How FAROL measures AI fluency.
Agentic fluency is describing a piece of work, handing it to an AI, following its execution and checking what comes back.
What people say, and what they do.
The diagnostic · self-reported
- A questionnaire answered by leadership and staff
- It shows what people say they do with AI, area by area
- And where the views of those who lead and those who do the work do not match
The Raio-X · observed
- It reads each person's conversations with AI, with their permission
- It shows how they actually work, across six dimensions
- Each score shows where it came from
Counting activity measures the tool, not the person.
Our first instrument counted activity. In July 2026, run on a field partner's test account, it gave a person just starting with AI the numbers of a professional. Those numbers were produced by the AI: an agent (an AI that carries out tasks instead of only answering) calls tools and creates files from any request, even a confused one.
The obvious fix, looking for good-practice words in the requests, reversed the result. The beginner wrote "make a plan" in 41% of requests; the author of the method, who plans when designing the work, in 9%.
Measured on two people, one of them the author of the method. It is an example, not a statistic.
That is why we rebuilt the method: today a reading with written criteria judges how each person directs the work, and each score must point to the passage that supports it.
12international AI maturity and readiness models
Where FAROL started.
The work began inside a software company, as a maturity index. Before writing a single question, we catalogued the models that already existed. Each block is one of them; its height is how many steps the scale has, from a continuous score to nine levels.
Consulting firms, vendors, universities and government.
Three come from consulting firms, three from technology vendors, four from academia and two from government. Tap a block to see the model.
7faithfully reproduced
MIT CISR, Gartner, Deloitte, Cisco, Microsoft, Salesforce and NIST had enough description (levels, dimensions and criteria) to be reproduced before any synthesis. The other five were kept as references.
Seven measure the company. None looks at how each person works with AI.
The models measure the organization, the risk, the technology or the agent in production. The only one that looks at the person, the TRI, measures the stated attitude toward technology. That is why FAROL moved from maturity to fluency: it began reading each person's practice.
The twelve models in a table
FAROL catalog of April 7, 2026 (model matrix), based on each organization's public documents. "Levels" is the number of stages the model itself describes; the NIST AI RMF has four functions instead of levels, and the TRI gives a continuous score. Forrester/IBM was included by mention only, with no text describing the levels.
Fluency, not maturity.
A maturity index describes what level the company is at, against a fixed yardstick. In AI, the yardstick moves too: what was advanced a few months ago becomes basic. That is why FAROL measures, at every reading, where the person and the company stand and where the frontier is.
AI use for work, in Brazil.
In Brazil, 57% of conversations on individual Claude accounts are about work; worldwide, 43%. This is FAROL's calculation from public Anthropic Economic Index data, for the May 2026 window. Anthropic itself lists Brazil among the countries with the highest share of work-related use, in its January 2026 report.
Anthropic Economic Index report, January 2026: anthropic.com/research/anthropic-economic-index-january-2026-report
The limits.
- The Raio-X scale is being calibrated, with a small sample and human judges.
- The score depends on the model doing the reading; we measure that variation.
- We have not yet followed an entire company over several cycles.
- Nothing here is a promise of productivity gains.
Tell us what your company wants to solve.
You will hear back from the person who leads the work. The first conversation is to see whether it makes sense and where to start.
- You write to us on WhatsApp or by email.
- We schedule a conversation with the person who leads the work, to see whether it makes sense and where to start.
- If it does, you receive a written proposal: team, timeline, deliverables and price, before any commitment.