I’ve been thinking about what AI-native means inside companies that already exist — not startups building from scratch, but organizations with histories, accumulated ways of working, and leaders who’ve been in their roles long enough to have seen several cycles of technological change. Part 1 was about individuals. This is the harder half.
There’s a pattern I keep noticing. Most established companies have already pushed AI adoption forward by now, and by most visible measures it’s going well — teams are faster, adoption is up, deliverables come back with traces of AI, boards are satisfied. At the same time, there’s a quiet recognition within the organization that the company itself hasn’t fundamentally changed. It runs faster, produces more, but the underlying shape — the products, the decision architecture, the way the business actually works — looks much like it did eighteen months ago.
This is not something that gets said easily. When all the metrics are green, saying “we’re still the same company” can sound like dismissing progress. But the feeling is there, and it’s specific.
Most of what comes in from the outside tends to fall into two directions. One is to keep extending deployment — bring more workflows into AI, push efficiency further. The other is to build outward — new products, new IP, new lines of business that AI makes newly possible.
Both make sense. Both are aligned with how most organizations currently understand AI: either it makes things faster, or it makes more things possible.
AI is being used to amplify existing systems, not to rewrite them.
Seen this way, it also explains why “transformation” is rarely framed as the core problem. Not because it isn’t important, but because most organizations haven’t quite reached that point yet.
Most companies aren’t resisting transformation. They just haven’t reached the point where it becomes necessary.
At the same time, in the same spaces, AI-native startups are building something structurally different. Not faster versions of the same products, but different ways of operating altogether. The gap isn’t showing up in quarterly metrics yet, but it’s beginning to widen.
Inside established companies, most AI work seems to fall into two categories. Workflow optimization — agents stepping into existing processes, AI assisting within predefined steps. And business expansion — new IP, new product lines, new ventures enabled by AI.
Both are real, necessary work. But both are additive. They extend capability without changing the underlying shape. And I think many leaders can already feel this, even if they haven’t fully put language to it.
This points to something more fundamental: the nature of the leadership role itself. Leadership is trained around judgment — reading situations, weighing trade-offs, knowing when to commit. In most domains, that’s exactly the right capability.
But AI seems to be a domain where judgment, without direct exposure, runs into a ceiling. What AI can actually do, where it changes the texture of work, what an AI-native operating surface even looks like — these things tend to become legible only from inside the work, after sustained contact with the material.
And leadership positions, by design, don’t usually produce that kind of time. Not because of reluctance, but because the role is structured around decision-making. And in this domain, direction seems to require grounding first.
If that’s true, it helps explain how AI initiatives tend to land inside organizations. They often take the form of what feels like presence requirements — AI needs to show up in the work — without being able to specify how it should reshape the work itself. That level of specificity depends on a kind of grounded familiarity that hasn’t had time to form.
So the question isn’t whether to adopt AI — that step is largely done. The question is how to develop a form of understanding that leadership roles haven’t historically been structured to accumulate, and to develop it in a way that can actually inform direction.
More deployment won’t produce it; deployment runs on the existing shape. Expansion without grounding tends to follow visible trends rather than redesign what sits underneath.
What I keep arriving at — not as a definitive answer, but as something that holds structurally — is the need for a different kind of container. A small AI-native team, five to fifteen people, given protected space and one clearly defined surface — a product line, a function, a workflow — with a mandate to rebuild that surface from the ground up, as if AI had been there from the beginning.
Small enough to be contained, real enough to produce something that runs.
The team itself needs to be hybrid. People deeply fluent in AI, often from outside because that fluency takes time to build, working alongside people who understand the business from the inside. The hands-on work happens there, not at the leadership layer — so leadership doesn’t have to become something else.
Two things come out of it. One is a new operating shape running on a defined surface — bounded, selective, but real. The other, and more importantly, is what I think of as grounded intelligence. The team translates what they learn back into the organization — what AI actually does well in this business, where the leverage sits, which parts of the operating shape can move.
This kind of artifact has been missing. Not necessarily because it’s difficult, but because nothing has been set up to produce it.
It also functions as proof. People rarely make large commitments based on abstraction alone. They need to see something working within their own context. A sandbox creates that, on a limited surface, with a contained level of risk.
If a company’s AI work today is mostly focused on deployment and expansion, that’s a very common shape at this stage. But structurally, there is a missing piece — and it’s one I don’t hear named often — which is the sandbox.
This is the direction I’m working in at Actual Fiction. Not a transformation program, not deployment support — more like a bridge. AI-fluent operating expertise alongside deep business knowledge, on one defined surface, trying to produce both a new working shape and the grounded intelligence the leadership role has been missing. I’m still early in this work, and still learning what it takes.
About the author
Qinqin (Stella) Yang is a creative director and strategist working at the intersection of innovation, technology, and futures thinking. Over the past decade she has built brand experiences, product design and speculative projects for Tencent, IKEA SPACE10, Burberry, and the Palace Museum etc. More recently she has been building AI-native systems through her practice, Actual Fiction, to help creative teams move from ambiguity to clarity.
This newsletter is where she thinks out loud about creativity, innovation and what it means to build towards a future that doesn’t exist yet.
For more details: qinqinyang.com