AI upskilling: how to build a programme for your team

What is upskilling in AI?

AI upskilling is building the AI skills people need in the role they already have. A recruiter learns to draft a job advert from a role brief and check it for invented requirements. A finance analyst learns to ask for a variance summary that stays within the data. The job stays the same; the way some of its tasks get done changes.

AI reskilling is different. It prepares someone for a different role, often because their current work is changing a great deal. Reskilling usually takes longer, involves new domain knowledge and is a workforce planning decision as much as a training one. Most small and mid-size teams asking about AI need upskilling first.

What is the difference between training and upskilling?

Training is the activity; upskilling is the change in what people can do at work. A team can complete a lot of training and not be upskilled, if nobody applies it or checks the result. The difference matters because it changes what you measure.

How do you find the AI skills gap on your team?

Compare the tasks each role does with the skills those tasks need, then look at actual work rather than asking people how confident they feel. A self-rating is still worth collecting, because it shows who is anxious and who is overconfident, but it is not evidence of skill.

Take a fictional 22-person digital agency, Northfold Digital. The founder lists three recurring tasks per role: account managers write client updates and meeting summaries, designers write briefs and image prompts, and the two developers write documentation. She asks each person to complete one short task with the approved AI tool, using a fictional client. The results show a pattern. Most account managers produce fluent updates that contain at least one claim the brief did not support. Designers write strong prompts but rarely check generated alt text. The developers are already careful.

That is the gap, and it is specific: checking claims for account managers, reviewing generated descriptions for designers, and almost nothing for developers. A one-size course would have spent the developers' time on basics and missed the account managers' real problem. The guide to assessing AI skills explains how to design such a task fairly.

What does a practical AI upskilling programme look like?

A workable programme for a small team has seven steps. Each step is small, and none needs a dedicated learning team.

What does one upskilling cycle look like?

Here is one cycle for a sales rep, Priya, at a fictional 40-person software company, Ledgerline. Her task is a follow-up email after a discovery call. The supplied notes say the prospect has 12 finance staff, uses spreadsheets for month-end close, wants a demo for their controller and asked about pricing for annual billing.

Her first request is: "Write a follow-up email after a good sales call." The draft thanks the prospect, says Ledgerline "will cut their close time in half" and offers a "special discount if they sign this month". Neither appears in the notes. Against the rubric, she scores well on tone but fails evidence: the email makes a performance claim and an offer nobody approved.

The feedback is specific: supply the notes, say what the email must and must not contain, and remove any claim you cannot support. Her second request reads: "Draft a follow-up to a finance lead using only these call notes: [notes]. Confirm the controller demo and propose two times. Say pricing for annual billing will follow from me. No performance claims, no discounts, under 150 words." The draft is accurate. She edits one sentence that sounded pushy and explains why.

The next session gives her a different prospect with different notes. If she briefs it well without being prompted, the skill is transferring. If not, the path repeats the explanation with a simpler example. That loop of explanation, attempt, rubric feedback and a new example is the core of upskilling, whichever tool or vendor delivers it.

What is the best way to upskill employees in AI?

The most reliable approach is short, regular practice on tasks from the person's own role, reviewed against criteria they can see. That is the same principle that works for most workplace skills. Watching content first can help, but behaviour changes when someone has to produce work and receive specific feedback on it.

Managers matter more than the content. When a manager shows how they check an AI draft, asks to see the edit rather than just the result, and protects practice time, people practise. When AI use is either ignored or treated with suspicion, people either avoid the tool or use it quietly without checking. Keep the review about the work, not about monitoring people's chat history.

Which AI upskilling options are available?

Options range from free public courses to dedicated platforms, and each suits a different need.

Why do AI upskilling programmes stall?

Most stall for practical reasons rather than a lack of content. People are not given time to practise, so the course competes with real deadlines and loses. The approved tool is unclear, so careful people wait while others use personal accounts. Examples come from other jobs, so learners cannot see how to apply them. Progress is measured by logins, which rewards activity rather than better work.

Fear is another reason. If people think the programme exists to find who can be replaced, or that their chats will be read, they disengage. Say plainly what the programme is for, what managers will see and what they will not. Tosheo.ai's planned manager dashboard shows skill progress and completion, not raw chat logs, for this reason.

How is Tosheo.ai planned to support AI upskilling?

Tosheo.ai is designed to run the cycle above for every learner without a learning team. All of this is planned scope for launch. A manager is intended to select roles and the tools the team uses, optionally add 1 to 5 company documents for context, and invite people. Each learner takes a 5-minute diagnostic, then gets an adaptive path of 10 to 15 minute lessons, each ending in a hands-on task graded by AI against a visible rubric with specific feedback. Scores can be flagged for review. A weekly digest goes to the manager, with nudges for inactive learners.

Planned pricing starts with a Team plan priced per seat with a five-seat minimum, and a 7-day free trial for up to 5 seats. The pricing page has the detail, and all pricing is a starting hypothesis.

When is upskilling the wrong lever?

If the process itself is broken, such as unclear ownership of client updates, AI practice will only produce broken work faster. Fix the process first. If a role is changing so much that the person needs different domain knowledge, that is reskilling and needs a longer plan than a skills product offers. If the goal is engineering work with AI models, look for technical courses.

If your organisation needs single sign-on, LMS integration or enterprise procurement before buying, those are not part of Tosheo.ai's planned launch scope, and a sales-led provider may be a better fit.

Sources and scope

GOV.UK, Free AI training for all as government and industry programme expands, 28 January 2026 ↗

Skills England on GOV.UK, AI foundation skills for work benchmark, 28 January 2026 ↗

Google, Google AI Essentials course page, checked 16 October 2026 ↗