By 2027, generative AI will augment 30% of knowledge worker tasks, up from zero in 2023. The numbers are clear, but most organizations are not doing enough to prepare. According to Gartner research from March 2025, only 18% of workers believe their company concretely supports GenAI integration into daily work. And just 12% have used AI tools to significantly reduce the burden on critical tasks.
The problem is not the technology: it is the training. Gartner has identified four specific skills that workers must develop to use GenAI productively. These are not theoretical skills -- they are operational capabilities built through practice and quickly lost without it.
The forgetting curve challenge
Before diving into the four skills, it is worth understanding why traditional training does not work for GenAI. The problem is called the forgetting curve, a phenomenon documented by Hermann Ebbinghaus in 1885: 50% of new knowledge is forgotten within one hour of learning, 70% within 24 hours, 90% within one week.
For GenAI, this is a concrete obstacle. A two-day workshop followed by weeks without practical application does not produce results. Gartner research confirms this with data: only 29% of workers use AI tools daily. Those who use them every day are 3.4 times more likely to report a significant increase in their productivity compared to those who use them once a week.
The proposed solution is the 70/20/10 framework: 70% of learning time should be dedicated to hands-on practice, 20% to social learning through communities and peer review, and only 10% to traditional structured training. The concise model is "see one, do one, teach one": first you observe, then you do, then you teach others.
Skill 1: identifying use cases
The first skill is the ability to recognize where GenAI can create concrete value. It is not about knowing all available features, but about being able to read work processes and understand where AI can solve a problem, save time, or improve the quality of an output.
Gartner uses the term "opportunity spotters" to describe workers who develop this capability at an advanced level: people who become a competitive advantage for the organization because they identify AI applications that others do not see. The cited case is that of Vizient, an American healthcare company, where Chuck DeVries, SVP and technology officer, observed that meaningful use cases almost always emerge from where you least expect them.
When GenAI is adopted without clear use cases, workers struggle to understand how to integrate it into their workflow. The result is sporadic, superficial usage that does not produce measurable results.
Skill 2: tech fluency
The second skill is understanding the basic mechanisms of GenAI: how language models work, what their capabilities and limitations are, how training data influences results, and what the ethical implications of their use are.
Tech fluency does not mean knowing how to code or manage infrastructure. It means understanding when to use one model over another, being able to evaluate whether an AI tool is suitable for a specific task, and knowing the ethical considerations that come into play when working with business data.
Gartner emphasizes that GenAI capabilities are a "jagged frontier": what it can and cannot do changes constantly. First-generation language models could not perform elementary calculations, while today they handle advanced mathematics. At the same time, they continue to fail on seemingly simple tasks. This jagged frontier requires constant updating of one's knowledge.
Skill 3: prompt engineering
The third skill is the most operational: knowing how to write effective instructions to get the desired results from AI. Prompt engineering is not a term reserved for developers; it is a cross-functional skill that applies to everyone who uses GenAI tools in their daily work.
The most common mistake among those who start using GenAI is treating the prompt like a Google search or a casual question. GenAI does not fill in the gaps of a vague instruction the way a colleague with contextual experience would. It needs clear instructions on what to do, how to do it, and with what constraints.
The other frequent mistake is treating the interaction with AI as a single exchange, rather than an iterative conversation. The best results come from progressively refining instructions, adding context, and correcting direction based on intermediate outputs. An experienced user knows how to build a multi-turn conversation that drives the model to produce exactly what is needed.
Skill 4: evaluating results
The fourth skill is the most undervalued: the ability to judge whether an AI-generated output is accurate, useful, free of bias, and consistent with the organization's objectives. Gartner calls this output discernment.
It is not enough to verify that the output is formally correct. A well-written text can contain inaccurate information, circular reasoning, or partial perspectives that stem from biases in the training data. The discernment skill requires evaluating sources, understanding the model's potential failure points, and knowing how to iterate on prompts to get better results.
Gartner also warns of the opposite risk: limiting human judgment to mere validation of AI output restricts the value of human-machine collaboration. The most effective model is bidirectional, where AI informs human analysis and human judgment guides AI usage.
How to build a development program that works
For each of the four skills, Gartner proposes a progression from beginner to expert, with concrete learning activities for each level. Those starting from zero can begin with guided exercises such as generating a document, creating captions for a set of images, or performing sentiment analysis on reviews. Intermediate levels include internal hackathons and peer review sessions. Those with advanced experience can work on building AI agents or custom GPTs on cloud platforms.
The principle guiding everything is reducing the time between learning and application. In traditional workshops, weeks or months separate training from practical use. For GenAI skills, this interval must shrink to minutes: you learn something new and apply it immediately, in the same work context.
By 2027, more than half of organizations will fund structured AI literacy programs, driven by the difficulty of realizing the expected value from GenAI investments. Those who build these skills now, before they become a widespread requirement, will have a significant advantage in recruiting, productivity, and the ability to innovate.
