How to Build AI Skills Across Your Team

4 November 2026 · 6 min read

AI upskilling can improve individual skills while leaving a team dependent on one or two people. Completing training does not automatically share the knowledge needed to maintain and adapt the team’s AI-enabled workflows.

This article covers five actions: map who holds the capability, make knowledge transfer part of everyday work, document design decisions, define AI responsibilities within roles and test whether the team could keep working if one person left. It also explains why course completions alone cannot show whether that capability has spread.

Map Where AI Capability Sits

Connect People’s Skills to Specific Workflows

Identify who can design prompts, run structured evaluations or build and supervise agents, and which workflows depend on those skills. A department-level assessment can miss how much rests with a few individuals.

Ask each person which AI-assisted tasks they perform regularly, which tools they use and whether they designed or inherited the workflow. Map their answers to the workflows they support so you can see where capability is shared, concentrated or missing.

Use the Map to Prioritise Knowledge Transfer

In a ten-person team, three people might produce almost all the AI-enabled output. The team may appear capable overall, but the map should show whether anyone else could take over those workflows. Where they cannot, you have identified both a dependency and a place to begin knowledge transfer.

Make Knowledge Transfer Part of Everyday Work

Pair Colleagues on Real Tasks

Pair a colleague with the person who built the workflow and give them regular time to run it together. The builder should explain decisions and trade-offs as they work. Continue until the colleague can run the workflow independently.

Screen-sharing sessions can bring that learning to the wider team, with one person working through a task and explaining their reasoning. These sessions can be organised internally, but they need protected time and a regular schedule.

Keep Shared Prompts and Working Notes

Store prompts, system instructions and evaluation criteria in a shared library with clear names and version history. Review the entries regularly so colleagues can find and use the current versions rather than rely on someone else’s chat history.

After each working session, record what was tested, what worked, what failed and why. A short summary lets others build on the findings without repeating the same work.

Keep Design Knowledge With the Team

Record the Decisions Behind the Workflow

Document why a model was chosen, where human review is needed and which failure cases were tested. Someone taking over needs the reasoning behind those choices to maintain and adapt the workflow.

For each workflow, keep a short record of its business purpose, key design decisions, known limitations and when a person must intervene or escalate. AWS’s guidance is that prompts and agent configurations should be version-controlled with the same discipline as code.

Make Documentation Part of Delivery

The people making design decisions should record their reasoning as they work, rather than leave a knowledge manager or technical lead to reconstruct it later. Include documentation in the completion criteria for each workflow and review the two together.

Define AI Responsibilities Within Roles

Set Expectations Across the Team

An AI specialist can fill an immediate skills gap, but passing every AI task to that person keeps capability concentrated. Define what other team members are expected to handle themselves.

An operations analyst’s role should specify which AI-assisted tasks they run, how independently and to what quality standard. A team lead’s role should include oversight of those workflows as part of the job.

Be Specific About the Work

“Uses AI to support reporting” gives little guidance. “Designs and maintains prompt templates for weekly performance summaries, checks output quality and escalates anomalies to the team lead” describes work that can be taught, practised and assessed.

Include responsibilities for protecting data, improving processes and deciding when AI should not be used, where relevant to the role.

Test for Key-Person AI Dependencies

Check Who Could Take Over

For each AI-enabled workflow, ask: if its builder or main operator left tomorrow, who could take over and how long would they need?

Use the capability map to rate each workflow’s dependence on that person. Check whether the documentation gives a colleague enough context to operate and maintain it, and where they would still need support.

Address the Most Disruptive Gaps First

Prioritise workflows that run frequently, have a high operational impact or produce time-sensitive outputs. Use pairing, documentation and cross-training to prepare a second person to run them independently. Schedule lower-risk dependencies for the next planning cycle.

Repeat the check after significant team changes and regularly as new workflows are introduced.

Measure Whether AI Capability Has Spread

What Course Completion Shows

A high completion rate does not establish whether people can apply their learning independently. A team may finish the training while still depending on two people for most of its AI-enabled work. Look at who regularly uses AI, what they produce and how much support they need.

What to Track Instead

Track three indicators across the team:

• Independent capability: how many people can run each key workflow without the original builder’s support.
• Design documentation: how many AI-enabled workflows have a record of their design decisions and reasoning.
• Role expectations: what proportion of roles include specific, measurable AI responsibilities.

Independent workflow performance shows whether capability has spread. Documentation and role expectations support that transfer, but are not proof that someone can do the work.

Build Capability the Team Can Sustain

Start with a critical workflow that depends on one person. Give a colleague time to understand the design decisions, practise maintaining it and demonstrate that they can work independently. Keep sharing that knowledge as the workflow changes.

Envisago’s AIVOM™ Reading shows where your operation stands across Value, Design, Capability and Performance, including where capability sits with too few people, and what to prioritise next.

Start your free reading at aivom.envisago.com.