Generative AI pilots stall with little or no measurable impact. The cited cause is a learning gap, not model quality.
Why this matters
Another tool won't fix how a team decides
Most organisations now have AI tools. Far fewer have a shared way for every role to use them safely, together.
The numbers
The tools are there. The results aren't.
People aren't prepared.
Employees who say they have received AI training.
Employees who say their workplace has a policy or guidance on generative AI.
Germany's rank among countries surveyed for AI literacy.
Mistakes are already happening.
People who rely on AI output without evaluating its accuracy.
Employees making mistakes in their work because of AI.
Leadership alone isn't reaching people.
Frontline employees who say they receive strong leadership support for AI.
Employees who feel positive about generative AI, without and with strong leadership support.
Frontline employees who see a strong link between what leaders say about AI and what the organisation does.
It's now a legal duty in the EU.
Providers and deployers must support staff AI literacy. No set level is required. It depends on role and context.
The gap isn't the technology. It's whether people in different roles know how to work with it, and with each other.
Figures from independent studies, 2025–2026. Surveys differ in method and sample.
Three common assumptions
The tool will solve it.
often heard from developers
Belief. A new platform or agent brings the structure.
Reality. Tools speed up the work. They don't decide who checks it, who's accountable, or what happens when it's wrong for the person on the other end.
Managing it is my job.
often heard from managers
Belief. Clear management keeps AI on track.
Reality. Managers can't see every AI output. The judgement sits with each role, and each role needs to know what to look for.
We've done the training.
often heard from compliance and L&D
Belief. A policy and an e-learning course cover it.
Reality. Generic training isn't tied to real work. People pass the course, then meet a case it never mentioned.
What's already out there, and what it leaves open
| Category | Examples | What it does well | What it leaves open | How role paths fit |
|---|---|---|---|---|
| AI coding agents and orchestrators | AIDEN, GitHub Copilot, Cursor | Speed up building. Some, like AIDEN, add a spec to approve and a review step before code is merged | They review the code, not the consequence for the person who receives the output | Each role learns what to check in its part of the chain, not only in the code |
| Code and delivery platforms | GitHub, GitLab | Version control, pull-request reviews, traceability of changes | A review approves a change, not a decision about a citizen | Adds a named reviewer for AI decisions that affect people |
| Work tracking | Jira, Azure Boards | Tasks, handoffs and status in one place | A ticket can close while nobody owns the outcome | Names an owner and a recovery route for critical handoffs |
| Team methods | Scrum, RACI | Rhythm and recorded responsibility | They say who, not how to judge AI output | Rehearses the decision, not just the name in the box |
| Generic AI training | E-learning, AI literacy courses | Shared basics and vocabulary | The same course for everyone, unconnected to the team's real work | Role-specific paths, practised on a real workflow |
| Governance frameworks | NIST AI RMF, ISO/IEC 42001, EU AI Act | Define what must be governed | Say what, not how each role does it day to day | Turns requirements into each role's tasks and handoffs |
| Monitoring tools | MLOps, model monitoring | Track model performance and drift | Can't see the citizen who got a confusing letter | Connects technical signals to frontline feedback |
Your tools record who does what. Role paths teach people how to do it, together, when AI is involved.
Why it has to be role-based
In my research, failures happened between roles: risk was "another team's job", fairness was left to the developer, and nobody owned the effect on the citizen. A tool or a course aimed at everyone misses those gaps. Role paths give each person what they need for their part of the chain, and the team rehearses the handoffs together.
See the four stories → Meet the personas → How the learning works →