The Application Layer and Ontology Are the Real Keys to Enterprise AI

Alex Karp didn't hold back in his latest CNBC interview.

Something has gone completely wrong. Enterprises are livid, paying for tokens that create no value, while these models steal the alpha of their business.

He went further. The winners won't be the companies with the biggest frontier models. They'll be the ones who own the application layer and a strong ontology: the structured layer that turns raw intelligence into something safe, useful, and sovereign.

He's right. And it's the clearest explanation yet of why so much AI spend has produced so little.

The Dirty Secret Karp Named

Every company runs on a pile of systems that don't talk to each other. CRM, billing, ERP, HR, support, infrastructure, product analytics. Each one knows a piece of the truth. None of them know each other. Your CRM calls a company an "account." Billing calls it a "customer." Support calls it an "organization." The warehouse calls it a "tenant." It's the same company every time, and no system anywhere holds that simple fact.

Point a powerful model at that mess and it doesn't get smarter about your business. It gets confidently wrong, because there's nothing honest for it to be right about. Ask it a real question and it fills the gaps with a fluent guess, and now you've got a hallucination in a nice font. Give it broad access to your systems and you've traded one problem for a worse one: you've handed the relationships that make your business your business to infrastructure you don't control.

That's the trap Karp is describing. The intelligence is cheap and everywhere. What's missing is an honest, current picture of the business for that intelligence to work from. No amount of model horsepower fixes a missing source of truth.

SixDegree Is That Application Layer

This is exactly why we built SixDegree.

SixDegree connects to your systems of record and keeps a live map of your business: the people, the accounts, the deals, who owns what, and what state everything is actually in right now. Those four different names for the same company collapse into one, with one owner, one history, and one honest answer to how they're doing.

Every AI agent then works from that same map, whether that's Claude, ChatGPT, a model you fine-tuned, or an open-source model like GLM, Nemotron, or Qwen running entirely inside your own walls.

• People, accounts, ownership, risks, and workflows stay consistent and current, not frozen at the last nightly sync.
• Agents can reason and act with context that reflects how the business actually runs, pulling across systems in a single question instead of one silo at a time.
• You keep control. The map lives in your walls. Your IP and your operational alpha stay yours.

Being model-agnostic isn't a checkbox for us, it's the whole point of sovereignty. Most of the value was never in the model. It's in the map: who your customers are, how they relate, and what tells you where they're headed. Because SixDegree holds that map, you're free to run a private, self-hosted model and keep the entire loop inside your walls. Nothing proprietary leaves the building, no vendor gets to study your business on your dime, and you still get an assistant that actually knows what's going on.

Here's the part that matters more than it sounds: you do not want most of the knowledge coming from the model. A model's knowledge is frozen at training time and stale by months or years. That's fine for a developer asking how a language feature works, because the answer barely moves. It's useless for someone who needs to know what's true about a specific account this quarter, this week, today. The model was never going to know that, and it shouldn't pretend to. The ontology holds the current, business-specific truth. The model just reasons over it.

That's what "application layer" really means. Not a chat box bolted onto a big model, but the layer underneath that holds what things are, how they relate, and who owns them. It's the part that makes AI trustworthy at work, and it's the part almost nobody has built.

Where It Gets Acute: RevOps

Nowhere is this sharper than in Revenue Operations, because RevOps lives at the seam of the whole company. Marketing hands off the lead. Sales works the deal. CS owns the relationship. Finance recognizes the money. Product knows whether anyone is actually using the thing they bought. Five teams, five systems, and no two of them agree on what a customer is.

So much of the job is just stitching that back together by hand:

• Forecasting. The number the CRM rolls up is built from stage fields reps update on Thursday afternoons. The real story lives in usage that's trailing off, tickets piling up, and an invoice nobody's chasing, and the forecast can't see any of it.
• Renewals and expansion. A renewal is at risk long before the CRM admits it. The login trend is sliding, the champion changed jobs, support is on fire, an invoice is 60 days late. Three systems each hold one clue, and nobody connects them until it's a save call.
• Ownership and handoffs. The moment three people have touched an account, who actually owns the renewal conversation? The org chart doesn't know, and the owner field in the CRM is two reorgs out of date.

None of that is a reasoning problem. It's a context problem. Give an assistant the live map and "which renewals are at risk this quarter" stops being a stale probability field and starts pulling from usage, support, and billing at once. It's the difference between AI that sounds confident about a deal and one that can actually see the tickets, the usage drop, and the AE who went quiet.

Why This Matters Now

The mood is shifting. The novelty of a clever answer has worn off, and teams are noticing that more tokens against a generic model mostly buys them cost and risk. What they want now is AI that actually understands their business.

That's the move: from a truth scattered across a dozen tabs to one map everyone and every agent can trust. From answers that sound right to answers that are right. From your alpha leaking out to your alpha staying yours.

We're just getting started. If you're tired of AI that sounds smart but doesn't understand how your business actually runs, we should talk.