Which part of your team's work can AI already do?
Most companies switched AI on before anyone could answer that, or say who on the team can actually run it. You can answer the first question for one process in about a minute.
No account. Describe the process in your own words or start from an example: hiring, onboarding, month-end close, a support request.
What the answer looks like.
Handling inbound customer requests at our demo company: 2,280 hours of work a year. Shares are of those hours, not of the number of steps.
42%
Moves to an agent
- Classify the request and set its priority
- Route it to the team that owns the area
- Search the knowledge base and past requests
49%
Moves to review
- Draft the reply in the company's tone
- Work out with the engineer on duty what's actually broken
9%
Stays with people
- Call the customer who is angry
- Decide on a refund or a contract exception
- Decide whether a request is about personal data
Review is still a person's time: it shrinks, it doesn't disappear. The steps that stay are where accountability sits, not where the model is weak.
Then: who on the team can run it?
“42% can move to an agent” is only true if someone can brief that agent, check what it returns and own the result. That is AI literacy in practice. A certificate of attendance doesn't show it; a short practical check does. For a whole team it takes three steps.
- 01
Break down the processes.
Each step gets a verdict in hours a year: an agent, a person with review, or a person alone. That's where the AI budget goes, or doesn't.
- 02
Check who can run each AI step.
Short practical tasks with a passing score for the role, not training hours and not self-report. A step only moves to an agent if someone can brief it, check it and own the result.
- 03
Keep the record.
Who, which role, what was checked, what came out, and when. Refresh it when the tools change. It's also the record the AI Act asks for.
In HRPulsar the first step is Coverage, the second is exams with a passing score per grade, and a person's coverage is read from those results, never from self-report. To see where you stand yourself, take the AI Fluency Scale self-assessment: five questions.
What the EU AI Act asks since July 2026.
The literacy rule went from a result to an effort. The record it asks for is worth keeping anyway: it's the same map as above.
- Measures and a record.
- Since 27 July 2026, Article 4 (as amended by Regulation (EU) 2026/1744) asks every company that provides or uses AI systems to take measures for its staff's AI literacy. An internal record of what you did counts as evidence.
- No required level.
- The law doesn't ask you to guarantee anyone's level any more. National authorities supervise Article 4 from August 2026, so doing nothing isn't an option.
- Stricter for hiring and staff evaluation.
- AI used there is high-risk. From 2 December 2027, the people who oversee it need the competence, training and authority to do so (Article 26).
Start with one process.
Pick the one your team complains about most. You'll have the three lists before your coffee gets cold.
The example is from the HRPulsar demo company. Not legal advice. Sources: Regulation (EU) 2024/1689 (AI Act), Articles 4 and 26; Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L, 24 July 2026; European Commission, AI literacy Q&A. Updated October 2026.