Companies are introducing ChatGPT, Gemini, Copilot, automation tools, AI design platforms and AI-assisted research into everyday work. On paper, that looks like progress. But there is a problem I think many companies underestimate: Giving employees access to AI does not mean they know how to use it well.
A company can pay for the tools, announce an AI initiative and still have employees who are unsure where AI fits into their work, how much they should trust it or even how to judge the output it gives them. That is why I think we need to look beyond AI adoption. The real question is not simply, “Do our employees have access to AI?”
It is: “Do they understand their work well enough to use AI intelligently?” Personally, an AI-inclusive workplace is one where people at different skill levels, in different roles have a fair opportunity to understand AI, apply it to their work and develop enough competence to know when the output is useful and when it is not.
That starts with five things.
- Access doesn’t mean adoption
There is a big difference between giving someone an AI tool and helping them actually adopt it. I have seen this across designers, developers, students, colleagues and other working professionals. Many people use AI, but that does not necessarily mean they know how to use it in a way that improves their work. Some use ChatGPT, Gemini, Claude, Grok and the likes mainly for basic writing without knowing essential use cases.
Some know that AI can automate tasks but do not know how to turn a repeated process into a workflow. Some have access to tools they barely use because nobody has shown them what those tools can actually do in their role. And that matters because AI adoption is not one-size-fits-all.
For a manager, AI may help with research, documentation, meeting preparation and minute taking, reporting or content development. For a graphic designer, it may support ideation, creative direction, image generation, research or faster production. For a marketer, it may support audience research, campaign planning, content workflows, email sequences, lead nurturing and reporting. For a developer, the usage may be different again.
So if a company says, “We are adopting AI,” the next question should be: Adopting it for what? That answer should connect directly to the work employees are already doing. Microsoft’s 2025 Work Trend Index shows why this gap matters. The report found that 67% of leaders were familiar with AI agents, compared with only 40% of employees. That suggests that leadership interest in AI can move faster than employee readiness. Access may be the first step but it is not the finish line.
2. Understand the work before automating the work
This is probably the most important point for me. Before you automate a process, you need to understand the process. Before you ask AI to do a job, you need to understand what good work in that job actually looks like. Take proposal writing. A good proposal is not just a document with a polished introduction, a list of services, and a price.
You need to research the client. You need to understand the organization. You need to know what problem you are solving. You need to understand the voice of the company you are writing for. You need to know what information matters and what should be left out.
If an employee does not understand those things, AI can still generate a proposal for them. The problem is that they may not know whether the proposal is actually good. The same thing applies to automation.
Imagine a company running a marketing campaign and collecting newsletter sign-ups. Those subscribers may need regular updates on industry trends. A company could assign people to research topics, write emails, format them, schedule them, and repeat the process every week. Or the company could design a workflow where AI supports the research, drafting, organization and distribution, while a human still sets the direction and reviews the final output.
That could save a significant amount of manual work. But to automate that process properly, someone first needs to understand the workflow. Where does the information come from? What makes an update useful? Who approves it? What should never be automated? What happens when the AI produces something wrong? If you cannot explain the workflow clearly you probably are not ready to automate it.
The better question is not: “What can AI automate?” It is: “Do we understand this workflow well enough to automate it responsibly?”
3. AI training should be role-specific and level-specific
One of the fastest ways to waste an AI training budget is to put everyone in the same room, teach them the same thing and expect the same result. From my experience teaching technology, even a small class can contain people with very different starting points. One person may have a technical background. Another may come from accounting, healthcare, administration, design or a completely different field.
Some people understand new tools almost immediately. Others need more time and more examples before the idea becomes practical. The workplace is no different. A beginner may need to understand what AI is, what it can do, what it cannot do and how to interact with it responsibly.
A more advanced employee may need training on automation, AI agents, advanced research or connecting tools into workflows. A designer should not have the same AI curriculum as someone in finance. A manager should not necessarily have the same training as a developer.
Training needs to answer two questions: What does this employee actually do? What is their current level of understanding? That is where training becomes useful.
World Economic Forum’s Future of Jobs Report 2025 found that employers expect 39% of workers’ existing skills to change by 2030, while 59 out of every 100 workers would need training.
International Labour Organization also argues that AI literacy is becoming a foundational skill, alongside adaptability, resilience, and human agency.
But “AI literacy” should not mean teaching everyone the same prompts. It should mean helping people understand how AI fits into their own work. That could involve onboarding sessions, periodic workshops, role-specific training, internal knowledge-sharing or bringing in experienced external professionals.
And it should not be a one-time event, AI is changing too quickly for that.
4. AI fluency is a shared responsibility
Companies have a responsibility to help employees adapt. They should provide access to useful tools. They should create clear policies. They should provide relevant training. They should help employees understand where AI fits into their work. But the responsibility does not stop with the company. Employees also have to keep learning.
I personally spend time reading AI newsletters, following new tools and technological developments, testing different use cases and thinking about where AI can improve the work I do and the people around me.
Nobody can completely outsource that responsibility for me. A company can provide the environment. It cannot manufacture curiosity. This matters because AI fluency is becoming less like a specialist skill and more like general digital literacy.
OECD has found that most workers exposed to AI will not need specialist AI skills such as machine learning or natural language processing.
They will however, see their tasks and skill requirements change. That means the goal is not for everyone to become an AI engineer. The goal is for people to understand AI well enough to use it within their profession.
Companies should create the conditions for that learning. Employees should take advantage of them. That is the shared responsibility.
5. Dependence without competence is the real risk
A lot of conversations about AI warn people not to become too dependent on it. I think that misses the real problem. Using AI heavily is not necessarily a bad thing. The problem is depending on AI for work you do not understand.
If you understand design, AI can speed up your creative process. If you understand research, AI can help you research faster. If you understand writing, AI can help you draft and refine faster. If you understand a workflow, AI can help you automate parts of it.
But if you cannot judge the output then you have a problem. You may accept incorrect information because it sounds convincing or send weak work to a client because it looks polished. You may even automate a broken process or produce more work without actually producing better work.
That is why companies should focus less on whether employees are “using AI enough” and more on whether they are competent enough to evaluate what AI produces. ILO’s recent work on AI and skills makes a similar point from another angle: as AI changes tasks, workers still need higher-order cognitive skills, adaptability, judgment and human agency.
AI can accelerate work. It cannot (at least till further notice due to rapid AI advancements) replace the need to understand what good work looks like. The better goal is not: Use AI more. It is: Know your work well enough to use AI intelligently.
What should companies do next?
If a company wants to become more AI-enabled without leaving employees behind, I would start with five questions:
- Do our employees understand their jobs well enough to use AI properly?
- Do they know where AI actually fits into their workflows?
- Are we training people according to their roles and current skill levels?
- Who owns continuous AI education inside the organization?
- Are employees learning how to judge AI output or are they only learning how to generate it?
Those questions are more useful than simply asking how many AI tools the company is subscribed to. Because the goal should not be to become a company with more AI. The goal should be to become a more capable company because of AI.
That requires tools. But it also requires training, judgment, curiosity, domain knowledge and people who understand the work well enough, among other things, to know when AI is helping and when it is not. AI adoption is easy to announce. AI readiness takes work.
If your organization is introducing AI into everyday workflows, start by mapping where employees already use it, where the biggest capability gaps are and which roles would benefit most from targeted training.
That is where meaningful adoption begins.