The AI Job Skill Gap to Address: Knowing When It’s Wrong
In all the hullabaloo about agentic AI, it can be easy to overlook how diverse the portfolio of artificial intelligence technologies is and how many of these AI technologies are already working in concert to take on increasingly more sophisticated information work. Fortunately, McKinsey Consulting’s Technology Trends Outlook 2025 is here to remind us how AI, as a whole, is changing the way we communicate and collaborate. Equally crucially, the report identifies some of the skill sets that are in growing demand in the modern workplace.
As the McKinsey report notes:
The AI talent pipeline is under pressure. Core skills such as machine learning, Python, and data science are in high demand, though supply still lags for the latter two. Cloud infrastructure expertise is especially scarce, particularly with platforms such as Amazon Web Services. Even as some programming and math-related capabilities are more readily available, gaps in foundational AI skills could slow momentum unless addressed through focused upskilling and development.
To quantify the gap between talent and demand for anyone interested in working with AI, McKinsey looked at two metrics: the percentage of jobs requiring specific skills, and the ratio of talent to market demand. There is almost a perfect ratio of available machine learning talent to employer demand, but if you’re looking for an AWS expert, the demand is nearly ten times the available hiring pool. For data science, there is slightly more demand than there is available science.
McKinsey’s also looked at how AI is shaping existing jobs, and it flags AI assistants as a significant development:
These AI assistants can rapidly browse the web, analyze vast amounts of data, and produce comprehensive reports, significantly reducing research time. This potential to reshape cognitive work has implications for knowledge workers, increasing concerns about output verification and displacement.
It’s worth reiterating that there is a distinct difference between information work, in which information is applied within the context of a specific task or specific workflow to produce specified results — like structured reports, say — versus knowledge work in which interpersonal collaboration and creative originality are components required to produce the outcome.
Another element not discussed in the report but raised in several real-life instances is the conundrum of AI doing these information-work tasks and producing entirely incorrect results, like the legal briefs citing made-up cases, the made-up citations in bibliographic sections in scholarly works, recipes that would be fatal to anyone who tried eating the results, and so on. In all these cases, the bad output was caught and corrected by people who had the experience and expertise to know how to tell good information from bad.
Which raises the ultimate question about AI training and the skills gap. It’s one thing to acquire the skills to build AI tools and integrate them across our communication and collaboration platforms so they optimize our workflows. It’s another to acquire the know-how to know when the AI is performing correctly and producing the best outcome. Investing in that upskilling will be just as important.