Choose the platform.
Buy the licences.
Give employees access.
Run a few demos.
Talk about productivity.
And then?
That is where things become more complicated.
Because giving people access to artificial intelligence is not the same thing as changing how work gets done.
A team may have access to some of the most advanced AI tools available and still use them primarily to rewrite emails, summarise documents or generate first drafts. Those tasks can certainly save time. But they are very different from redesigning a workflow, improving a decision-making process or changing the way an entire function operates.
This distinction is becoming increasingly important.
AI adoption is now widespread, but genuine organisational transformation remains far less common. McKinsey’s 2026 State of AI research found that organisations continue to expand their use of AI and AI agents, while many are still trying to translate individual productivity gains into measurable business value at scale.
The gap between those two things — having AI and knowing what to do with it — may be one of the most important workforce challenges businesses now face.
The tool is usually the easy part
Technology implementations traditionally focus heavily on the technology itself.
Does the platform work?
Can it integrate with existing systems?
Is it secure?
What does it cost?
Who gets access?
All sensible questions.
But AI introduces another one that is harder to answer:
What exactly should people do differently once they have it?
That question sounds obvious. In practice, many organisations have never answered it clearly.
Imagine a marketing team receives access to an enterprise AI assistant.
One employee uses it to brainstorm headlines.
Another uses it to summarise reports.
Another avoids it completely because they are concerned about confidential information.
Another copies AI-generated content directly into client material without checking it carefully.
Technically, all four employees have “adopted” the same technology environment.
Operationally, they are doing four completely different things.
That is why measuring AI transformation purely through licences, logins or tool usage can be misleading.
The more useful question is whether the technology has improved the way a job is actually performed.
Productivity is not the same as transformation
One of the easiest AI wins is speed.
Write this faster.
Summarise this faster.
Research this faster.
Analyse this faster.
Create a first draft faster.
Those benefits matter. But saving twenty minutes on a task is not necessarily transformation.
Transformation happens when the workflow itself changes.
Consider the difference.
A consultant uses AI to rewrite a client email.
That is productivity.
A consulting team redesigns its research process so that AI helps structure initial research, compare sources, identify inconsistencies and prepare a first analytical framework — while consultants validate the evidence, apply judgement and develop the recommendation.
That is workflow redesign.
A developer asks an AI assistant to explain an error message.
Productivity.
A software team redesigns parts of its development lifecycle around AI-assisted coding, testing, documentation and code review, with clear controls for security and human verification.
Workflow redesign.
An HR professional asks AI to rewrite a job description.
Productivity.
An HR function develops a structured process in which AI helps analyse role requirements, identify skills gaps and prepare initial workforce-planning scenarios — while HR professionals remain responsible for context, fairness and decisions.
Workflow redesign.
The difference is not the sophistication of the AI.
It is the sophistication of the way people use it.
“Everyone needs AI training” is probably the wrong starting point
One common response to the AI skills gap is to train everyone.
That sounds sensible.
The problem is that everyone does not need the same AI skills.
A developer, a finance manager, a salesperson, a project manager and a cybersecurity professional may use the same underlying technology in completely different ways.
Teaching all of them the same generic collection of prompts will therefore have limited value.
The more useful question is:
What does AI literacy look like for this role?
For a developer, it may involve:
-
AI-assisted development
-
code generation and validation
-
secure use of coding assistants
-
model APIs
-
AI application architecture
-
testing AI-generated code
-
understanding where human review is essential
For a manager, it may involve:
-
deciding which tasks should and should not be augmented by AI
-
evaluating AI-generated analysis
-
redesigning workflows
-
measuring productivity and quality
-
understanding governance and risk
-
managing teams where humans increasingly work alongside AI systems
For someone working in marketing, it may involve:
And for cybersecurity professionals, AI brings an entirely different set of responsibilities around threat detection, data protection, identity, model security and emerging attack surfaces.
This is one reason role-based capability building is becoming increasingly important. McKinsey has identified role-specific training as one of the organisational practices associated with more mature AI adoption efforts.
The objective is not to make everyone an AI expert.
It is to make people competent users of AI within the context of their actual work.
Prompting is useful. It is not the whole skill set.
For a while, much of the conversation around AI skills centred on prompt engineering.
And prompting does matter.
Someone who can give an AI system clear context, constraints and objectives will usually get better results than someone typing vague instructions into a chat box.
But good AI use involves much more than asking good questions.
Employees increasingly need to know how to:
Judge the output
AI can produce an answer that sounds confident, structured and completely reasonable — and still be wrong.
A skilled user does not simply ask:
“Does this look good?”
They ask:
“What evidence supports this?”
“What might be missing?”
“What assumptions has the model made?”
“Can I verify this independently?”
Understand the data
Employees need to know what information can safely be entered into an AI system.
Customer data?
Internal financial information?
Source code?
Employee information?
Confidential documents?
The answer may depend on the tool, account type, company policy and regulatory environment.
That makes AI literacy partly a data-literacy and security skill.
Recognise when AI is the wrong tool
Not every task becomes better simply because AI is involved.
Some decisions require accountability.
Others require specialist expertise.
Some require empathy, negotiation or contextual understanding that cannot simply be delegated.
Knowing when not to use AI is part of knowing how to use it well.
Work with AI rather than simply consume its output
The most capable users increasingly treat AI as an iterative working partner.
They provide context.
Review the answer.
Challenge it.
Add missing information.
Request alternatives.
Verify.
Refine.
The human remains responsible for the outcome.
That distinction becomes increasingly important as AI systems become more capable.
The uncomfortable part: AI exposes weak processes
There is another reason AI transformation can become difficult.
Sometimes the technology is not the problem.
The process is.
Ask a team to automate or augment a workflow and suddenly uncomfortable questions appear:
Why do we do this step?
Who actually owns this decision?
Why is the same information entered three times?
Why does approval require five people?
Why is this report produced every week?
Does anyone actually use it?
AI has an interesting way of exposing organisational habits that nobody previously questioned.
And that can be valuable.
Because the objective should not always be:
“How can AI make this existing process faster?”
Sometimes it should be:
“If we designed this process today, knowing what technology can now do, would we design it this way at all?”
That is a much more powerful transformation question.
AI readiness is partly a management problem
Employees often get blamed when new technology is not adopted quickly enough.
“They are resistant to change.”
“They are not using the tools.”
“They need more training.”
Sometimes that is true.
But employees cannot redesign an organisation’s workflows alone.
Managers need to create clarity around:
-
which tools are approved
-
what employees are allowed to use them for
-
which information can be shared
-
where human review is mandatory
-
what good AI use looks like
-
how performance will be measured
-
how experiments can safely become standard processes
Without that clarity, employees are left to improvise.
Some become enthusiastic power users.
Others become cautious observers.
Others quietly use public AI tools without telling anyone.
And organisations end up with very uneven adoption.
That is not simply a training issue.
It is an operating-model issue.
The organisations getting more value are redesigning work
The emerging pattern across current AI research is increasingly clear: organisations that capture greater value from AI are not simply deploying more tools.
They are changing how work is organised around those tools.
McKinsey’s research has highlighted practices such as redesigning workflows, defining clear adoption roadmaps, establishing governance, creating role-based capability programmes and measuring AI-related performance.
Microsoft’s recent Work Trend Index research points in a similar direction: organisations are increasingly thinking about work as a combination of human judgement and AI or agent-based execution rather than simply adding AI as another piece of software.
That shift matters.
Because the question is gradually moving from:
“How do we get employees to use AI?”
to:
“What should humans do, what should AI do, and how should the two work together?”
That is a much more strategic conversation.
A practical way to approach AI capability
Organisations do not need to transform everything at once.
A better approach is often to move systematically.
1. Start with the work, not the tool
Identify the tasks and workflows that consume significant time, create bottlenecks or require large amounts of repetitive analysis.
Do not begin with:
“What can this AI platform do?”
Begin with:
“Where could our work be better?”
2. Map capabilities by role
Identify what different groups actually need to know.
Avoid the temptation to give everyone exactly the same AI training.
A finance team and a software engineering team may share basic AI literacy, but their advanced capability requirements will be completely different.
3. Build a safe experimentation environment
People need opportunities to practise.
Not theoretical demonstrations.
Actual scenarios based on their work.
That is where employees begin to understand both the possibilities and the limitations.
4. Create rules people can understand
AI governance should not live only inside a 40-page policy document.
Employees need clear answers to practical questions:
Can I upload this document?
Can I use AI with customer information?
Do I need to disclose AI-generated content?
What needs human approval?
Which tools are approved?
Good governance makes responsible adoption easier, not harder.
5. Measure outcomes, not activity
The number of prompts employees submit is not a business outcome.
Instead, measure things such as:
AI adoption should eventually connect to something the organisation actually cares about.
The real AI skill may be learning how to learn
There is one final challenge.
Whatever employees learn about AI today will probably not be enough indefinitely.
The tools are changing too quickly.
Interfaces change.
Capabilities expand.
New models appear.
New regulations emerge.
New risks appear.
Entire workflows that seemed impossible a year ago become normal.
So perhaps the most valuable capability organisations can build is not mastery of one AI platform.
It is the ability to continuously learn, experiment, evaluate and adapt.
That means creating employees who are comfortable asking:
Could this process work differently?
What can this tool do now that it could not do six months ago?
Where is human judgement still essential?
What new risk does this capability introduce?
What should we learn next?
Technology changes.
The ability to learn with it is what lasts.
So, we bought the AI tool. Now what?
Use it.
But do not stop there.
Identify where it genuinely improves work.
Teach different roles how to use it responsibly.
Redesign the workflows that deserve redesigning.
Give employees clear boundaries.
Measure whether anything actually became better.
And keep learning.
Because ultimately, AI transformation is not really about how much technology an organisation owns.
It is about whether its people know what to do with it.