What I mean by an AI-native company

Someone asked me recently what I mean by the term “AI-native company”.

Picture this: you have been sent a document, obviously written by a neural network, some 40 pages long. Hardcore AI slop, in that infuriating style too. And your boss asks: “what do you think?” The answer depends on your mood and on your own value on the job market = tell them where to go/quit, apologise/no time, look important/make them rewrite it...

Or better, an example from my recent arguments on social media. You work in the IT department of a bank. A colleague from the business side comes over and says: “look what a great tool I vibe-coded! where do I send it? let’s roll it out!” And straight away the corporate bureaucracy kicks in: where is the technical documentation, how will we support it, “how do we do backups?” (c)

It seems to me the main trouble in situations like these = asymmetry in tooling. Instead of “what do you think”, it would be great to feed the document to a decent model and ask for a point-by-point summary. Compare it with whatever methodologies/standards already exist. Ask your own model about the problem spots that are well known from the height of professional experience: are they covered in the proposed document?

Same with development the business brings in. It is a blessing that the business itself knows what it needs! How much effort I once spent on all sorts of clickable prototypes, just to be sure the mobile app would not have to be laid out again. Even if you make them sign the spec in blood, something important still slips through until you look at a half-finished result.

But yes, if someone has dumped fresh “vibe code” on you and now you have to work out by hand, on your own, what to back up and how, that is a problem, of course. Shifting it onto the business makes no sense either: it will just vibe-code you the answers about “backups” as well. But if you have, not even architecture standards, just a plain list of the previous handmade wonder-services, any experienced model will stop proposing MySQL where PostgreSQL is already the de facto standard.

For me, an AI-native company = no significant asymmetry between employees in their tools and in how well they command them.

As long as one person is ready to pay for a top model out of their own salary while the others stopped at a half-hour course on “all the secrets of AI in MS Copilot”, AI will stay a migraine for the company’s internal climate and its competitiveness.

And how exactly people work together in an “AI-native organisation”, whether you have some powerful internal MCP gateway, sync markdown files through git or simply copy the answers of smart models into email the old-fashioned way = does not matter.

p.s. Slava sent me blog drafts in Montenegrin today, and I thought: thank god this is not mine to do personally. Montenegrin it is :-)