Survey shows a drop in AI deception but confidence soars
At a recent SAP conference in Madrid, Dan Adika, co-founder and chief executive of digital-adoption platform WalkMe, asked attendees to raise a hand if they were using artificial intelligence in a meaningful way. Only two executives responded, far fewer than the 8% he expected from a Gartner survey.
Adika's anecdote frames the findings of WalkMe's third annual AI at Work Pulse Survey, carried out by Propeller Insights among 2,037 working adults in the United States. The study reports that the proportion of workers who admit to pretending to understand AI in a meeting fell from 45.2% in 2025 to 28.3% in 2026. Likewise, those who say they have passed off AI-generated output as their own dropped from 48.7% to 32.5%.
Why confidence may be misplaced
Despite the decline in self-reported deception, 90% of respondents say they feel confident using AI, while only 24.6% believe it works on the first try. Half of those surveyed say they spend more time coaxing AI to perform a task than they would have spent doing it manually.
"Workers are not becoming more deceptive; they are becoming more comfortable," the survey states.
Adika argues that this comfort is mathematically indefensible when examined against company profit-and-loss statements. He estimates that a firm with 50,000 employees should save roughly 200,000 hours a month if AI delivered on its promises, yet the expected multi-million-dollar savings have not materialised.
He also points to a double standard: organisations expect AI to be near perfect, while tolerating a higher error rate from humans. This mismatch fuels resistance to AI-driven decisions.
Impact on hierarchy and job security
More than half of employees feel that senior leaders do not fully grasp the AI strategies they publicly champion. KJ Kusch, WalkMe's global field chief technology officer, notes that everyone from new hires to executives is learning AI in real time.
Adika warns that the current model of building a "company brain", a shared memory layer that lets AI act like a well-trained employee, could jeopardise jobs. He explains that a manager who once needed a team of ten could, after training an AI on that team's collective knowledge, operate with just three people, leaving the remaining staff vulnerable to redundancy.
Generational differences and the road ahead
The survey highlights a stark generational divide. Gen Z workers are the most confident, with 94.1% saying they feel comfortable using AI, but they are also the most likely to have overstated their skills, 45% admitted to doing so, compared with 13% of baby-boomers. Moreover, 31% of Gen Z respondents said their over-confidence led to a real workplace problem, versus 7% of older workers.
When asked about the future, Adika offers no grand vision. He suggests that without a major breakthrough, improvements will be incremental and the workforce will continue to "muddle through".
What next for AI in the workplace?
Adika believes that more training alone will not close the confidence gap, as overconfident employees are unlikely to enrol in courses they think they already master. Instead, he calls for AI onboarding that mirrors human onboarding, a longer, more attentive process.
WalkMe respondents indicated that better integration of AI into existing applications (33.7%) and built-in guidance (30.2%) would be more helpful than standalone training.
In the meantime, organisations must reconcile the gap between perceived AI competence and actual business outcomes, and consider the ethical implications of turning employee expertise into replaceable algorithms.