Dear Zuck, You just announced Meta Superintelligence Labs.

Dear Zuck,

You just announced Meta Superintelligence Labs. You brought in Alexandr Wang as Chief AI Officer, Nat Friedman for applied research and product, and swept the field clean of talent from OpenAI, DeepMind, Anthropic, GitHub. A war chest north of $14 billion. Llama as infrastructure, Meta AI reaching over a billion users. The ambition is clear, the velocity undeniable.

The Mourinho Parallel: Orchestrating Talent

But I’ve seen this kind of acceleration before… in football, not in AI. Mourinho did not have a Galáctico résumé. But in the span of six years, he won the Champions League with Porto, conquered the Premier League with Chelsea, and secured a historic treble with Inter Milan. Different clubs, different tools, same outcome.

He didn’t just assemble players. He choreographed systems. He didn’t need to be the most talented person in the room. He needed to make the talent work: together, under pressure, and toward one goal.

And yes, that includes managing locker rooms full of divas, tactical savants, and men with million-dollar contracts and teenage emotions. Anyone who has run a team of research scientists knows the feeling: brilliant minds, sharp edges, endless questions, and egos tuned like particle accelerators. You don’t dominate that by force. You do it with rhythm, patience, structure, and the occasional well-timed joke.

You are assembling stars. Talent, traction, tenacity. But who’s tuning the tempo? Who’s stitching the structure? Who’s making sure the brilliance bonds?

You’re not just building better models. You’re racing toward AGI. It’s no longer a question of if, but when. And that when is strategic. The first to cross the threshold may not take it all, but it will take the lead. The strange part is, we still don’t agree on what AGI actually is. (I have my own ideas. It isn’t anthropomorphic, but that’s another article.)

What matters here is that the road to AGI isn’t blocked by weak models. Intelligence is scaling. New paradigms may emerge. Architectures evolve. But the system around them lags behind. Research moves fast. Safety adds friction. UX polishes the surface. Infrastructure strains under the load.

Yet no one defines the language between these functions. That’s the bottleneck. Not compute. Not capability. But cohesion.

Grothendieck's Genius: Reshaping the Terrain

Let me offer another analogy, this time from mathematics. Alexander Grothendieck was not known for clever tricks or tactical flair. He didn’t attack problems head-on. He reshaped the terrain beneath them. He invented entire languages: schemes, toposes, categories. These gave structure to problems that once resisted solution.

His genius was definition. He didn’t climb the mountain. He changed the altitude.

That same shift is what AGI needs now. Not more force. More framing.

LLMs are not the whole answer, but they are not a side note either. They are part of AGI because they create structure through language. They define, they compose, they respond. But left alone, they are islands. What Grothendieck offered mathematics is what orchestration must offer AGI: a unifying grammar. A way to make disparate insights work together, not just coexist.

The Language of the Lab: Affordances and Signifiers

In your lab, as in Grothendieck’s mind, there will be models, teams, constraints, and modalities all whispering across a chasm. What’s missing is a shared framework. A conceptual grammar. A set of affordances and signifiers.

In the language of design:

• An affordance is what a system allows you to do.
• A signifier is how it shows you that the action is possible.

A button labeled "Submit" offers submission. A blinking cursor invites you to type. These cues make a system usable.

In a research lab, the same principles apply. An affordance might be the ability to explore a different loss function, to fine-tune with novel priors, or to plug a new agent module into the reasoning stack. A signifier could be a dashboard, an open channel, a documented API, or even a well-timed nudge from a teammate. Without affordances, nothing moves. Without signifiers, no one knows how.

The path to AGI will not be paved by intelligence alone. It will require discoverability inside the model, inside the lab, and inside the system that binds them together.

The Missing Role: Orchestration, Not Management

And that leads to the missing role. Not a PM. Not a TPM. Not a researcher or a VP. A different kind of presence. A presence rooted in what those roles were meant to be.

You could call it TPM 4.0. A synthesis of research intuition, engineering literacy, product judgment, and conceptual design.

In some ways, it echoes how Apple once worked under Jobs. No bloated org charts. No product managers in the modern sense.

Just a rare, central brain connecting engineering to user need, someone who could hold the whole shape. Jobs could do it because he fused taste, technical depth, and narrative clarity.

AGI labs don’t need to scale that model. They need to return to it. In a talent-dense, high-stakes, small-team environment, the goal is not layers of process. It’s radical coherence. And that coherence doesn’t come from structure alone. It comes from someone who sees the model, the user, the language, and the system… all at once.

The role is not management. It is orchestration. Not to lead every section, but to make sure they can hear each other, feel the timing, and enter in sync. To build the affordance layer of the organization itself. To define the signifiers that make discovery contagious. To shape the environment where brilliance becomes inevitable. A kind of playground. Precise. Demanding. Alive. A place where the work plays itself forward.

My theory of action:

• Define shared affordance languages across research, engineering, and product. Let discovery, iteration, and deployment speak the same dialect.
• Design signifiers that make these affordances visible and actionable, at the right moment, in the right context.
• Structure lean discovery loops. Small, cross-functional cycles where teams prototype, integrate, and validate not just what the model can do, but how it plays with the rest.
• Measure coherence as a first-class metric. Track where effort multiplies, where signals connect, and where structure breaks down. Because incoherence kills faster than inaccuracy.

You have the intelligence. You have the infrastructure. You have the will. Now choose your conductor.

Grothendieck changed mathematics not by calculating more, but by framing better. AGI won’t be built by force. It will be composed. And the system that learns to compose will be the system that composes the future.

— Christophe