If you can't define it, you can't store it

Jaya Gupta and Ashu Garg published “AI’s trillion-dollar opportunity: Context graphs” in late December. It went viral for good reason. The thesis is precise: the last generation of enterprise software became trillion-dollar companies by owning systems of record. The next trillion-dollar opportunity isn’t adding AI to existing systems. It’s capturing what enterprises have never systematically stored - “decision traces”. The reasoning connecting data to action was never treated as data in the first place.

So many folks responded in January with “Context is the next data platform.” A lot of folks wrote about building the usual suspects: connectors, indexes, semantic understanding, enterprise memory. Almost all of them have been looking to solve this problem as an observability one - capturing the digital trail of actions, collaborations, and decisions [which is still better than the median where everyone else was still arguing about whether context mattered.]

But amidst all this, everyone seems to miss the structural question underneath.

A decision isn’t an event in a log. It’s not a Slack thread or a deal desk conversation or an exception someone made once. A decision is a composite structure: a goal evaluated against constraints, affecting specific entities, within a governance framework, at a specific point in time, producing an action that creates an outcome. That’s not a graph problem. That’s an ontology problem.

Jaya’s framing of decision traces capture that something was decided. Most approaches [for eg - Glean’s context platform] captures how work happened. But decision anatomy - the structural understanding of what makes this decision this decision and not some other kind of event, requires a semantic layer that no one is building.

We’ve spent the past year building a [8 x 4] cognitive ontology - right from Entity/Concepts all the way to Governance. The hardest layer was Decision itself. It took the better part of our company existence to get right.

Because “what is a decision?” sounds like a philosophy seminar topic until you have to encode it computationally. Then it becomes an engineering problem with real trade-offs.

• Is a decision an event or a structure? Structure.
• Is it defined by its inputs or its outputs? Both [and the relationship between them.]
• Can two decisions have the same inputs but different outputs? Yes, because governance and temporal context change.
• How do you model a decision that contradicts a previous decision? That’s the governance layer interacting with the temporal layer, and it took us three iterations to answer.

Context graphs without ontology are memory without comprehension. The graph remembers. The ontology understands.

The trillion-dollar question isn’t who stores the most traces. It’s who gives those traces “structure”. The cognitive layer that makes context graphs useful for decisions- the thing that lets you ask “which deals should I deprioritize?” and get an answer that understands pipeline coverage, rep capacity, board commitments, and competitive dynamics as a structural relationship - that layer doesn’t exist yet in the modern enterprise.

Should the ontology live inside the context graph, or above it as a separate reasoning layer? We have a bet. And the proof that it is the correct one has been coming in a plenty in the past few weeks.

The cognitive layer that makes context graphs useful for decisions — that's not a storage problem. That's an infrastructure problem. Decision Infrastructure: The Next Competitive Advantage.

‍