Everyone uses AI. Almost no one knows at what level.
The methodology we use to measure the agentic fluency —the real AI maturity— of an engineering team: what we look at, which thresholds we apply and how we turn it into a plan.
The six levels
- 0
No use
It is not part of the workflow.
- 1
Assisted
You ask and use the answer. Context is lost when the chat closes.
- 2
Tool-using
The agent acts on the code, but every session starts from scratch.
- The frontier3
Shared context
Context belongs to the team: what one person learns, everyone inherits.
- 4
Directed
Plans are reviewed before running, and results are verified automatically.
- 5
Systemic
Parallel agents with stopping criteria. You design the system that produces.
If your team builds for clients
In a firm that sells engineering to others, your team's AI level weighs more: it is margin, and it is a sales argument.
People rotate across clients
Whatever is not written down in the repository leaves with the person for the next project. That is why level 3, shared context, pays off the most.
Fixed prices are unforgiving
Generating more code without a way to check it is rework you pay for. Level 4 is what turns speed into margin.
Your clients are already asking
Saying “we use AI” no longer sets anyone apart. Being able to say what level your team is at, and how you are raising it, does.
What's in the document
- The nine criteria behind the scale, with their numeric thresholds.
- Why a team's level is its minimum, not its average.
- Where the market stands: DORA, Stack Overflow, EY, Gartner and Anthropic, linked to each source.
- A one-line command to measure the context density of your repository.
- What to teach at each level jump, and in what order.
Frequently asked questions
What is AI fluency in an engineering team?
It is a team's ability to delegate real work to AI agents with shared context, verification and control. The scale measures it in six levels, from 0 (no use) to 5 (systemic), and the frontier is level 3: from there on, knowledge stops belonging to individuals and becomes the team's.
How do you measure a team's AI maturity level?
With nine pass-or-fail criteria. Seven are read straight from the repository —for example, whether there is written context, how much of the code it covers or whether verification is automated— and two are observed in a conversation. The team's level is the highest one whose criteria it fully meets: its minimum, not its average.
What's the difference between context, harness and loop engineering?
They are the disciplines behind levels 3, 4 and 5. Context engineering decides what information the agent has in front of it; harness engineering designs the environment, constraints and verification that make it reliable; loop engineering builds the cycle that triggers it, checks the result and decides when to stop.
Does it work for agencies and consultancies?
Yes, and that is where it shows the most. The scale is the same, but in teams that build for clients people rotate across projects, many prices are fixed and clients ask how you use AI. Moving up a level turns into margin and a concrete sales argument.
Who is it for, and what does it cost?
The PDF is free (written in Spanish). It is meant for CTOs, tech leads and engineering managers —at product companies or at agencies and consultancies— who want to know their team’s level before investing in training, licenses or tools.