Anthropic Maps What Makes ‘Fluent’ AI Users Different

When MIT economist Esther Duflo won the Nobel Prize in 2019, she became the youngest person ever—and only the second woman—to receive the award in economics. Her work revolutionized how we think about fighting global poverty through rigorous, evidence-based methods. Today, a different kind of evidence-based revolution is unfolding in the AI space, and Anthropic is trying to measure something equally elusive: what separates skilled AI users from the rest.

“The most common expression of AI fluency is augmentative—treating AI as a thought partner, rather than delegating work entirely.” — Anthropic Research Team

The 4D Fluency Framework

Anthropic’s new AI Fluency Index represents one of the first systematic attempts to quantify how people actually develop competence with AI tools. Working with Professors Rick Dakan and Joseph Feller, the company developed a framework identifying 24 specific behaviors that exemplify safe and effective human-AI collaboration.

Of these, 11 are directly observable when humans interact with Claude. The researchers analyzed 9,830 conversations across a 7-day window in January 2026, measuring the presence or absence of these behaviors to establish a baseline for what AI fluency looks like in practice.

Iteration and refinement emerged as the single strongest predictor of fluency. Users who treated AI responses as starting points rather than final products—asking follow-up questions, pushing back on unclear sections, and refining their requests—exhibited more than double the number of fluency behaviors compared to those who accepted first responses and moved on.

The Polished Output Paradox

One of the study’s more counterintuitive findings concerns what happens when AI produces tangible outputs. When Claude generates artifacts—code, documents, interactive tools—users actually become less critical rather than more.

Conversations involving artifacts showed users were 14.7 percentage points more likely to clarify their goals and 14.5 percentage points more likely to specify formats upfront. But they were also 5.2 percentage points less likely to identify missing context and 3.7 percentage points less likely to check facts.

“As AI models become increasingly capable of producing polished-looking outputs, the ability to critically evaluate those outputs will become more valuable rather than less.” — Anthropic Research

The researchers speculate this pattern might stem from the appearance of completeness. When code compiles or a document looks professionally formatted, users may assume the underlying reasoning is sound—a potentially dangerous assumption when the most complex tasks are precisely where AI struggles most.

Three Paths to Better AI Fluency

Stay in the conversation. The data is unambiguous: iteration correlates with every other fluency behavior. Treat initial responses as drafts, not deliverables. Push back, ask follow-ups, refine what you’re looking for.

Question polished outputs. When AI produces something that looks finished—a working app, a formatted report, clean code—this is precisely when skepticism matters most. The surface quality can mask deeper issues.

Probe for reasoning. Highly fluent users consistently ask models to explain their rationale. They don’t just want the answer; they want to understand how the answer was constructed.

Implications for the Workforce

Anthropic positions this index as a foundation for measuring AI skill development in education and workplace training. As organizations rush to integrate AI tools, the question of who can use them effectively—and how to teach those skills—becomes increasingly urgent.

The research suggests that AI fluency isn’t about technical sophistication or prompt engineering tricks. It’s about maintaining an active, critical engagement with the technology. The users who get the most from AI aren’t those who delegate blindly; they’re the ones who treat it as a collaborative partner in an ongoing dialogue.

As AI capabilities advance, the gap between adoption and fluency may become one of the defining divides in the modern workplace. Anthropic’s framework offers a starting point for understanding—and potentially closing—that gap.

This article was reported by the ArtificialDaily editorial team. For more information, visit Anthropic Research.