The civil servant's guide to the 5 days of learning, and how to spend them on AI

Five days a year, sitting there unused

The AI Playbook for the UK Government says it in one line: "Every civil servant can take up to 5 days of learning per year." Five days. Already yours. No bid, no business case, no new budget line to defend.

Now ask around your team and see how many people took them last year. The honest answer, most places, is somewhere between nobody and one person, and that person used half a day on a mandatory course somebody else booked for them.

Meanwhile the Cabinet Office pointed its whole 2025 One Big Thing campaign at this. The theme was "AI for All", it went live on 14 October 2025, and the stated aim was for every civil servant to become a confident, responsible user of AI. The number published alongside it explains the urgency: only 28% of civil servants felt confident using AI at work.

So the ask exists, the gap exists, and the time to close it already exists. What's missing is a plan for what to actually do with the five days once you've put them in the diary.

Start with the free stuff. Don't finish there.

The Playbook also says: "We've launched a series of free online learning resources for all civil servants." They sit on Civil Service Learning and they cost you nothing but time. What's in them:

• Fundamentals of AI and generative AI, meaning what these systems can do and, more usefully, what they can't.
• AI ethics.
• The business value of AI.
• An overview of the main generative AI tools and applications.
• A technical curriculum covering LLMs, machine learning, deep learning and NLP, leading to certificates.
• The Government Campus AI prospectus also sorts learning into three tiers, Foundation, Working and Expert, which is a useful way to work out roughly where you are and what to aim at next.

Take them. They're free, they're decent, and they give you a shared vocabulary so you can have a sensible conversation with your digital colleagues. But here's the part nobody says out loud at the all-staff: most people will complete a module, tick the box, feel briefly informed, and change nothing whatsoever about how they work on Monday morning. A course can explain what a large language model is. It cannot tell you whether to trust the paragraph it just wrote about your policy area.

The three skills that survive contact with your actual job

When I train people, the ones who come out genuinely more capable have picked up three things. None of them are facts about AI. All of them are habits.

• Judgement, meaning knowing what to hand to AI and what to keep. Summarising forty pages of consultation responses into themes? Hand it over. Deciding which of those themes matters to the minister? That's yours, and it always will be. Most bad AI use is a judgement failure, not a prompting failure.
• Workflow design, meaning thinking in steps rather than one giant prompt. People type a paragraph, get something mediocre, and conclude AI isn't up to it. The people who get good results break the job into stages: extract, then structure, then draft, then challenge the draft. Same model, completely different output.
• Verification, meaning knowing when an answer is safe to act on. Public sector work has to be defensible. That means knowing which claims to check, where the model tends to be confidently wrong, and what you'd never send without reading it end to end yourself. I've written separately about what responsible AI use actually means if you want the longer version.

You cannot get any of these from a video. You get them by doing your own work with AI, getting a couple of things slightly wrong in a low-stakes setting, and adjusting.

How I'd actually spend the five days

This is the bit you can copy into your objectives. It assumes nothing except access to a tool your department allows you to use, and that you're willing to work on your own real tasks rather than a made-up exercise about a fictional company.

• Day 1, foundations. Do the free Civil Service Learning modules. Fundamentals, ethics, the tools overview. Half a day if you move quickly. Spend the other half writing down five tasks you do regularly that eat your time. Be specific: "turning meeting notes into actions", not "admin".
• Days 2 and 3, applied practice on your real work. Take those five tasks and do them with AI, alongside how you'd normally do them. Compare the two. You're not trying to save time yet; you're finding out where the model helps, where it flatters you with a plausible answer, and where it wastes your morning. Keep a running note of both. That note is the actual output of these two days.
• Day 4, build one workflow. Pick the single task that went best and turn it into something repeatable. Write down the steps, the wording that worked, the source material you paste in, the checks at the end. One good workflow you use every week beats twenty prompts you tried once and forgot. Save it somewhere your team can find it.
• Day 5, verification, then teach it. Spend the morning deliberately trying to catch the model out on your own subject matter. Ask it something you already know the answer to. Ask for sources and check them. Work out your own rule for what you'd never send without reading it yourself. Then spend the afternoon walking one colleague through your workflow. Teaching is where you find out what you actually understand.

One caveat: don't take the five days as a block week. It's tempting, and it doesn't work. Space them out, a day a month if you can, because the learning happens in the gaps, when you go back to your normal work and notice a task you'd now do differently.

If you lead a team

The Playbook is direct about the talent problem. It tells departments to "consider adopting strategies that combine new hires, working with contractors or third parties, and internal upskilling", and to set out an AI sourcing-and-partnership strategy, meaning a clear view of which capabilities you build in-house and which you source. Read that as permission to invest in the people you already have, rather than waiting for a hiring round that may never clear.

Three things make the difference between a team that has done AI training and a team that has changed how it works. Give people the days as genuinely protected time, or they'll get eaten by delivery. Ask for an artefact at the end, not a certificate: a workflow, a checklist, a before-and-after on a real task. And pair people up, because the person who has to explain their approach to someone else is the one who ends up actually owning it.

If you want the department-level version of this, I've written about how teams get hands-on Claude and AI training in 2026, and what we cover in AI and Claude training for public sector teams. Straight answer on where we sit: Fautons is not an approved government supplier and we're not on a framework. I'd rather tell you that up front than have you find out three emails in. Plenty of what's above you can do without us, and you should.