From Curious to Capable: A Three-Level Roadmap for Upskilling Your Team in AI.

What a three-level AI upskilling roadmap looks like.

A team becomes capable with AI the way it becomes capable with anything else: in stages, each one built on the last. The roadmap we use has three levels. Level 1 builds a shared foundation across the whole company. Level 2 turns that foundation into workflows inside each function. Level 3 moves a team to agentic work, integration and measurement.

Thinking in levels tells you two things at once: where your people actually are, and what the next honest step is. A course catalogue cannot do that. A maturity model can.

It also prevents the most common planning mistake, which is buying training for the level you wish you were at rather than the one you are on.

Level 1: the shared baseline everyone works from.

Level 1 is about vocabulary and safety. What these tools are, what they are not, where they fail, and how to use them without putting company information somewhere it should never go.

Companies often want to skip this because some people are already using AI. That is usually the reason not to skip it. Scattered self-teaching produces scattered standards: one person pasting client data into a public tool, another convinced the whole thing is a toy, and no way for a manager to tell the difference.

What a foundation level gives you is a floor. Everyone shares a language, everyone knows the limits, and everyone leaves with a few prompts that work for their own job. Level 1: Foundations runs as a single day for exactly this reason. The floor should be quick to lay and identical for everybody standing on it.

Level 2: applied workflows inside each function.

Level 2 asks a different question. Not can you use this tool, but what does your job look like with this tool inside it. Marketing, finance, operations, sales and HR all answer that differently, so the training splits by function.

The output is documented workflows: a named task, the steps, the prompt or template, and the check that catches errors before the work leaves the team. A function that finishes with [STAT - verify before publish] written workflows has something to build on. A function that finishes with enthusiasm has not.

Quality control matters more here than speed. A team producing fast, unreviewed work has not improved. It has moved the problem downstream to whoever reads the output next.

This is also the level where capability stops depending on individuals. A written workflow is inherited by the next person to join. A workflow that lives in one enthusiast's head leaves when they do.

Level 3: agentic work, integration and measurement.

Level 3 is for teams ready to run processes with AI rather than assist tasks with it. That means delegation with guardrails, human review at the points where a mistake would be expensive, and integration into the systems the work already lives in.

It also means measurement. At this level, leadership stops asking whether people enjoyed the training and starts asking what changed in the process: cycle time, error rate, volume handled, hours returned to the week. Level 3: Advanced closes with a four-week project on one real process for that reason. The proof is a working process, not a certificate.

Not every team needs Level 3. Some functions are genuinely finished at Level 2, and saying so is part of an honest roadmap.

How to sequence the levels across functions.

Sequencing is where most roadmaps come apart. The instinct is to run the foundation for the whole company at once, then stall, because nobody agreed what happens next.

We prefer a different order. Give everyone Level 1, because a shared floor is inexpensive and it removes the shadow-use problem quickly. Then pick one or two functions for Level 2, chosen where repetitive work is heaviest and where the manager actually wants it. Let those functions finish and show their workflows before you open the applied level to everyone else.

Level 3 goes to the function already producing work that other teams copy. Depth in one place travels further than a thin layer everywhere, and internal examples persuade sceptics far better than an external trainer can.

Leadership sits slightly outside the sequence. Executives need the foundation concepts and the advanced questions at roughly the same time, usually in a shorter, private format rather than a company course.

How to measure progression between levels.

Attendance is not progression. Neither is the energy in the room on the day. We look at three things.

• Artefacts. Can the team show prompts, workflows and standards they wrote themselves? Written artefacts are the difference between a team that attended and a team that changed.
• Behaviour on real tasks. Choose a handful of recurring tasks before the training and look at them again afterwards. Is the work being done differently, by more than one person, without being reminded? A fair benchmark is how many of the workflows written during the applied level are still in use a quarter later [STAT - verify before publish].
• Independence. Does progress continue when we are no longer in the room? A level is complete when the team improves it without us.

Agree these measures before the first session. Chosen afterwards, they become a report. Chosen beforehand, they shape what gets taught.

Shortcuts that backfire.

The one-off inspiration day.

A keynote raises interest and changes nothing on its own. Interest without a next step decays fast. If you run an inspiration day, put the foundation dates in the same email.

Training only the volunteers.

The people who sign up are usually the ones already experimenting. Train them alone and you widen the gap between the confident and the stuck, which is the gap causing your inconsistency in the first place.

Assuming familiarity is literacy.

Teams that skip the foundation tend to arrive at the applied level with confident habits that are wrong. Correcting a habit takes longer than teaching the thing properly the first time.

Buying tools before capability.

A licence is not a skill. Rolling a platform out to a team with no shared standard produces low usage and a renewal conversation nobody can defend.

Measuring the wrong thing.

Seat counts, completion rates and satisfaction scores are easy to report and tell you nothing about the work. Pick the operational measure instead, and pick it early.

Where to start.

Start honestly. Ask what proportion of your team could explain, in plain language, what these tools do badly. If that question is uncomfortable, you are at Level 1, and that is a perfectly good place to be. Every capable team was there recently.

If you want a roadmap built around your functions rather than a generic curriculum, begin with Level 1: Foundations and tell us where your team stands today. We will come back with a sequence and a clear recommendation.