The AI Productivity Paradox: Why Your Team Feels Faster But Isn't

MOUNTAIN VIEW, CA – June 10, 2026 – A significant disconnect is emerging at the heart of the modern workplace. Employees report that artificial intelligence is saving them nearly a full day and a half of work each week, yet a staggering majority of businesses are not seeing the revolutionary impact on their bottom line. This is the central finding of the inaugural Work AI Index, a landmark report from Glean’s Work AI Institute, which suggests the much-hyped AI revolution is stuck in a new kind of productivity paradox.

The comprehensive study, which surveyed 6,000 digital workers and analyzed millions of workplace AI interactions, found that while 87% of employees use AI and 75% feel more productive, a mere 13% of organizations report that AI has significantly improved performance. Workers claim AI automation saves them roughly 11 hours a week, but these gains are evaporating into a fog of hidden labor, rework, and unverified outputs, challenging the core assumption that more AI usage automatically equals more business value.

The Hidden Costs: ‘Botsitting’ and ‘Botshitting’

The report introduces two new terms to the business lexicon that pinpoint the source of this productivity drain: 'botsitting' and 'botshitting.'

‘Botsitting’ refers to the unrecognized, manual labor required to make AI usable. This includes feeding the tools necessary context, supervising their outputs, debugging errors, and cleaning up AI-generated work. According to the Work AI Index, employees spend a startling 6.4 hours per week—or 37% of their total time spent on AI—just 'botsitting.' This is more time than they spend actually using AI to perform their primary tasks (36%). This hidden work is a direct consequence of AI models that, while powerful, still require significant human oversight. Independent research corroborates this need; even the most reliable models from tech giants have documented hallucination rates, which can skyrocket for complex reasoning tasks, making unassisted output a significant business risk.

This leads to the second, more troubling behavior: ‘botshitting.’ The report defines this as the act of shipping AI-generated work that an employee has not verified, does not fully understand, or cannot confidently stand behind. A concerning 69% of AI users admit to this practice. Within that group, 41% confess to delivering work they couldn’t explain if asked, and 28% have even blamed AI for mistakes they personally caused. This creates a dangerous quality-control problem, where polished-looking outputs mask a lack of rigor and accountability.

“Too many companies are treating AI adoption like a vanity metric—more seats, more prompts, more usage,” said Dr. Rebecca Hinds, Head of the Work AI Institute at Glean. “But adoption alone doesn’t equal transformation. If employees are spending the productivity dividend on botsitting and botshitting, companies haven’t eliminated work—they’ve created a new layer of overhead.”

Tool Sprawl and the Context Crisis

At the root of these issues is a fundamental flaw in how most companies have deployed AI: a lack of context. The report found that 53% of workers say important information they need is not accessible from their AI tools. This forces them to act as a human integration layer between disconnected systems.

Today’s enterprise landscape is a patchwork of specialized software. While powerful AI assistants are now embedded in suites like Microsoft 365 and Google Workspace, their knowledge is often confined to their own digital territory. This creates a ‘tool sprawl’ where 77% of AI users find themselves juggling multiple AI tools each week, and 60% admit to rerunning the same prompt across different platforms because the first output was insufficient. This fragmentation is a primary driver of 'botsitting,' as employees waste time manually piecing together information that the AI cannot see.

This environment pushes employees toward risky behavior. The study found that workers using this kind of 'context-poor' AI are significantly more likely to feel worn out (50% vs. 18%) and to use unapproved AI tools (53% vs. 21%) in a desperate search for better results. Ironically, the most proficient AI users are often the biggest rule-breakers; 54% of 'high AI achievers' admit to using unapproved tools or using approved tools in noncompliant ways.

The Human Toll of Flawed Implementation

The push for AI adoption without the proper infrastructure is taking a toll on the workforce. The pressure to appear 'AI-busy' and leverage these new tools is creating an environment of burnout, blame, and concealment. Workers are caught between the mandate to innovate and the reality of wrestling with inadequate tools. The fatigue reported by users of context-poor AI is a clear signal that employees are bearing the brunt of poor implementation strategy.

Furthermore, the report reveals a breakdown in transparency. Among high-achieving AI users, 38% admit to downplaying AI's assistance to their manager, and 36% actively hide how much AI helps them. This behavior prevents leaders from getting an honest assessment of which tools are effective and where the true challenges lie. When organizations reward visible activity—like the number of prompts run—over tangible outcomes, they inadvertently encourage employees to generate 'botshit' to prove their engagement with the new technology, further eroding quality and trust.

The Path Forward: Redesigning Work for the AI Era

Despite the sobering findings, the report offers a clear path forward by examining a small cohort of 'transformative organizations' that are successfully harnessing AI. These companies aren't just buying more tools; they are fundamentally redesigning how work gets done.

In these leading organizations, 90% of workers say their employer provides sufficient AI training and support, and 84% say their employer formally rewards AI skills—figures that are dramatically lower in other companies (52% and 48%, respectively). Crucially, 90% of employees in these firms say their employer treats AI as an opportunity to redesign work, not just automate old processes. This involves grounding AI in deep enterprise context by unifying data from across the company's disparate systems into a single, accessible knowledge base.

Success also hinges on establishing robust governance and training employees not just on how to use AI, but when to use it and how to verify its output. By building guardrails into daily workflows, these companies empower employees to move faster without sacrificing quality. They treat workarounds not as violations, but as valuable signals that official tools are falling short.

As Dr. Hinds concluded, “The next phase of enterprise AI will not be won by the companies that buy the most tools or drive the highest usage numbers. It will be won by the companies that make AI part of how work actually gets done, grounded in the right context, measured against real outcomes, and governed in a way that helps employees move faster without lowering the bar for quality.”