Satya Nadella is right that the firm of the future must own its learning loop. But the loop he specifies is, in the terms of my own work, a reputation-weighted neoclassical labor market – not the generative architecture this moment actually requires. We agree on the destination. We diverge on the economics.

On June 14, 2026, Satya Nadella published a short essay, A frontier without an ecosystem is not stable. It is the most honest thing a hyperscaler CEO has said about the AI economy: it names the concentration risk directly, and it tells every company that the durable asset is not the model but the loop on top of it. I have spent the last several years formalizing the architecture that claim requires — and that work is exactly why I can say where his essay stops, and what it stops short of.

Here is his piece in full.

I’ve been thinking a lot about the future of the firm in an AI-driven economy.

This transition is different than any previous platform shift. In the past, we used digital systems to enhance human capital. This is the first time we can create a real cognitive loop between people and digital systems. That is a mind-bender, because it changes how we even conceptualize work inside an enterprise.

What is at stake is not some digital tool or system and its use, but how organizations continue to learn, build IP, differentiate, and thrive in a world where AI models can continuously absorb the expertise of humans and organizations and commoditize it.

Every company is going to have to build what I think of as human capital and token capital. Human capital comprises the knowledge, judgment, relationships, ingenuity, and pattern recognition of its people, while token capital is the firm’s AI capability it builds and owns.

Importantly, human capital does not become less valuable as token capital grows. It only becomes more valuable! I believe human agency will be the driver of token capital growth. Humans will set ambitious goals, connect dots across domains, build relationships, and recognize patterns that matter most. Without human direction, you have compute running in circles.

This means the real opportunity is not in picking the best model but instead in building a learning loop on top of models where human capital and token capital compound. You can offload a task, or even a job, but you can never offload your learning. The future of the firm is the ability to compound that learning across people and AI.

This requires a new architectural approach where every business is able to build agentic systems that improve over time, while still retaining control over their IP. A company should be able to switch out a “generalist” model without losing the “company veteran” expertise built into their learning system. This is the key “test” of your control and sovereignty in the era ahead.

Companies need to turn their workflows, domain knowledge, and accumulated judgment into AI systems that improve with each use. Private evals should capture whether a model is actually improving against outcomes that matter to the business (not just external benchmarks!). Private reinforcement learning environments should let models grow stronger on real traces from inside the organization. Its knowledge base makes institutional memory queryable and use of tokens more efficient.

This loop becomes the new IP of the firm. I think of it as a hill climbing machine. And unlike most assets, it compounds. Every improved workflow generates better training signal, which accelerates the accumulation of tacit knowledge unique to the firm. The companies that build this early will have an advantage that is hard to replicate, regardless of any new individual model capability.

The last thing any of us want is a world where every company across every sector is ceding value to a few models that eat everything they see. If all the value is accrued by only a few models, the political economy will simply not tolerate it. There is no societal permission for an AI future that hollows out entire industries.

Think about what happened in the first phase of globalization where entire industrial economies were hollowed out by outsourcing. The GDP numbers looked fine on the surface, but the displacement was real and the consequences are still being felt. Let us not bring that dynamic into the AI era, with a small number of AI systems capturing all the economic returns, while entire industries find their knowledge commoditized right out from underneath them.

In my view, our priority has to be building a frontier ecosystem, not just a frontier model, so value flows broadly across every company, every industry, and every country. One where every organization can own the learning loop that encodes its institutional knowledge, compounding its human and token capital.

This is the ethos I’ve grown up with where platforms enable more value on top than is captured inside, and where every company can continuously innovate and build value of its own.

When that happens, companies will create value for themselves and for the economy around them. Employees will see their expertise amplified and their judgment become part of systems that make it replicable and scalable and the benefits accrue to the companies and communities around them.

That is how companies drive value for themselves and the broader economy. And it is the stable equilibrium we should build together.

— Satya Nadella, A frontier without an ecosystem is not stable, June 14, 2026

Where his essay ends is where my argument begins.

Three claims from my own work frame the disagreement. They are not refinements of Nadella’s picture; they are a different architecture for the same goal.

First, the economics changed – not just the tooling. In The Collapse of Scarcity Economics, I argue that the institutions we inherited take scarcity as their organizing principle, and that this assumption expires under computational abundance. When intelligence becomes abundant, inexpensive, and self-improving, what I call Agentic Decoupling, the severing of output from human labor input, neoclassical optimization stops describing the binding problem, and the transaction-cost logic that explained why firms exist begins to collapse (the Coasean Singularity). The successor framework, which I develop in Computative Economics, replaces the scarce good with the generated possibility space. Value no longer comes from allocating known options efficiently. It comes from generating better options at all.

Second, the loop has a precise form, and most “learning loops” do not have it. In Possibility Loops, I specify the loop as a generative cycle: agents propose new actions, the system validates them, funds the ones that survive, and reflects outcomes back so that each completed cycle leaves the organization with a strictly enlarged space of the possible. This yields the sharpest diagnostic in the whole debate, and it is where Nadella’s loop fails it: any loop whose action set is fixed from outside the loop – what I call a Human-Derived Coordination Architecture – implements, at most, a reputation-weighted neoclassical labor market. It makes existing work cheaper. It does not generate capability.

Third, the value and the fairness both live in the attribution layer. Across my work on reputation systems – Domain-Specific Reputation Systems, Citation Honesty in WDAG Governance – the durable, ownable asset is not the workflow and not the fine-tune. It is the validated, attributed record of whose judgment improved which outcome: citation-weighted attribution, non-transferable (soulbound) reputation, a tamper-evident history of contribution. That layer is what stays portable across models, and it is what makes the human contribution real instead of rhetorical.

With that frame in place, the four places where Nadella’s vision and mine part company come into focus. They are one divergence seen four ways.

1. Optimization versus generativity – scarcity economics versus abundance economics.

Nadella: a “hill climbing machine” – private evals against known business outcomes, reinforcement learning on past internal traces, a queryable knowledge base.

My work: that is a machine that climbs a hill someone else already shaped. Optimizing execution over a fixed menu of actions is, by the diagnostic above, a labor market – and the part of it Nadella calls “hard to replicate” is exactly the part the base models are racing to absorb, after which the advantage evaporates. The frontier firm does not need a faster climber; it needs a loop that moves the hill – an endogenous action set, a generated possibility space that grows every cycle. The disagreement is not that his loop is badly built. It is that the optimization frame itself is the obsolete part. He is doing better scarcity economics. The moment calls for Computative Economics.

2. Amplification versus attribution – who the loop is actually for.

Nadella: employees “will see their expertise amplified,” and “the benefits accrue to the companies and communities around them.”

My work: that outcome has a mechanism or it does not happen, and his essay supplies none. A loop built without an explicit attribution-and-stake layer does not amplify human expertise – it expropriates it, converting tacit judgment into firm or vendor capital while the people who supplied that judgment accrue nothing durable. The mechanism I have specified is reputation plus stake: soulbound attribution, skin in the game (AI’s Mother’s Instinct), citation-honest knowledge graphs. And the governance principle that keeps a human a principal rather than feedstock is one with no analog in Ostrom’s commons work – what I call generation parity: the right to propose new actions into the loop cannot be monopolized. Amplification without attribution is extraction with better narration.

3. Sovereignty versus lock-in – infrastructure neutrality as a design property, not a slogan.

Nadella: the test of sovereignty is whether you can “switch out a generalist model without losing the company veteran.”

My work: it is the right test, and his own components fail it. Private reinforcement learning on a vendor’s model bakes the veteran’s expertise into that vendor’s checkpoint; swap the model and you lose it, or you pay the incumbent to migrate it – lock-in wearing the costume of independence. In Possibility Loops, the architecture is infrastructure-neutral by construction: the veteran lives in firm-controlled surfaces – knowledge, validation, reputation – explicitly decoupled from any model’s weights, so the test passes by design rather than by hope. This is also where the conflict of interest is structural and worth naming: Microsoft is a model and cloud vendor. The party urging you to own your loop profits most from the version where the loop runs on its rails. That is not a reason to dismiss the argument. It is a reason to finish it more carefully than the vendor has any incentive to.

4. A real ecosystem versus a federation of silos – the Ostrom inversion and the visibility paradox.

Nadella: every company owning its own private loop adds up to a “frontier ecosystem.”

My work: a thousand private loops is an archipelago, not an ecosystem. Reputation, attribution, and identity are network goods – their value is in portability, in validated judgment and trusted agents moving between organizations. A genuine ecosystem requires shared, neutral connective tissue with credible exit and the right to fork, governed as a commons. That is the Ostrom inversion I develop in Computative Economics: Elinor Ostrom’s principles for governing a shared resource against over-consumption, re-derived to govern a generative resource against degradation. And my empirical work names the precise way Nadella’s version fails in practice. In a study of forty DAOs, the dominant pattern is a visibility paradox: organizations over-invest in visible artifacts – dashboards, workflows, token launches— and systematically under-provision the invisible institutional infrastructure (reputation ledgers, validation, values-drift detection) that actually predicts long-run resilience. Left to default incentives, the substrate does not get built. And so concentration does not disappear in his world. It moves down one layer, from the model to the substrate, and re-forms at whoever owns the rails beneath all the silos.

The disagreement is not about values. It is about whether the equilibrium is built or merely hoped for.

Nadella and I want the same end state: human capital and machine capability compounding, value flowing broadly, no single layer eating the returns. Where we part is that he treats the stable equilibrium as something that emerges if enough firms each run a private loop, and my entire body of work argues that it is a built artifact – and specifies what has to be built. Not a faster optimizer over known work, but a generative loop over an endogenous possibility space. Not “amplified” expertise, but attributed and staked expertise. Not a loop on the vendor’s rails, but an infrastructure-neutral substrate the firm and the ecosystem own in common. Not many private silos, but a forkable, commons-governed standard for portable reputation and identity.

His title is correct, so let me finish the sentence. A frontier without an ecosystem is not stable — and an ecosystem without a generative, attribution-bearing, infrastructure-neutral substrate is not an ecosystem. It is the staging ground for the next concentration. The model is rented. The loop, as he specifies it, is necessary but not sufficient. The substrate is the thing worth owning — and it has to be owned in common to do the work he wants it to do.

---

I have argued this formally in The Collapse of Scarcity Economics, Computative Economics, Possibility Loops, Evolution of Domain-Specific Reputation Systems, and Citation Honesty Mechanisms in WDAG Governance.

— Wulf A. Kaal