In brief: Over half of the time saved by generative AI flows back into verification and correction of AI outputs, resulting in hidden control overhead and employee demotivation.
Generative AI saves employees approximately eleven hours per week – yet over half of this time flows back into verification and correction of AI outputs. The phenomenon is termed “bot-sitting” and leads to demotivation and increased attrition among qualified professionals.
A study by the Work AI Institute (research unit of AI provider Glean) documents a previously underestimated effect in the integration of generative AI: employees invest considerable time making AI systems usable in the first place. They supply missing context, review outputs, correct errors, and perform necessary post-processing. The calculations are sobering: of the approximately eleven hours saved per week, over half flows directly back into this verification process.
The reason lies in how generative language models work. They produce content that appears superficially convincing but may be incomplete or factually incorrect. Employees must therefore verify and validate each output before passing it on – an additional burden in both cases: those who work carefully bear the control load themselves; those who pass on unreviewed results shift the risk to downstream process steps. There is also a perceptual bias: a study by research organization METR involving 2025 experienced open-source developers found that these developers took an average of 19 percent longer with AI tools but subjectively believed they saved approximately 25 percent of their time.
For CTOs, this phenomenon is critical because qualified professionals who remain stuck in this verification function are significantly more likely to consider changing jobs. The original AI promise – freeing employees from routine so they can focus on more demanding tasks – is thus undermined. Instead, the workforce trades one form of monotony for another.
However, not all verification is wasted time. The critical distinction lies in competence: those who can judge what a model can do and where it fails create genuine value – for example, through targeted verification of consequential results or by bringing in expert knowledge the model lacks. Tasks with legal or financial implications must remain in human hands, as must final quality assurance. A language model can draft a quarterly report; the responsibility for the accuracy of the figures ultimately rests with a person. Companies should strategically deploy qualified employees where their judgment is genuinely needed – with focus on analytical thinking, subject-matter expertise, and critical information assessment.
Source: www.it-daily.net · Published 19 July 2026
Lumi AI News — AI-assisted curation in accordance with Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.