ICYMI
• Is AI Blunting Your Strategy? As strategy development relies increasingly on LLMs, true differentiation from competitors is becoming more elusive
• Double (AI) Agents. What intelligence services can teach companies about AI agents.
• The Centaur Age Is Here. Which of Your Are Lame? Human-AI teams are proliferating across the enterprise. Most of them are making things worse.
In May 2026, Eric Schmidt gave the commencement address at the University of Arizona. He was booed repeatedly, and the boos grew louder when he turned to artificial intelligence. “When someone offers you a seat on the rocket ship,” he told them, “you do not ask which seat. You just get on.” They were not convinced. A March 2026 NBC News/Hart Research poll put AI’s overall net favourability among registered US voters at minus 20, with 57 per cent saying the risks outweigh the benefits. Student anger, consumer unease and employee resistance are not the same thing, but together they point to a widening trust gap around AI.
It is easy to dismiss this as youthful idealism. But the people closest to the work can often see things that leaders cannot. The Post Office Horizon scandal ran for years because the organisation trusted its software over its own people. More than 900 postmasters and branch staff were wrongfully prosecuted as a result.
Employee scepticism is useful data. Not all of it is insight. Some is fear, self-interest or poor information, so leaders have to separate obstruction from genuine warning. But the gap is striking. In a WalkMe/SAP survey of 3,750 respondents across 14 countries, only 9 per cent of workers trusted AI for complex decisions, compared with 61 per cent of executives. The people doing the work tend to see three things their leaders miss.
The first is hidden risk. In early 2024, a Microsoft engineer named Shane Jones spent months warning that the company’s AI image tool was producing violent and sexual content. He was directed elsewhere and, he says, asked to delete a post about it. He took his concerns to US senators and the regulator, and was proven right.
The second is hidden cost. AI makes it easy to produce work that looks finished but is empty “workslop.” Two in five US desk workers said they had received it in the previous month, and each piece costs downstream colleagues time to interpret, check and redo.
The third is hidden failure. When IBM was promoting Watson for Oncology, its own medical specialists and client physicians flagged “unsafe and incorrect” treatment recommendations while the product was still being sold. MIT later studied enterprise AI and found that 95 per cent of generative-AI pilots delivered no measurable return. Dashboards show how much AI is used; they rarely show whether it helped.
Together, these create the Verification Tax: the extra human time and judgement needed to check AI before it can be trusted. Workers can see this tax before leaders, because they are usually the ones paying it.
Resistance is more specific than it looks. A Harvard Business School study found that people are much more comfortable with AI augmenting human work than replacing it: 94 per cent of respondents favoured current AI as an augmentation tool, while many remained ambivalent or morally opposed to full automation in specific professions. The same person who resists AI sending client advice on their behalf may happily use it to analyse data. That tells you where AI needs boundaries, redesign or a different use case, but only if you are listening. The best organisations build a channel to capture that signal. NatWest’s AI ethics panel uses volunteers from different grades and areas of the bank to review new AI uses before they go live. At Cisco, the internal AI assistant reached mostly technical staff at first. After the IT team rebuilt it around what employees said they actually needed, it became the company’s most widely used AI tool, serving more than 100,000 people, most of whom report saving around five hours a week. Where employees get no such channel, they build their own: Hollywood’s writers and actors won some of the first high-profile AI protections after a combined 266 strike-days across two unions.
If you override your employees instead of listening, the cost often surfaces later. Duolingo pushed an AI-first policy in 2025 and drew enough staff and public backlash that it had to soften it. A Writer/Workplace Intelligence survey reported that 29 per cent of employees who use generative AI at work said they had “sabotaged” their company’s AI strategy, including by using unapproved tools or refusing to use AI; the rate rose to 44 per cent among Gen Z.
Getting real value from AI takes shift in approach. First, track how employees feel about it and treat a fall as an early warning of problems with quality, trust or control. Edelman’s 2025 research suggested that acceptance is driven more by workplace experience and trusted peers than by top-down messaging. Second, stop scoring people on how much AI they use, and start asking whether it improved their work. Third, give staff a safe, anonymous way to raise concerns; the Post Office’s postmasters and branch staff had to start their own campaign because no such channel existed. Finally, ask every team to map where AI helps, where it makes the work worse, and where it is beside the point. The teams that push back hardest may understand the limits best.
Schmidt closed with an invitation: “The future is not yet finished. It is now your turn to shape it.” In the story of the emperor’s new clothes, the courtiers who stayed silent were complicit. The child who spoke up was the only one willing to say what everyone else had learned not to see.
Monday Morning Actions for Executives
• Add a question to your next employee engagement survey asking whether AI feels imposed rather than useful.
• Remove AI-usage volume from performance reviews. AI fluency is a reasonable competency, but usage volume is not.
• Designate a formal channel for AI dissent, using patterns already established for whistleblowing.
• Ask each function to map its AI frontier: where AI helps, where it hinders, and where it is irrelevant.
Questions for Board Members’ Back Pockets
• Do we understand our employees’ attitudes towards, and concerns about, AI?
• Are we measuring AI adoption volume or AI deployment quality?
• If an employee raised a concern about AI output quality tomorrow, what would happen in practice?
• Has our AI strategy been shaped by the people who will use it, or only by the people who will buy it?
Footnotes & Sources
• Eric Schmidt, University of Arizona commencement, 15 May 2026. Official transcript and video; also reported by NBC News, Fox Business, The Verge and Business Insider.
• NBC News/Hart Research poll, March 2026. 1,000 registered voters; ±3.1pp. Overall AI favourability: 26% positive, 46% negative, net −20; 57% said AI’s risks outweigh its benefits.
• Post Office Horizon Inquiry, Volume 1, July 2025. More than 900 postmasters and branch staff wrongfully prosecuted; failures included software defects, governance failure, legal failure, disclosure failure and institutional defensiveness.
• WalkMe/SAP, Fifth State of Digital Adoption Report, 2026. 3,750 respondents across 14 countries. 9% of workers trust AI for complex decisions vs 61% of executives.
• Writer/Workplace Intelligence report, released April 2026. Survey of 2,400 workers across the US, UK and Europe: 1,200 C-suite executives and 1,200 employees. 29% of employees said they had “sabotaged” their company’s AI strategy; 44% among Gen Z.
• BetterUp Labs/Stanford Social Media Lab “Workslop” study, published in HBR, 2025. 1,150 US desk workers. Estimated cost: $186/employee/month; 40% of workers reported receiving workslop in the prior month.
• IBM Watson for Oncology, internal documents, 2017–18. IBM’s own medical specialists and client physicians flagged “multiple examples of unsafe and incorrect” treatment recommendations; the system had been trained largely on hypothetical cases. Reported by STAT News, July 2018.
• MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, July 2025. Review of 300+ publicly disclosed AI initiatives, representatives from 52 organisations and 153 senior leaders: 95% of organisations reported no measurable P&L return from generative-AI pilots.
• Edelman Trust Barometer Flash Poll, October 2025. AI acceptance depends more on trusted workplace experience and peer influence than on top-down messaging.
• Duolingo CEO reversals, May 2025 and April 2026. AI-first posture softened; AI use removed from performance reviews. Von Ahn: “I’m not going to force you.”
• Shopify CEO memo, 7 April 2025. Tobi Lütke, reported by TechCrunch and CNBC.
• WGA/SAG-AFTRA AI provisions, 2023. A combined 266 strike-days across WGA (148) and SAG-AFTRA (118); one of the first high-profile, labour-negotiated AI governance frameworks, covering AI consent, compensation and human-only credit provisions.
• NatWest Group AI & Data Ethics (AIDE) Panel, 2025–26. Volunteers from different grades and areas review AI use cases against seven AI System Principles before deployment; Best Data Governance with AI Initiative at the 2025 DataIQ Awards.
• Cisco internal AI assistant (CIRCUIT), 2023–26. Initially used mainly by technical staff (about 30% of the workforce); after an IT-led redesign around employee needs, 100,000+ users across Cisco’s global workforce, 45M+ interactions, with 73% reporting higher productivity and ~5 hours/week saved.