We already know what "native" means, because we already use it for something else every day. A native speaker of a language is whoever acquired it as a child, during the window where a brain builds language instead of just learning it. Someone who picks the same language up at twenty-five can study it for the rest of their life, live inside it, think in it, even write better sentences in it than plenty of people who were born speaking it — genuinely, honestly fluent, no asterisk on the skill. What they don't get to become, no matter how long they work at it, is a native speaker. That's not a skill level. It's a fact about when they started, and no later effort moves the start date.
I've only run into this twice, but both times left a mark. The first was a single meeting with a company we ended up not working with — the phrase came up once, I filed it away, and didn't think much more about it. The second is the client I'm working with right now, and this time I can't just file it away, because I don't think I've talked myself all the way through it yet.
Both times shared the same shape: a company that's been operating for decades, under real, time-boxed pressure to show something concrete for its AI investment soon. That pressure is legitimate — it's a specific capability question, and whoever's sponsoring the push is right to take it seriously.
Both times, though, the answer that surfaced in the room wasn't a capability answer. It was a phrase: AI Native. And both times it showed up right after somebody had seen a vendor demo a platform with that phrase already printed on the slide. I don't think that's a coincidence, even at a sample size of two. Nobody sits down cold and decides a decades-old company should aspire to be native to something that didn't exist for most of its history. Somebody saw the word first, and it stuck, because it sounds like ambition.
My gut says it's the wrong target. Not the wrong effort — the wrong target.
"AI Native" isn't a new coinage. It's borrowed, the same way "cloud native" was borrowed before it — and "cloud native" already ran this exact argument once. A system that gets lifted out of a data center and shifted into a VM in AWS is, technically, running in the cloud. Nobody in infrastructure calls it cloud native, because the term was never about where something runs. It's about what was assumed on day one — elastic scaling, distributed failure, statelessness — baked into the architecture before the architecture had ever known anything else. A lift-and-shift carries the old assumptions with it forever, no matter how long it's been in its new home.
"AI Native" is asking the same question about a company instead of a system: did it grow up assuming AI, or did it have to make room for it later? For a company that's been operating since well before AI was a business consideration, it had to make room. That's not a flaw to fix. It's just true, the way when a company started is just true, and no amount of near-term effort changes the start date.
Here's what worries me about letting the phrase stand unchallenged: in both cases, the pressure driving the conversation was a capability question with a real deadline on it. "AI Native" answers an identity question that has no deadline that could ever be met, because the thing it's asking for — having been built this way from the start — was foreclosed the moment the company was founded, long before AI made it a business consideration. If that framing wins the room, the whole organization inherits a strategy that fails on its own terms no matter how well it's executed, because the goalpost was never reachable to begin with. That's a worse outcome than picking the wrong tactic. It's picking a target that can't be hit by construction, and spending real budget finding that out.
This is the same shape of problem Conway's Law points at, aimed at the organization itself instead of its software: a company's communication structure, decision rights, and review chain were designed around a pre-AI world, because that's the only world that existed when those things got designed. You can hand every person in that structure an AI tool tomorrow. The structure itself doesn't become native to anything. It just gets faster at being whatever it already was.
My word for what I think is actually achievable is AI enabled — not because it's a lesser ambition, but because it's an honest one. It doesn't ask a long-established company to have acquired something it never had the chance to grow up with. It asks what a company that's spent decades accumulating real institutional knowledge, real client relationships, and real working systems can integrate deliberately, the way any capability gets integrated into something that already exists and already works. That's the part of this I actually know how to do — map what's there before deciding what moves.
I haven't brought this into a room yet. I'm still doing the thing I always do before a conversation like this one: expanding it out to what I'd want it to look like in the best case, then contracting it back down to what's actually gettable on a near-term timeline, and seeing how much daylight is left between the two. Right now there's real daylight. I don't have the ending to this yet — just the argument I'm walking in with, and a growing suspicion that the most useful thing I can say in that room is that the word on the slide was never actually available as a destination, and that's fine, because it was never the point.
Field Mappings