EY Global Vice Chair: AI’s largest paradox comes all the way down to what AI cannot do

That is exactly the skill AI is now forcing into every job, whether the title on the door says analyst, technician, or machine operator. The paradox of the AI era is that the more capable the technology becomes, the more the market rewards the other things. Sought-after skills are intrinsically human—it all comes down to what AI can’t do.

While investments in AI rise, companies are still missing out on up to 40% of AI productivity gains due to gaps in talent investment. Many training programs focus on technical AI experience, skipping over critical skills that will determine whether an AI strategy pays off.

What AI still can’t do

Technology thrives at repetitive tasks and surfacing data. But humans are responsible for defining the goal, setting the purpose and providing the context that gives information meaning. People decide which objectives are worth pursuing, recognize when context changes and remain accountable for the results. Without that human direction, data and insights may not be applied effectively or generate their full potential value.

This reframes what an AI-orchestrated workplace looks like. It doesn’t remove the person, but instead the parts of the job that never required human judgment in the first place. What remains is the part only a human can do.

Train AI fluency as a leadership skill

To take full advantage of an AI-enabled world, AI fluency has to become more than a specialty skill – blending the operation of AI tools with evaluation of outputs for relevance and knowing how (and when) to intervene when things go wrong.

Machine identities outnumber human employees 82:1 inside the average organization. Scale like that doesn’t dilute human responsibility. It concentrates it. As AI systems become more autonomous, human judgment becomes even more important.

Analytical thinking is the most desirable core skill employers search for in potential hires, according to a survey by the World Economic Forum, with roughly seven in ten employers calling it essential.

Leaders want an employee who avoids blind delegation and knows how to challenge AI when it matters.

The bottom line: use AI to make your job more efficient, not to take over the judgment decisions that continue to be the most important part of the job.

Don’t navigate alone

In practice, no one organization is solving AI workforce integration alone. Isolated approaches can be time-consuming and costly, as organizations waste cycles rebuilding workstreams others have already solved.

That’s the alliances and ecosystems work I lead at EY, where we help organizations create the right foundations to efficiently scale technology without needing to rebuild what someone else already solved.

Our recent work with the beverage manufacturer Lion evidenced this as collaboration with ecosystem partners led to 75% faster customer response speed and roughly 30% improvement in operating costs across its people function. Rather than trying to build every layer of an AI-enabled people function from scratch, Lion worked with partners who had already solved pieces of that problem elsewhere.

Redesign workflows so humans stay at the center

Workflows must be reimagined to keep humans at the core of decision-making and maintain trust. When hybrid AI-human workflows are built around rigid handoffs, expecting AI to complete its tasks before passing the result to a human, the fundamental assumption is that tasks can be divided cleanly.

In practice, this structure collapses opportunities for meaningful judgment calls and can add more iterative work.

The more durable model organizes work around shared tasks with a truly integrated process, where humans and AI are contributing at every stage rather than in a sequence.

Daikin, a multinational company specializing in heating, ventilation and air conditioning, illustrated this by leveraging hybrid AI and human teams working in tandem to roll out a new enterprise resource planning (ERP) system. While AI assisted with code generation and automated testing, employees maintained oversight for exceptions and high-risk scenarios, reinforcing that judgment is something that no algorithm can absorb.

This approach accelerated delivery by approximately 30%, and the pilot produced a 10% gain in counter efficiency and a 20% faster financial close.

The skill that doesn’t expire

With AI model capabilities constantly in flux, the upskilling conversation falls behind if it only tracks technical updates. Technical skills tied to a specific model version have a short shelf life, the judgment to know when and how to intervene doesn’t.

Judgment is essential to establish secure and trusted AI.

None of this happens by accident. It takes fluency at every level, partners who share the building burden and workflows designed to keep humans in the loop at every stage to provide appropriate oversight.

The organizations getting AI scale right are making deliberate choices about which parts of the job stay human. And AI adoption done well clarifies that role rather than shrinking it.

The views reflected in this article are the views of the author and do not necessarily reflect the views of the global EY organization or its member firms.