Why do workers feel confident with AI but get poor results?

WalkMe's AI at Work Pulse Survey, fielded June 10-20, 2026 by Propeller Insights among 2,037 working U.S. adults who use AI on the job, finds that "90% of employees feel confident using AI," but "only 24.6% say it works on the first try, and 50.2% say they have spent more time trying to get AI to do a task than the task itself would have taken manually." The report names this gap the "AI confidence trap": workers feel fluent, but the results frequently don't match. Confidence is highest, and most overstated, among the youngest workers — Gen Z self-reports 94.1% confidence and a 45% rate of "pretended to be more skilled with AI than they actually are," versus 13% for Baby Boomers, and a resulting work problem (a mistake, missed deadline, bad decision, or lost trust) for 31% of Gen Z versus 7% of Boomers.

WalkMe reads the gap as an adoption-support failure rather than a worsening honesty problem: year-over-year, "the share of employees who say they have pretended to know AI in a meeting dropped from 45.2% in 2025 to 28.3% in 2026," and passing off AI work as one's own fell from 48.7% to 32.5%. Its CTO KJ Kusch frames it as workers "becoming more comfortable," not more deceptive, while tools still fail to "meet people where the work actually happens." The report also cites two corroborating studies — SAP/Oxford Economics' Value of AI Report 2026 (2,600 business leaders, 13 countries), where 79% report rework, delays, or backlogs from low-quality AI output; and WalkMe's own State of Digital Adoption 2026 (3,750 enterprise workers/executives, 14 countries), where only 12% are fully confident AI understands their work's context and 9% trust it for high-impact decisions. Asked what would help, employees rank in-tool guidance and integration (33.7% and 30.2%) above separate training.

This is population-scale, self-reported evidence of the same split that Can self-ratings replace objective performance scores for AI competence? finds in a pooled correlation of literacy instruments: a confidence measure and a performance measure are "targeting different things." The WalkMe numbers don't replicate that finding directly — "works on the first try" is still a self-report, not a scored task — but the same split shows up at the level employees experience it: feeling capable with AI and getting a usable result on the first pass are not the same thing, and the gap between them is where the 50.2% "spent more time" figure originates.

The excerpt does not establish that AI use itself caused the lost time, the missed-deadline incidents, or managers' rising output expectations (51% report the latter) — all figures here are self-reported perceptions from a single pulse survey, not task logs or output audits, and the sample is restricted to workers who already use AI on the job, not a workforce-wide baseline. WalkMe sells an AI adoption and in-app guidance platform, and its headline recommendation — build guidance into the tools people already use — is also the product its survey implicitly argues for; that commercial interest doesn't make the confidence/first-try gap untrue, but it should discount how strongly the "in-tool guidance over training" conclusion is treated as disinterested.