If Most Workers Will Not Build AI, What Should They Learn Instead?
Why working effectively with AI may depend less on technical expertise and more on knowing how to think, analyse and decide
Whenever a major new technology arrives, there is a natural tendency to focus on learning the technology itself.
When spreadsheets became important, people learned Excel. As organisations became more data-driven, employees learned analytics, dashboards and visualisation tools. Now that AI is becoming part of everyday work, many professionals are wondering whether they need to learn Python, machine learning, how large language models work or even how to build AI systems themselves.
For some people, the answer will certainly be yes. But for most business professionals, probably not.
That is one of the more interesting messages from the OECD’s 2026 review, AI and Skills: What We Know So Far. The OECD estimates that fewer than 1% of workers will need advanced AI-specific skills, such as programming and developing AI systems. That statistic needs to be handled carefully. It does not mean that fewer than 1% of workers will need AI skills. The OECD’s broader conclusion is that many more workers will need digital capabilities and the ability to use, analyse and interpret data, alongside managerial and human skills such as problem-solving, creativity and innovation.
That changes the question. Instead of asking “Do I need to become an AI expert?”, perhaps most of us should be asking: “If AI can do more of my work, what do I need to become better at?”
Working with AI is not the same as building AI
Imagine a marketing manager using AI to analyse thousands of customer comments. She does not need to know how to build the language model doing the analysis, but she still needs to know what business problem she is trying to understand.
Are customers becoming less satisfied? What are they dissatisfied about? Is the feedback representative of the wider customer base? Are the themes identified by AI genuinely supported by the comments? If something has changed, what might explain it?
These questions lead naturally to possible explanations. Perhaps prices have changed, service standards have deteriorated or customer expectations have shifted. In analytics, these possible explanations become hypotheses that we can investigate. Only then do we ask what data we need and how we should analyse it.
This is why the OECD distinction matters. A relatively small group of people will build, develop and maintain AI systems and need deep technical expertise. A much larger group will work with AI. They need enough understanding of AI to use it productively, enough understanding of data and analytics to evaluate what it produces, and enough understanding of their business to decide what the output actually means.
AI changes the tools, not the need to think analytically
This connects strongly with something we have taught at FYT for many years: analytics does not start with data.
It starts by defining the problem and asking the right questions. From there, we consider possible causes and develop hypotheses. Only then do we determine what data we need, collect and process it, analyse it, interpret the results in the appropriate business context, and communicate the findings so decision makers can act.
Our FYT Analytics Thought Process is therefore:
Define the problem → Develop hypotheses → Collect & process data → Analyse → Interpret results → Communicate & influence
AI does not make this process obsolete. It can already help generate hypotheses, clean and restructure data, perform statistical analysis, identify patterns, create charts, interpret results and draft presentations. What it cannot remove is the need to understand why we are doing each of those things.
If AI suggests a hypothesis, is it plausible? If it analyses the data, is the method appropriate? If it identifies a relationship, what does that relationship actually mean? If it recommends an action, is there enough evidence to support it? AI can increasingly help us do the analytics. We still need to know how to think analytically.
When AI does more, where does human value move?
There is an assumption underneath much of the discussion about AI and work: if AI can perform more of a task, perhaps people need to know less.
The OECD evidence suggests a more complicated picture. Its research points towards continuing demand for broader digital, analytical, managerial and human capabilities as AI becomes more widely used. Earlier OECD workplace studies also found AI implementation frequently reorganising jobs rather than simply eliminating them, with work shifting towards tasks where people continued to provide value.
Consider a business analyst preparing a monthly management report. What once took two days of extracting data, cleaning it, calculating variances, creating charts and drafting commentary might increasingly be completed with AI and automation in a few hours.
The opportunity is not simply to produce more reports. The analyst can spend more time investigating why an important variance occurred, testing possible explanations, challenging unusual results and helping management understand what the numbers mean for the decision in front of them.
The mechanics become easier, but the value of the role does not necessarily disappear. It moves towards interpretation and judgment.
This does not mean every worker will benefit equally or every job is safe from disruption. But “AI can do more” and “people need fewer skills” are not the same statement. In some roles, AI may actually raise the bar because once the software can produce the first analysis, the human contribution increasingly lies in deciding whether that analysis is appropriate and what should happen next.
So what should most business professionals learn?
I would start with AI fluency, but not simply the ability to use the latest chatbot. AI fluency means understanding what these systems are reasonably good at, where they can fail and when an answer deserves closer scrutiny. It includes knowing how to give AI useful context and clear instructions while recognising that a polished answer is not necessarily a correct one.
Alongside that, we need analytical thinking: the ability to define the problem and develop plausible hypotheses before rushing into the data. We need data literacy, which the OECD specifically highlights as increasingly important, so that we can recognise poor data, inappropriate comparisons and conclusions that go beyond the evidence. And throughout the process, we need critical evaluation and business judgment to question assumptions, consider alternative explanations and decide what can responsibly be concluded.
How much technical AI knowledge someone needs will depend on the role. An AI engineer obviously needs very different expertise from a sales manager. A data analyst may need greater technical depth than an HR business partner, while a senior executive may never write code but still needs to understand when an AI-supported recommendation deserves confidence and when it should be challenged.
So rather than asking whether someone “knows AI,” I would ask: “What does this person need to be able to do with AI in their job?” That is a much better starting point for deciding what they need to learn.
What this means for training
This is where the OECD research connects naturally with the direction we have taken in our own training at FYT. Our programs span data management, analytics, visualisation, Excel, Power BI, Tableau and AI-enabled analytics. The tools differ, but the underlying objective is consistent: helping business professionals turn data into meaningful insights that support better decisions.
In our Data Analytics in the Age of AI program, for example, participants work through the analytics thought process, from defining the problem and developing hypotheses through data preparation, analysis and interpretation. Generative AI is used alongside conventional tools such as Excel so participants can compare approaches and, importantly, evaluate whether the results make sense.
The same philosophy applies to our Power BI training. The objective is not simply to learn the software or build an attractive dashboard. Learners use dashboards to answer business questions and gain insights.
That matters because the technology will keep changing. The AI platform we use today may look very different in a few years, but the ability to define a problem, develop plausible explanations, work with evidence, interpret an analysis and communicate what it means has a much longer shelf life.
So I do not think the future of workplace training is simply about adding more AI courses. It is about making the skills people already need to think, analyse and decide more effective in an AI-enabled workplace.
AI can close skill gaps. It can also expose them.
There is another interesting side to this. AI can help organisations compensate for capabilities they do not currently have.
The OECD’s 2025 survey of more than 5,000 SMEs across seven countries found generative AI being used by 31% of surveyed SMEs. Among users that had experienced skills gaps, 39% said generative AI helped compensate for them.
That can be particularly valuable for smaller organisations without dedicated research, writing or analytical teams. But being able to produce something is different from being able to judge whether it is good.
AI can create a chart or calculate a correlation in seconds, but that does not mean the chart is appropriate or the relationship has been interpreted correctly. It can produce a polished recommendation, but polish does not tell us whether the evidence supports it.
In that sense, AI can expose a skill gap even while helping to fill one. Once the mechanics become easier, weaknesses in reasoning and judgment become harder to hide.
These are not another methodology. They are capabilities that help us move intelligently through the analytical process, regardless of whether we are using Excel, Power BI, Tableau, an LLM or whatever tool comes next.
This also helps us avoid several common mistakes. The first is misreading the OECD statistic: fewer than 1% needing advanced AI-specific skills is not the same as saying fewer than 1% need AI capability. Another is equating AI readiness with prompting. Prompting matters, but a better prompt only produces a better response; someone still needs to determine whether that response is reliable and useful. Finally, different roles should not receive identical AI training when their decisions, risks and uses of AI are different.
What if AI becomes much more capable?
What if AI eventually becomes much better at defining problems, generating hypotheses, analysing evidence, interpreting results and recommending decisions? That is entirely possible, and AI is already moving in that direction. But greater capability may make human judgment more consequential rather than irrelevant.
The OECD’s 2025 research into algorithmic management offers a glimpse of why. In a survey of more than 6,000 mid-level managers across six countries, managers using algorithmic management tools reported benefits including access to more information and faster decision-making. At the same time, nearly two-thirds reported concerns about their impact on workers, including unclear accountability when algorithmic decisions went wrong.
As AI becomes more capable, the question therefore becomes less about whether it can analyse or recommend and more about who is responsible for what happens next. Organisations still need people who understand the context, recognise when something does not make sense, consider the consequences and take responsibility for action.
Knowing which buttons to press may become less valuable. Knowing when to trust AI, when to challenge it and what to do with its output may become more valuable.
Perhaps we are asking the wrong skills question
Whenever a major technology arrives, it is natural to focus on learning the technology. But perhaps that is too narrow a way to think about AI.
For a relatively small group of workers, deep technical AI expertise will be essential. For most business professionals, the challenge may be less about learning how to build AI and more about learning how to work intelligently alongside it.
That means understanding enough AI to use it well, enough analytics to know what question we are trying to answer, enough data literacy to question the evidence, and enough judgment to interpret the result and decide what should happen next.
The OECD evidence gives us an important reminder: the future of work is not simply a story about what AI will be able to do. It is also about the skills people will need because AI can do more. For most business professionals, becoming more technical may certainly be useful.
Becoming better thinkers may be even more important.
Quick Reference
Advanced AI-specific skills
Technical capabilities associated with programming, developing, maintaining or building AI models and systems. These are the skills referred to in the OECD’s “less than 1%” estimate.
AI fluency
A practical understanding of what AI can do, where it can help, where it can fail and how to use it responsibly.
Analytical thinking
A structured way of approaching a business problem by asking the right questions, considering possible causes, examining evidence and interpreting what the findings mean.
Define the problem
The first stage of the FYT Analytics Thought Process: understanding the business issue and asking the right questions before analysing data.
Hypothesis
A possible explanation for the business problem that can be investigated using appropriate evidence. Data can be used to test whether the evidence supports or challenges that explanation.
Data literacy
The ability to understand, analyse, interpret and question data rather than simply consume numbers, charts or AI-generated conclusions.
Critical evaluation
Assessing whether an analysis or AI-generated answer is supported by appropriate data, assumptions and reasoning.
Interpretation
Understanding what analytical findings mean in the relevant business context. A statistical result does not necessarily explain itself.
Business judgment
Using evidence, context, uncertainty and consequences to determine what can reasonably be concluded and what action may be appropriate.
Algorithmic management
Software used to automate or support managerial activities such as allocating, instructing, monitoring or evaluating work.
References
OECD (2026), AI and Skills: What We Know So Far. The primary source for this article. It examines how AI is changing skill requirements, including the distinction between advanced AI-specific technical skills and the much broader digital, data, managerial and human capabilities required across the workforce.
OECD (2025), Generative AI and the SME Workforce: New Survey Evidence. Based on a survey of more than 5,000 SMEs across seven countries, examining generative AI adoption, performance, skills gaps and workforce preparation.
OECD (2023), The Impact of AI on the Workplace. OECD research examining how AI adoption affects jobs, tasks, skills and the organisation of work.
OECD (2025), How Widespread Is Algorithmic Management in Workplaces? Examines the use of algorithmic management and its reported effects on managerial decision-making, accountability, explainability and worker well-being.
FYT Academy FYT’s current training portfolio spans data management, analytics, data visualisation, Excel, Power BI, Tableau and AI-enabled analytics, with an emphasis on helping business professionals turn data into insights and decisions.
Comments