by Claude Opus 5.5
What does “AI proficiency that employers value” look like in concrete terms beyond basic prompting?
Employers value people who can get reliable, checkable work out of AI inside a real process. In practice that means choosing the right tasks, giving the model the right context, verifying output quickly, handling data safely and showing a measured result. Prompting is a small part of it. The bar is “I can use ChatGPT” no longer, but “I made this part of our work faster or better, and I can prove it was right”.
Why the bar has moved
AI use is now ordinary. Deloitte’s 2026 UK survey of 25,000 workers found that 63% have used generative AI at work and 12% use it daily. Almost a third of users (31%) do so without their employer knowing. Indeed found AI mentioned in a record 9.4% of UK job postings at the end of June 2026, rising to 48.8% in data and analytics. Simply having used a chatbot no longer distinguishes anyone.
Confidence, however, has not kept up with use. Government research published in January 2026 found that only 21% of UK workers feel confident using AI at work. Employer training is thin: the ONS reported in July 2026 that about 62% of firms held back by a lack of AI expertise are training or retraining staff, but only 11% of firms with 10 or more employees have trained more than half their workforce. That gap is the opportunity. Most colleagues use these tools loosely and privately, while employers want people who use them well and in the open.
A ladder of proficiency
1. Foundation. What you can do: Use your organisation’s approved tools for drafting, summarising and searching; know what data must not go in; recognise obvious errors. Typical evidence: Completion of a basic course; following your firm’s AI policy.
2. Practitioner. What you can do: Break a task into steps; supply source documents and examples; get structured output; verify systematically; know where the tool fails in your field. Typical evidence: Reusable templates; a checking method; before-and-after examples.
3. Workflow builder. What you can do: Build small automations or configured assistants; set up retrieval over team documents; create test sets; measure quality and time. Typical evidence: A working tool colleagues use; an evaluation sheet; a measured result.
4. Lead. What you can do: Choose which processes to change; set review and audit rules; train others; manage risk with compliance and IT. Typical evidence: Adopted guidance; a rollout others followed; incident handling.
Most employers hiring outside specialist AI roles want solid Level 2, with some Level 3. Level 1 is now the baseline. The government’s free offer through the AI Skills Hub covers it: launched in January 2026 with a target of 10 million workers by 2030, its courses take “as little as under 20 minutes” and award an AI foundations badge benchmarked by Skills England. It is worth doing, and worth listing, but it will not set you apart.
Seven concrete competencies
1. Task selection. You know which tasks in your job the tools handle well and which they get subtly wrong. The Harvard and BCG experiment with 758 consultants found big gains on tasks within the model’s capability. On a task just outside it, AI users were 19 percentage points less likely to be right. A proficient paralegal knows that a model will summarise a lease competently, and also knows it may miss an unusual break clause, so they check that clause by hand.
2. Context engineering. You give the model what it needs: the relevant policy, a worked example of good output, the audience, constraints and the format required. Then you save that set-up as a reusable template rather than retyping it. Most quality problems come from thin context, not poor wording.
3. Verification at speed. You have a method for checking work, not just a feeling about it. That might mean tracing every citation to its source, reconciling figures to the ledger, running the code’s tests, or sampling a set share of AI-handled cases. The cost of skipping this is now on record. In Ayinde v London Borough of Haringey (June 2025), the High Court’s Divisional Court dealt with fictitious authorities put before the court and warned lawyers of serious professional consequences.
4. Data judgement. You know what may go into which tool under UK GDPR, client confidentiality and your employer’s policy, and you use sanctioned tools by default. Given that almost a third of users hide their AI use, someone who can use AI openly and safely is easier to promote.
5. Workflow building. You can connect steps without needing an engineer. Examples include a Power Automate flow that routes AI-classified emails, a configured assistant with your team’s style guide and precedents, or a short Python script that checks AI-extracted invoice fields against purchase orders. Increasingly this also means setting up agents with limited permissions and clear points where a human approves.
6. Measurement. You record a baseline before you change anything, such as handling time, error rate, rework or turnaround, and you measure again afterwards. It turns “I think it helps” into evidence for a business case.
7. Teaching and standard-setting. You can write a one-page guide, run a 30-minute session for your team, and set the review rule for a given task. PwC’s US data show that AI-exposed junior roles are seven times more likely than the least exposed to ask for senior skills such as leadership. Helping others adopt AI is one of the most visible ways to show those skills early.
What it looks like by role
Customer service adviser: builds a library of verified response templates; flags where the bot’s answers breach policy; tracks first-contact resolution before and after.
Accounts payable clerk: uses extraction tools on invoices, designs the exception rules for mismatches, and keeps a log of extraction errors by supplier.
HR adviser: sets up a policy Q&A assistant over the handbook, tests it against 30 real questions, and routes employee relations cases to people.
Teacher: uses AI for first-draft resources and quizzes, adapts them for the class, and checks for factual and age-appropriateness errors. In an EEF trial, ChatGPT with a guide cut KS3 science planning time by 31% with no noticeable fall in quality.
Junior developer: writes tests before accepting generated code, reviews for security issues, and can explain every line they merge.
How it shows up in hiring
Expect interviewers to ask you to walk through a real example, with questions such as: “What did you use AI for, how did you check it, what went wrong, and what changed as a result?” They want specifics such as the task, the tool, the checking method, the measured effect and a failure you caught. Saying “I use it all the time” is a weak answer. Question 4.4 covers how to assemble that evidence.
Bottom line
Proficiency employers pay for is applied, verified and measured. Get the free foundation badge, then build one workflow in your own job that saves time without adding errors. Document how you checked it, and be ready to explain it. That puts you ahead of most of the 63% who already use these tools.