How good are you really at AI? The A1 to C2 scale, applied to work

“Proficient in AI” on a CV today is worth roughly what “proficient in Microsoft Office” was worth fifteen years ago: everyone writes it, and nobody knows what it covers. Between someone who asks a chatbot the odd question and someone who has cut four hours a week out of their job with a chain of automations, the gap is the one between a tourist with a phrasebook and a fluent speaker.

Languages solved this problem long ago with the Common European Framework of Reference — the six levels from A1 to C2 that language schools and employers use well beyond Europe. The same grid maps surprisingly well onto AI, as long as you understand what actually moves you from one level to the next. It isn’t the number of tools you’ve tried, the hours you’ve put in, or knowing which model came out last week.

The six levels, from search-engine habits to system design

A1 — Beginner: AI as a search engine

You ask a short question, read the answer and use it as it is. The exchange ends after one turn. You don’t follow up, because it doesn’t occur to you that you can.

The ceiling at this level: you get generic answers and often conclude that “it’s not much use for my job”. The next step fits in one sentence — give it context and tell it what format you want back.

A2 — Elementary: producing usable text

You generate emails, summaries, translations and rewrites. You’ve worked out that the quality of the output depends on your instructions, and you adjust when the result disappoints.

The ceiling: you start every request from scratch. The time saved is real but one-off, never cumulative. The next step is to structure your requests instead of improvising them.

B1 — Intermediate: structured instructions

Your requests now include a role, some context, a format constraint and sometimes an example or two of what you expect. You know that one good example is worth three paragraphs of explanation. You keep your best prompts so you can reuse them.

This is the level where the gains start to show up across a working week. The ceiling: each task still lives on its own, in a chat window, disconnected from the rest of your tools.

B2 — Upper intermediate: chaining and combining

You break a complex task into successive steps, each one building on the output of the last. You combine several tools — one for research, another for drafting, a third for formatting. You have tried-and-tested templates you reach for without thinking.

You also know what you won’t hand to a model. That’s a more reliable marker than anything else: at this level, saying no is a reasoned decision, not a nervous one.

C1 — Advanced: automating and integrating

You no longer kick things off by hand. You connect tools together, plug into an API, and let a process run on data that arrives without you. You design for failure: what happens when the output is wrong and nobody reads it?

The risk changes nature at this level. It’s no longer one bad answer; it’s the same bad answer repeated a thousand times with nobody watching.

C2 — Mastery: designing and teaching

You’re no longer just solving your own tasks: you build ways of working for other people, you train them, and you decide what should and shouldn’t go through a model in an organisation. You can explain why an approach will fail before anyone has tried it.

One overall level isn’t enough: five dimensions to score separately

An average hides the most important part. The most common profile among experienced professionals is lopsided: very comfortable producing content, almost no grasp of what a model actually does — and therefore unable to spot an answer that is wrong but well written. Five dimensions deserve their own score.

• Understanding: do you know, roughly, how a model produces its answer? That’s what lets you predict where it will go wrong — and why it invents references with such confidence.
• Basic use: writing, summarising, rewording, translating. The most widespread dimension, and the one that plateaus fastest.
• Intermediate use: structured instructions, examples, iteration, reusable templates.
• Advanced use: chaining, combining tools, automation, working at volume.
• Ethics and security: what you share, where it goes, what you have to disclose. Usually the weakest dimension — and the only one where a failure has legal consequences.

How to move up a level

Progress doesn’t come from a course; it comes from a change in practice, and the change is different at each level. Three principles hold at every stage.

1. Work in a field you know well. It’s the only place you can spot an error: you can’t improve by checking answers you’re in no position to judge.
2. Always look for the mistake. Ask the tool to critique its own answer, then check its critique. That back-and-forth builds judgement faster than fifty answers accepted at face value.
3. Only automate what you’ve done by hand ten times. Automating a process you don’t fully understand doesn’t remove the work; it mass-produces the error.

Which tools you use matters less than how you practise, but it isn’t neutral: depending on how you work, the same tools will either help you progress or set you up to fail. We go into this in AI tools at work: how to choose the right ones for the way you work, and the framework for working out your own style is in our guide to the 34 talents.

What your level doesn’t tell you

A high level isn’t an end in itself. A C1 whose job contains nothing worth automating has invested a lot for very little. A B1 in a writing-heavy role has probably clawed back several hours a week. The useful question isn’t “what’s my level?” but “what level does this role call for, and where’s the gap?”.

That’s why ethics and security deserve attention first, whatever your overall level: it’s the only dimension where a shortfall can’t be fixed after the fact.