The data
Anthropic's research team studied 9,830 Claude conversations over a week in January 2026, measuring 11 observable behaviors that indicate skilled AI use. The headline findings matter if you're thinking about career positioning:
Iteration is the single strongest signal of AI skill. 85.7% of conversations showed users refining and building on responses rather than accepting the first answer. Those conversations exhibited 2.67 additional fluency behaviors — roughly double the rate of quick, one-and-done exchanges. Users who iterated were 5.6x more likely to question AI reasoning and 4x more likely to catch missing context.
Polished outputs make people stop thinking. When AI produced artifacts — code, documents, interactive tools — users became more directive upfront (clarifying goals +14.7pp, specifying format +14.5pp) but less evaluative of the results. They were less likely to identify missing context (-5.2pp), check facts (-3.7pp), or question the model's reasoning (-3.1pp). The better the output looked, the less scrutiny it received.
Almost nobody sets the terms. Only 30% of users explicitly told AI how to collaborate — things like "push back if my assumptions are wrong" or "tell me what you're uncertain about." The other 70% just accepted whatever dynamic emerged.
Why this matters for your career
If you're in the job market or trying to make yourself more resilient, AI fluency is rapidly becoming a differentiator — not "can you use ChatGPT" but "can you use it in a way that actually makes your work better rather than just faster."
The iteration gap is a hiring signal. Managers are starting to notice the difference between someone who pastes a prompt and ships whatever comes back, and someone who treats AI as a thought partner — pushing, refining, catching what's wrong. The Anthropic data shows that the second group is dramatically more skilled across every measured dimension. That's the kind of working style that compounds over time.
The polished-output trap is a real risk. If your workflow involves AI generating reports, analyses, code, or client-facing documents, the data suggests you're statistically less likely to catch errors in exactly the moments that matter most. For knowledge workers, this is where reputational risk lives. The fix isn't to stop using AI for artifacts — it's to build a habit of evaluating outputs with the same rigor you'd apply to a junior colleague's first draft.
Setting collaboration terms is an underused lever. Telling AI upfront how to work with you — "challenge my reasoning," "flag assumptions," "tell me what's missing" — changes the quality of what you get back. Only 30% of people do this. If you're in the 30%, you're working with a fundamentally different tool than the other 70%.
The bottom line
AI fluency isn't about adoption — everyone's adopting. It's about whether you're developing the judgment to use it well. The people who iterate, question, and direct are building a compounding skill advantage. The people who delegate and accept are building a dependency.
The question for your career isn't "am I using AI?" It's "am I getting better at using AI, or am I just getting faster at not thinking?"
That's a different kind of leverage — and it's one Career Ladder is designed to help you see. Which parts of your work reward the kind of judgment that AI fluency develops, and which parts are heading toward full automation regardless of how well you prompt.