The workforce reality

AI saves 11 hours a week and takes 6.4 of them back in botsitting: feeding context, supervising outputs, debugging errors. Beneath the ledger, three slower erosions: knowledge decay, organizational amnesia and the training gap.

Redesign, or friction

Nearly half of organizations introduced AI without redesigning the workflows or roles it sits within, and impact at scale depends on exactly that redesign. Middle management is where roles shift the most, towards orchestrating and enabling both human and agent contributions.

Botsitting

A survey of 6000 digital workers reveals that the productivity story is more complicated than the headlines suggest. AI saves roughly 11 hours per week, but employees lose 6.4 of those hours managing the technology itself. This means feeding it context, supervising outputs and debugging errors. What is even worse is that 69% of workers admit shipping unverified AI-generated work because oversight demands have become overwhelming. This is the still unrecognized labor of "botsitting". To combat this phenomenon, high-performing organizations do not adopt heavier AI usage; instead they invest deliberately in defining quality standards, building human judgment and knowing when not to deploy agents at all.

Knowledge decay

Academics and analysts have warned that unchecked AI adoption creates "knowledge decay": repeated AI summarization and synthesis gradually degrades the original judgment, context and expertise organizations depend on. Three failure modes accelerate this. Verification becomes costly when employees must disentangle authentic reasoning from AI-generated errors. Validation breaks down when organizations can no longer confirm genuine human expertise contributed to a deliverable. Knowledge entropy compounds as content passed repeatedly through LLMs drifts away from ground truth.

Organizational amnesia

Organizational amnesia complements knowledge decay: headcount reductions outpace institutional knowledge transfer at the organizational level, causing AI systems to operate confidently but incorrectly because they lack the contextual understanding that departed along with the impacted employees. Having clean data, capable models and strong infrastructure means nothing without "context intelligence": a machine-interpretable representation of how the business actually works.

The training gap

Lastly, the training gap. Only 40% of employees using generative AI at work say their companies provide training. The remaining 60% are, in effect, deploying the tools and hoping for the best. At the same time, organizations spend heavily on training models while largely neglecting the humans operating alongside them. McKinsey projects 75% of roles may require significant redesign in the symbiotic enterprise, because 60% of work hours are now theoretically automatable. Three emerging role categories are already forming: integrative supervisors orchestrating human-agent teams, deep specialists that agents cannot replace, and AI-enabled augmented operators.