Utilities need two kinds of AI literacy at once. Customer-facing teams in billing, debt and field scheduling must be able to judge an agent's held actions, especially where a vulnerable household is involved. Everyone near network operations must understand why agents read but do not act there. Evidence it with role-based training records tied to approval rights.
Where do agents sit in a utility?
Most utility agents work in the customer and back-office layer: answering billing queries, drafting payment plans, correcting meter reads, booking engineer visits, preparing debt letters. A few sit closer to operations, summarising outage data or drafting work orders. Almost none should act on network control systems, and literacy starts with knowing why.
The reason is the difference between harm you can reverse and harm you cannot. A wrong bill can be reissued. A wrong instruction to a field crew, or a missed flag on a household that depends on powered medical equipment, cannot be taken back. Staff who understand depth, from 1 for reading to 4 for actions that are hard to undo, can see why the same agent drafts a payment plan freely but holds a disconnection-related step for a named person every time.
What does each team need to know?
| Team | Literacy they need | What stays with the person |
|---|---|---|
| Billing and customer service | Reading a held refund or bill correction, checking the meter read's source | Approving credits above limits, any change to a priority services flag |
| Debt and collections | Recognising when the agent holds because of a vulnerability marker | Payment plan terms for vulnerable customers, any step towards restricting supply |
| Field scheduling | Checking a drafted work order against safety rules and site access notes | Dispatching crews to unusual or hazardous jobs |
| Network and control room | Knowing agents stay at depth 1 here and why | Every operational action |
| Managers and owners | Reading the agent's depth and the depth each action needs, certificate status and the record | Setting limits and answering for outcomes |
The specific skill for utility reviewers is reading provenance on meter data. A smart meter read from the utility's own data system is sealed. A customer's photo of their meter is declared. An estimated read is a calculation, not a fact. A reviewer approving a back-bill needs to see which one the agent used. Risk literacy in a utility is largely this: knowing which facts can carry which decisions.
An example: a held debt letter
Take a regional water company whose collections agent drafts arrears letters at depth 2. One account has a declared note from a support worker that the customer has a long-term illness, but the note is not on the formal priority register. The agent holds the letter and names the deciding bound: the customer's own limits route any account with a vulnerability indicator to a person.
A trained reviewer reads the note, checks whether the register should be updated, changes the letter to offer a payment break, and records why. An untrained reviewer sees a standard arrears letter and approves it. Both decisions take two minutes. Only one is defensible, and the record shows which was made.
How do you show the literacy is real?
Three pieces of evidence carry weight with a regulator or auditor:
- A training requirement per role, written into policy, with the reviewer level required for anyone who approves held actions.
- A record per person of what they passed and when it lapses.
- Proof that approval rights follow the record: people with lapsed training cannot approve, and the workspace shows it.
Utilities in the EU that fall within NIS2 as essential or important entities already carry duties on cybersecurity risk management and incident reporting, so staff who handle agents should know how an agent failure or a suspected prompt injection is reported. The EU AI Act's Article 4 asks deployers to ensure a sufficient level of AI literacy among staff; the detail is in EU AI Act Article 4 AI literacy for AI agents in energy and utilities. Managers who set limits need governance literacy as well: who may change a bound, and what evidence a change leaves.
Utilities can borrow from finance. What AI literacy means for banks covers the same split between people who read and people who approve, under heavier supervision. The overall structure is in the guide to AI literacy and certification for people who work with agents.
Where the Academy fits
Pryme Intelligence Academy sets module tests in the learner's own industry and role, so a debt adviser at a utility is tested on held arrears letters, not on bank payments. Foundations gives every team the vocabulary of depth and verdicts. The Operator track, ending in Certified Pryme Operator, is for the people who approve held work daily. A free readiness assessment for teams of up to 25 shows where the gaps are before the first agent goes live.
Questions people ask
Should control room staff take agent training at all?
Yes, at foundation level. They need to know what agents elsewhere in the business can and cannot do, so they recognise an agent-drafted request reaching them and treat it accordingly.
Who owns AI literacy in a utility?
The business owner of each agent owns the training of its reviewers. A central team owns the policy and the records, so the evidence is consistent across billing, debt and field teams.
How often should utility staff refresh?
Whenever the agent's scope or limits change, and at least when their certificate comes up for renewal. A change to vulnerability rules should trigger a short update for every reviewer straight away.