a point in _Children of a Modest Star_ that should be better absorbed in AI policy circles: the nation-state as the near-universal unit of global governance is not ancient. It is barely a lifetime old. What we now treat as self-evident, that legitimate coordination across borders happens between sovereign states and nowhere else, only became a fully global default around 1960. Before that the world ran on a messier arrangement of principalities, empires, chartered companies, religious authorities, scientific societies, and civilizational blocs, all exercising real authority without waiting for a seat at Turtle Bay. Their alternative, which they call planetary subsidiarity, argues for putting authority at the scale where the problem actually lives rather than pooling it at the national level by default. Climate does not respect the Westphalian grid. Neither does a pandemic. Neither, obviously, does a model. That reframe opens a door most AI governance conversations walk straight past. * * * Here is the door. You do not need six billion people to agree. You do not need a hundred and ninety-three countries to agree. You need a few hundred people to agree, maybe a few thousand. The population of humans who actually push the capability frontier is startlingly small. The people training frontier models, the people running evaluations, the people refereeing what counts as a credible safety claim: that is a community, not a nation. And critically, unlike a treaty delegation, it is not composed of parties whose incentive is to use the process as leverage in some unrelated fight. Dourado's central complaint about international negotiations, that everyone is there for a reason other than the stated one, applies far less to a room full of researchers who share a technical vocabulary and a professional reputation market. Dean Ball's recent argument for independent verification organizations circles this same insight from the policy side. His case is that frontier lab auditing will require extreme technical experts, recruited with lab-competitive pay and real compute access, which government hiring rules structurally cannot deliver. He is right about the mechanism. I want to push on the institutional form, because the form he is reaching for already has a precedent. The precedent is not the UN. It is the [Bulletin of the Atomic Scientists.](https://thebulletin.org/biography/jerry-brown/) In 1945 a handful of Manhattan Project physicists decided that they, not any government, bore an obligation to tell the world what they had built. They did not wait for a treaty body. They started a magazine and, two years later, a clock. Their founding charter was almost aggressively modest: formulate the responsibilities of scientists regarding nuclear energy, and educate the public about what those scientists already understood better than anyone alive. Einstein convened the board of sponsors. Oppenheimer chaired it. It had no enforcement power. It never has. What it had was the accumulated credibility of people who actually knew how the thing worked, willing to put their names in public on what should happen next. In a field small enough that everyone reads everyone, that turns out not to be a weak mechanism. Frontier AI has its Manhattan Project physicists right now, distributed across a handful of labs in the Bay Area, London, Beijing, and Hangzhou. What it does not have is its Bulletin. * * * The raw material is more international than the discourse admits. Academia remains a genuine republic of letters, one that predates the current geopolitical moment and will likely outlast it. AI research in particular carries decades of cross-pollination between American and Chinese institutions that the last few years of export controls and visa friction have strained without severing. The researchers who did their PhDs at Berkeley and now run labs in Beijing. The conference program committees that still span both countries. The author lists on foundational papers that read like a roll call of everywhere. These people are not incidental. They are the closest thing the field has to intermediaries who are legible, if not always trusted, on both sides of the Pacific. And they are already embedded in an institution with centuries of practice at coordinating across hostile borders. The republic of letters kept functioning through the Thirty Years War. It is not obvious that it cannot function through this. A body built from that community would not need to resolve the US-China rivalry to be useful. It would need to do something narrower: establish shared technical protocol. What a credible pre-deployment evaluation actually looks like. How you test for cyber-offensive capability without publishing a manual. What a dangerous capability threshold means in terms specific enough to check. These are questions where the honest answer is that only a few hundred people on Earth can give a competent one, and where the current alternative is each lab writing its own homework and grading it. It would need a home base too, and probably not Washington or Beijing. Track-two diplomacy has always needed neutral ground, and the Chinese diaspora world, Singapore especially, has been playing that role for decades across trade, finance, and technology. A convening built around Chinese and American AI researchers meeting on ground that is trusted by both, with hosts who are fluent in both worlds without being fully of either, is not a new idea. It is the oldest trick in the diplomatic book, just pointed at a new problem. Get people in a room to co-create, as we like to say here in hippie California, and the protocol tends to follow the relationship rather than the other way around. The enforcement is reputational, which sounds soft until you consider the field. Being the researcher who signed off on the model that caused the incident is a career-defining fact in a community this small. Being the lab that would not submit to the protocol everyone else submitted to is a recruiting problem, an insurance problem, and eventually a customer problem. Ball notes that even a voluntary certification regime generates hard market consequences, because few large enterprises would touch an uncertified lab. That logic does not require a treaty. It requires a critical mass. * * * One more thing worth saying plainly, because governance conversations tend to drift toward guardrails and forget the road. An institution like this should not exist only to stop bad outcomes. The same critical mass of researchers who can agree on what a dangerous capability threshold looks like are also the people best positioned to point frontier AI at the problems actually worth solving at planetary scale. Accelerating the parts of medical research that are bottlenecked on search and simulation rather than lab time. Modeling crop resilience and famine early warning well enough to act before the hunger, not after. Building the climate models precise enough to make adaptation and stabilization decisions defensible rather than merely hopeful. A Bulletin does not just ring an alarm. It can also point a telescope. We are looking back at ourselves from a genuinely new vantage point right now, closer to that pale blue dot Sagan described than we have been since the photograph itself, a species with the tools to see its own fragility clearly and, for the first time, some of the tools to actually do something about it at the scale the problems require. That is worth building an institution for, not just defending against one. None of this displaces what only governments can do. Liability rules, export controls, deciding what gets built inside a border, the security work that requires classified threat intelligence. Roosevelt's formulation cuts both ways and I do not want to be the person who reinvents the failure I opened this essay complaining about. But there is a large space between the treaty apparatus Dourado correctly describes as broken and the pure national sovereignty that seems to be the assumed alternative. Blake and Gilman would call the thing in that space planetary subsidiarity: authority sitting at the scale where the problem actually lives. For the question of what constitutes a dangerous capability, that scale is not Washington and it is not Beijing. It is a few hundred people who could probably fit in a conference hotel, most of whom have already met. Silicon Valley's unworldliness is real. I have watched a version of it play out in water and municipal data for ten years and it is genuinely a correctable failure of imagination. But the correction is not to import the General Assembly wholesale into AI policy. It is to notice that history has already run this experiment once, with physicists, under considerably worse conditions, and that the artifact they left behind is still ticking. * * * Further Reading --------------- [**"The Wisdom of Eli Dourado"**](https://marginalrevolution.com/marginalrevolution/2026/07/the-wisdom-of-eli-dourado.html)**, Tyler Cowen on Marginal Revolution.** The post that prompted this one, preserving a tweet that deserved to outlive the timeline. Dourado's vantage point matters here: he has actually sat on the treaty delegations and UN technical committees he is describing, which is rarer in AI policy argument than it should be. [**"What Should Be Done"**](https://www.hyperdimensional.co/p/what-should-be-done) **by Dean W. Ball.** Thirty-five numbered propositions on the current frontier AI licensing mess and how to get out of it. Ball is a former White House AI policy staffer writing with unusual candor about advice he lost arguments over. His case for private verification bodies over government auditors is the strongest section, and his point that you cannot think your way to safety without real-world deployment experience deserves more attention than it has gotten. [**_Children of a Modest Star: Planetary Thinking for an Age of Crises_**](https://www.sup.org/books/politics/children-modest-star) **by Jonathan S. Blake and Nils Gilman.** The Berggruen Institute's argument that nation-state-centric governance is a mid-twentieth-century invention rather than a law of nature, and that planetary subsidiarity is a more honest architecture for problems that do not respect the grid. The historical chapters are the ones to read even if you find the proposal too ambitious. [**The Bulletin of the Atomic Scientists, on its own founding**](https://thebulletin.org/about-us/)**.** Worth reading the Bulletin's account of why it exists. Physicists in 1945 concluding they owed the world a warning, with no mandate and no enforcement power. The Doomsday Clock is the famous artifact. The founding charter is the more instructive one. Writing Coda ------------ I wrote this with voice to text after putting the little ones down to bed. A bespoke Claude skill did the fetching and the checking: pulling Dourado's actual tweet instead of letting me paraphrase from memory, tracking down the Roosevelt line, making sure _Children of a Modest Star_ said what I remembered it saying rather than what would've been convenient. The argument itself, the 311 anecdote, the Bulletin analogy, the instinct to route this through Singapore rather than Geneva, that's mine, worked out in real time across a few passes back and forth. Which is more or less how I want AI to fit into the writing: not generating the thesis, but clearing enough of the research friction that I can spend the walk actually thinking instead of half-remembering a footnote. --- *Originally published on [Pioneering Spirit](https://pioneeringspirit.xyz/license-to-coordinate)*