Ask ten executives whether their company is AI-native and many will say yes, or at least “getting there.” Often, they’re referring to their license count: Everyone uses a chatbot, and a handful of teams are piloting agents. But AI usage alone is not sufficient for a company to be considered AI-native. Buying tools is just the first step. Changing how work gets done, who does it, and how the company plans for it is where things stall. In our work with clients, that change plays out in four stages: individuals, functions, operations, and then the organization as a whole. Each stage pays off on its own. Each one also has a trap that keeps companies from reaching the next, and the traps are easier to spot once you know which stage you’re in.
AI-Enabled Individuals
Many companies are currently in this category. Each person uses AI tools at their own pace, for their own purposes. As such, the benefit of AI adoption is limited to the individual level, as opposed to the team or organization at large. Employees get better at their own responsibilities, and the company hopes that this results in cumulative gain, but progress is uneven and hard to measure. One analyst may use AI to get through compliance box-checking, while their coworker could have ten times the output. Managers usually can’t meaningfully differentiate between the two, so when that high-output employee moves teams or leaves, their prompts and judgments about AI go with them, and the company is back where it started.
AI-Enabled Functions
The second stage is when a company points AI at a whole function instead of a person: support tickets get answered faster; finance turns out more analysis in the same week. There are two overarching types of traps that show up at this stage. The first is investing in a bottleneck: Engineers adopt coding agents and velocity jumps, and then code review, QA, security, and release management can’t keep up with the output. Finance doubles its analysis volume, but the review cycle hasn’t changed, so the backlog grows. The result is that the constraint is moved to a different part of the workflow instead of being eliminated. The second trap is more insidious. If the goal is “make this function faster at what it does today”—which it often is—AI amplifies everything, including any inefficiencies. A broken intake process with AI on top is a broken intake process that produces more brokenness. An opportunity for improvement is to ask whether the function should work the way it does at all, but many companies fall into the trap of skipping that question—speed is easier to measure than redesign. So functions “improve,” but the organization doesn’t, because the AI implementation either lacks the proper infrastructure to improve productivity in a consequential way or it exacerbates the issues of a flawed system.
AI Operations
At the third stage, AI stops speeding up existing workflows and starts replacing them. The question changes from “how do we do this faster?” to “given what AI can do, what is this work now?” Moderna is a clear public example. The company has used AI across the business since 2023, and employees have built hundreds of custom GPTs for tasks like dose selection and contract review. In 2025, they merged HR and IT into a single function under one leader, on the reasoning that designing jobs and designing the systems people work in had become the same problem. The org chart needed to evolve because the work changed. Individual roles also undergo big changes at this stage, even if the titles remain the same. Take product strategy: The traditional loop is gather, review, discuss, repeat, and most of a typical product leader’s working hours go to the first two. When research synthesis and first-draft specs are largely automated, the job shifts toward decision quality and how fast they can run that loop. Knowledge also starts to compound at this stage. Agents, prompts, and context built in one function get reused in another. Someone in ops might build a workflow that an employee in finance adapts. For the first time, the company, and not only its people, is getting better at using AI.
AI-Native Organizations
At the fourth stage, AI is assumed. Budgets and planning cycles account for AI capacity up front, as infrastructure, rather than as a line item added after the headcount plan is done. And the workforce looks different: fewer people completing tasks, more people building and managing the agents that complete them. McKinsey’s most recent State of AI survey, published in August 2026, gives a sense of how rare this stage still is. Of the 1,719 organizations surveyed, about 6 percent qualified as AI high performers (companies that attribute at least 5 percent of their earnings to AI). Nearly three-quarters of these businesses have fundamentally redesigned workflows because of AI, compared with about a quarter of everyone else. They are more than twice as likely to put over 15 percent of their technology budget toward AI, more than three times as likely to be scaling agents across most business functions, and twice as likely to say senior leaders are visibly committed to the effort. Anthropic’s enterprise transformation guide describes the same end state: AI built into how the business plans and runs rather than added on top of it. The companies closest to that tend to be small and young, and they were built that way from the start.
The So What
A growing number of companies will tell you they’re AI native, but they’re most likely at stage one, with pockets of stage two. The problem lies in conflating a license rollout with organizational transformation and then wondering why the results never show up at the company level. If you want your organization to move to a higher stage, the sequence matters. Start by making individual usage visible, so the people doing ten times the work become patterns the company can copy instead of exceptions it loses. Pick a function and redesign it rather than accelerating it, and fix the bottleneck you create downstream before you declare victory. Build the shared infrastructure (agents, context, skills) so what one team learns reaches the next. Then put AI in the budget and the org design the way you would any other infrastructure. This rarely happens in one planning cycle, and few companies get through all four stages on their own. Blank Metal works with enterprises at each one of them: figuring out where adoption has stalled, redesigning the function or workflow that’s holding things back, and building the shared agents and context that let the rest of the organization move together. If that sounds like where you are, we’d be glad to talk.




