TL;DR
The short version
"AI-native" isn't a company that bolts agents onto old workflows. It's one that rebuilds how the work runs — around context, cost-per-task, and self-correcting loops — and still names a human owner for every outcome.
On The AI Daily Brief, Nathaniel Whittemore walked through Alex Lieberman's list of 30 features of an AI-native company. Strip out the specifics and a handful of disciplines keep repeating — the ones a leader can act on.
Built on The AI Daily Brief with host Nathaniel Whittemore, riffing on Alex Lieberman's "30 features of an AI-native company."
Stop making agents work the way people did
The instinct when you map a process is to hand the map to an agent and say "do this, but faster." Whittemore's sharpest disagreement with the list lands here.
Mapping is worth doing — a lot of how work actually happens lives locked in people's heads, passed around in hallway conversations and Slack side-chats that never reach an operating manual. Agents can't use what was never written down.
But the map is context, not a cage. Artificially constraining agents to the pattern of an old workflow is, in many cases, the wrong approach. The better move is to give an agent the goal and the guardrails, then let it find the path. The best way to do something next year probably won't look like an efficient version of how you did it last year.
Context is the load-bearing discipline
If one thing separates AI-native orgs, it's how well they organize the context their agents need. The list calls for a single intelligence layer — structured and unstructured data, documents, and business logic in one queryable place. Whittemore prefers a mesh or lattice for larger organizations: sources of truth that interface with each other and that agents can traverse, rather than one master file everything has to reconcile against.
It gets concrete fast. Progressive disclosure means structuring knowledge so an agent loads only the slice it needs — put the metadata an agent checks at the top of a file so it doesn't burn its context window, and its budget, on things it won't use. That same instinct drives the cost features: route tasks to the cheapest model that can do them, split heavy planning (expensive models) from execution (fast, cheap ones), and track cost-per-accepted-pull-request as a real metric.
Prompting is out; loops are in
The clearest operating change is moving from prompting to loops: instead of prompting agents, you give them a goal and bumpers around what they can do, then design a process they can run over and over until they hit it.
A loop only works with a success metric that's objective, not a matter of taste. "Achieve X% on this test" can drive a loop; "make the interface look good" can't. That's why evals stop being a QA afterthought and become core infrastructure — without something to check against, the loop has nothing to converge on.
Finance is the concrete case. OpenAI CFO Sarah Friar has described pushing her team toward a zero-day close and continuously updated forecasting — accounting that runs as a live process instead of a monthly scramble.
Everyone builds — but someone still owns the result
The line between technical and non-technical staff is blurring. This doesn't mean marketing replaces engineering. It means non-technical people can now build things to do their own jobs, and contribute in a format agents can turn into implementation plans. A citizen-developer SDLC keeps that work contiguous with engineering — governance, access, and versioning built in, not bolted on afterward.
Autonomy is earned, not granted. Agents climb a ladder — observe, suggest, act with approval, act alone — until they can run a whole workflow inside a defined boundary. Governance is the enabler here, not the brake: the strongest AI-native orgs treat it as a transformation partner, with legal, HR, and IT designing the policies that make new work possible.
The feature the list misses came from a viewer.
Every AI workflow should have a clear owner, measurable goal, and someone responsible when things go wrong.
Benti Jal, via The AI Daily Brief
The best AI-native companies won't just ask whether AI can do the work. They'll ask who owns the result. That's the new management discipline — and it lands on everyone, because everyone is becoming a manager of agents alongside doing their own work.
