His thesis: the durable advantage in AI is your own knowledge, not the model you pick. Ashu Garg sharpened it: the asset isn't the data, it's the decision trace. In regulated industries, that trace isn't just an advantage, it's the compliance layer.
This past Sunday, Satya Nadella published a piece on the future of the firm in an AI-driven economy that has since crossed tens of millions of views. Strip it to the load-bearing claim and it is this: the durable advantage in the AI era is not the model you choose, it is the ecosystem around it, and the most valuable part of that ecosystem is your own company’s knowledge feeding back into the system. The model is the engine. Your accumulated decisions, expertise, and workflows are the fuel.
We agree. But “your company’s knowledge” is the kind of phrase everyone nods at and almost no one operationalizes. What knowledge, stored how, captured when. That is where a sharper version of the same argument matters.
The asset is not the data. It is the decision trace.
Months before Nadella’s post, Foundation Capital’s Ashu Garg made the more precise case in a piece on context graphs. His framing: the last generation of enterprise software won by becoming systems of record. They captured what happened. What they never captured is why. His subtitle says it in six words: “Salesforce knows what happened. It doesn’t know why.”
The missing layer is what Garg calls the decision trace. Not the general rule, but the record of what happened in a specific case and why it was allowed: which policy version applied, which exception was invoked, who approved it, what precedent governed. That reasoning, the part that connects data to action, usually lives in Slack threads, escalation calls, and people’s heads, and was never treated as data in the first place.
Stitch those traces together across entities and time and you get what he calls a context graph: a queryable record of how decisions were actually made, one that becomes the real source of truth for autonomy because it explains not just what happened but why it was permitted. His claim is that this, not the model and not the raw data, becomes the single most valuable asset a company holds in the AI era.
And here is the structural point that matters for everyone building in this space. You can only capture the decision trace if you sit in the execution path at the moment the decision commits. Systems of record receive the data after the fact, by which point the context is already gone. Whoever is in the path when the action happens is the only one who can write the trace.
In regulated industries, the decision trace is not optional
Garg makes the case as a competitive one: the context graph is the next great enterprise asset. In regulated industries it is also something more immediate. It is the difference between an action you can defend and a liability you cannot.
When an AI issues a refund, approves a claim, or moves money, the question a regulator asks is not “what did the system do.” It is show me what happened, on whose authority, under which rule, with what consent, and prove it. A context graph answers the first question. A compliance-grade decision trace answers all of them. Same idea, higher stakes, and no tolerance for “the model decided.”
This is the layer we build at Callvu, and it is why the decision trace sits at the center of it. Because the platform is in the execution path, it writes the evidence as the work happens: the disclosures shown, the consents captured, the signatures collected, the identity-verification proofs, the execution events, and timestamped attestations. Written automatically. Immutable. On every interaction. Not reconstructed later from logs that were never designed to answer the question, but produced at decision time as a first-class artifact.
That is what turns Nadella’s “knowledge as fuel” into something an enterprise can actually stand behind. The knowledge makes the AI capable. The decision trace makes its actions provable. Without the second, the first is just faster exposure.
Three voices, one shape
We describe the working arrangement in a single line: AI evaluates, the system executes within enforced rules, and humans govern. The decision trace is the proof that all three happened. The AI’s judgment is recorded, the rule that constrained it is recorded, and the human authority behind any exception is recorded, immutably, every time.
Step back and the three arguments are the same shape at three altitudes. Nadella, at the level of the economy, warns there is “no societal permission for an AI future that hollows out entire industries” and argues the answer is a frontier ecosystem, not just a frontier model. Garg, at the level of enterprise architecture, says the enduring asset is the decision trace, captured by whoever sits in the execution path. We, at the level of a single regulated transaction, see the same truth every day: value, control, and proof live in the layer around the model, not in the model itself.
Build the ecosystem.
Capture the trace.
Govern what it does.
- Satya Nadella, “A frontier without an ecosystem is not stable,” posted on X, June 14, 2026: https://x.com/satyanadella/status/2066182223213293753
- Ashu Garg, Foundation Capital, “AI’s trillion-dollar opportunity: Context graphs,” December 2025: https://ashugarg.substack.com/p/ais-trillion-dollar-opportunity-context
- TechRadar, coverage and direct quotes from Nadella’s post: https://www.techradar.com/pro/the-last-thing-any-of-us-want-microsoft-ceo-satya-nadella-warns-ai-dominance-could-hollow-out-entire-industries
