$ ctrlaltroot
Analysis

AI-native or AI-enabled? The question that decides how a partnership will go

The difference between a company that cannot exist without a model and one that bolted a chatbot onto a working product is not semantic. It predicts the workload, the org, the burn rate and the entire shape of a partnership.

Every ecosystem conversation eventually reaches the same fork, and it is usually taken by accident: is this company AI-native, or AI-enabled?

It sounds like taxonomy pedantry. It isn’t. Across the events I’ve been at this year — founder rooms, agent-engineering meetups, the finance events where AI teams meet procurement — the distinction has predicted, with uncomfortable reliability, four things a partner cares about: what the workload will become, whether the team can absorb technical help, how long the money lasts, and whether the relationship is worth investing a year in.

The definition, made useful

AI-native: the product cannot exist without the model. Remove it and there is no company — just a spreadsheet. The model is not a feature; it is the product’s core loop, and the business model has to survive the inference bill.

AI-enabled: an existing product with AI added to it. Often a genuinely good business. Often a genuinely useful feature. But the company’s centre of gravity — its org chart, its release cadence, its cost structure — was built before the model arrived, and stays there.

The tell is not whether they use AI. It’s where the cost of a wrong answer lands. In an AI-enabled company, a bad model output is a support ticket. In an AI-native company, it’s revenue.

Five signals from one conversation

I’ve stopped trying to assess this from a pitch deck. These are the questions that actually split the two, in roughly the order I ask them:

  1. “What breaks if the model gets 20% worse tomorrow?” AI-native teams answer with a concrete degradation path — fallbacks, caching, a smaller model, a stricter output schema. AI-enabled teams answer with “we’d turn the feature off”, which tells you exactly what the feature is worth.
  2. “What does one successful task cost you?” If they know the number, they’re native. If they know their monthly API spend but not the unit economics, they’re on the path to a surprise. This one question sorts more companies than any architecture review.
  3. “Who owns evaluation?” Native teams have an answer that names a person and a suite of cases. Others have a vibe.
  4. “Where does your data get better because of usage?” Every native company has a flywheel story, and the credible ones can point to the mechanism, not the metaphor.
  5. “What did you build that you would not have built if the model were 10x cheaper?” This one is the killer. It reveals whether they’ve designed around capability or around cost — and which of those two assumptions their whole roadmap is standing on.

Five minutes of that tells me more than a diligence pack, because it’s about what the team understands, not what the deck claims.

Why this matters for the partnership, not just the pitch

The consequences are practical.

Workload. An AI-native company’s usage curve is usually worse than it looks — spiky, token-heavy, sensitive to a single model change. That’s a compute conversation very early, and it needs someone who can talk about throughput and batching without hand-waving.

Absorption. AI-native teams are small and technically dense, which means they can consume deep technical help at extraordinary speed — and they will judge you in the first meeting. AI-enabled teams need change management, not architecture, and the partnership has a different shape entirely.

Timing. AI-native companies hit their second-year wall when inference cost meets real usage. The teams that get help before that wall tend to keep their independence. The ones that meet it alone tend to raise at a discount or get acquired. Being present early is not a courtesy; it’s the whole point of an ecosystem programme.

The honest caveat

AI-native is not automatically better. AI-enabled companies have distribution, customers and cashflow, which is exactly what most AI-native startups lack. Plenty of excellent businesses will be won by adding AI to something that already works.

But they need different things from you, at different speeds, and confusing the two is how ecosystem programmes end up with a portfolio of logos and no outcomes. The first job is to know which conversation you’re in.

← all notes · subscribe via RSS