Artificial intelligence promises to free up time, but it may also weaken a resource companies cannot automate: judgment. A global study released by the IBM Institute for Business Value on September 21, 2026 puts critical thinking at the center of the future-of-work debate. Its findings do not simply pit people against machines. They reveal a management contradiction: organizations ask employees to supervise AI more effectively while sometimes removing the experiences through which that ability is developed.
Conducted with Oxford Economics between April and June 2026, the research combines two surveys: 1,500 chief human resources officers and senior executives responsible for workforce strategy across 21 geographies and 23 industries, and 8,800 full-time employees in 28 countries. IBM reports that 60% of employees worry about skills erosion. Among those concerned, three out of four believe AI has already begun to diminish at least some of their abilities.
The human-control paradox
Executives clearly recognize the need for supervision. Seventy-one percent of HR leaders identify the ability to monitor, validate or override AI outputs as the workforce’s most essential skill. Yet only 29% of employees rank judgment as important. IBM also finds broader agreement around critical thinking and problem framing: 57% of executives and 49% of employees consider them essential.
This gap does not necessarily mean workers are ignoring the risk. It may also reflect how AI is introduced to them: as a tool for speed, rarely as a system to challenge. Checking a recommendation, spotting a weak assumption, restoring missing context and deciding when to stop automation are now productive tasks in their own right. They take time even when they remain invisible on performance dashboards.
The invisible work created by automation
Four in five HR leaders surveyed by IBM acknowledge that AI adoption creates invisible work, including validating recommendations, correcting mistakes, adding context and managing exceptions. Among employees, 42% say AI either increases their workload or creates work that goes unrecognized. More troublingly, 43% say they are blamed when an AI system gets something wrong.
The issue is therefore no longer just time saved. It is about how authority and risk are distributed. Who signs off on the decision? Who is allowed to disagree with the model? Who documents the exception? In organizations where HR shares responsibility for defining which decisions must remain human-led, 76% of employees feel safe questioning or overriding an algorithmic recommendation. That falls to 43% when HR is merely advisory.
Automating without draining the learning pipeline
The quietest danger concerns the development of future experts. A first analysis, an imperfect memo, preparing a case file or patiently comparing sources can look like low-level work. Yet these are the experiences through which professional intuition and discernment are built. If AI absorbs them completely, companies may produce more quickly today while shrinking their supply of judgment tomorrow.
Recent work by the International Labour Organization points in the same direction. Its August 2026 report says AI adoption is increasing demand for higher-order cognitive and socioemotional skills, adaptability, resilience and human agency. The ILO also describes the ability to understand and use AI safely and ethically as a new foundational skill. The OECD similarly emphasizes the need to use, analyze and interpret data, while embedding training in broader policies for transparency, explainability and accountability.
Governance becomes an economic advantage
IBM finds that organizations clearly defining work as human-led, AI-assisted or AI-executed report an 18% reduction in risk and a 20% improvement in quality. Yet only 26% of the companies in the study make that distinction explicit. Nearly half, 46%, do not involve the CHRO when AI strategy is being defined.
That absence is strategic. Installing a tool is a technology decision; redesigning responsibility is an organizational one. A mature company does more than teach employees to write better prompts. It identifies decisions that require human sign-off, creates a genuine right to challenge automated outputs, measures verification work and protects spaces where employees continue solving problems without assistance.
AI return on investment should therefore not be measured only in hours saved. It should include decision quality, trust, intercepted errors, skills growth and the ability to handle unfamiliar cases. Without those indicators, an organization can appear more productive while becoming less capable of recognizing its own mistakes.
Human capital after the novelty wears off
The signal from IBM’s study extends beyond the HR function. When the same models are available to every competitor, owning them is no longer enough to set a company apart. Advantage shifts toward the quality of its questions, practical knowledge, creativity and the ability to reject an answer that sounds convincing but is wrong.
AI can accelerate work without guaranteeing better decisions. The next stage is not about choosing between automation and human expertise, but designing the relationship between them. Companies that deliberately preserve learning, dissent and accountability can turn saved time into new capability. Others may discover too late that short-term efficiency consumed the knowledge required to keep it under control.
Sources
- IBM Newsroom — New IBM CHRO Study: AI Puts Critical Thinking at the Center of Workforce Priorities, September 21, 2026
- IBM Institute for Business Value — Designing the Thinking Organization, 2026
- International Labour Organization — Changing landscape of skills in the age of AI, August 13, 2026
- OECD — AI and skills: What we know so far, June 5, 2026