The hidden cost of enterprise AI and why teams keep babysitting bots

AI adoption increases hidden human labor costs

AI is everywhere now. It’s becoming the silent partner in almost every digital workflow. The problem is that the efficiency looks better than it really is. Many employees spend real time doing what the machines can’t, feeding data, checking errors, verifying context, and stitching together results from different systems. This “botsitting” work, as Glean’s Work AI Institute calls it, adds invisible labor that isn’t measured in performance reports but shows up in employee fatigue and slower cycles of delivery.

A recent survey of 6,000 full-time digital workers by Glean’s Work AI Institute found that 87% are using AI. It’s automating about a quarter of their tasks and saving roughly 11 hours each week. Yet they also waste around 6.4 hours per week maintaining or correcting AI output. That means almost half of the time gained is pulled back into supporting the system itself. For companies seeking real ROI on digital transformation, this is core to productivity strategy.

Rebecca Hinds, Head of the Work AI Institute, described this as “a vicious cycle that feeds itself.” She’s right. Every improvement in AI performance demands additional human oversight, data tuning, error correction, and context input. Until organizations redesign workflows that take this human overhead into account, AI deployment will feel faster but actually move slower in real terms.

Executives should look closely at how much invisible work AI generates across teams. Recognizing and quantifying that hidden labor is the first step in building smarter systems that truly deliver productivity.

Fragmented tools and lack of enterprise context fuel frustration

Most large language models were trained on general internet data. That’s useful for broad knowledge but limited for enterprise accuracy. They rarely know company-specific terminology, internal products, or local workflows. Employees must repeatedly fill in these gaps, explaining and re-explaining context with every new prompt. Multiply that across several unconnected tools, and what should be “automation” turns into extra work with diminishing returns.

Rebecca Hinds noted that workers often feel frustrated because “the tools don’t understand enough about day-to-day work to be useful.” They end up rewriting prompts for different platforms and manually checking inconsistent answers. Much of this friction comes from poor integration, a lack of unified data environments and standard APIs. It’s not about the AI’s intelligence level; it’s about how well it talks to the rest of your business systems.

For decision-makers, this should signal a design flaw. Deploying multiple disconnected tools without centralized governance or shared context simply shifts complexity from the system to the employee. The smarter move is investing in tightly integrated AI ecosystems that understand enterprise-specific data and communicate across departments seamlessly.

By addressing fragmentation and pushing for more connected architectures, companies can reduce repetitive input, lower frustration, and make AI an actual productivity amplifier instead of a new layer of hidden labor.

Unverified AI use (“Botshitting”) endangers output quality

Speed is important, but not at the cost of trust. Many employees are pushing out AI-generated work without proper verification, a behavior Glean’s Work AI Institute calls “botshitting.” It’s often the result of pressure to deliver faster or manage too many AI tools at once. The risk here is simple: when accuracy drops, credibility follows.

According to Glean’s survey, 69% of users admitted to shipping unverified AI outputs. Forty-one percent said they’ve sent work they couldn’t fully explain, and 28% have blamed mistakes on the AI itself. These numbers show what happens when verification time is sacrificed for speed. The issue isn’t a lack of capability, it’s a lack of structure around how AI is used and monitored.

Rebecca Hinds, Head of Glean’s Work AI Institute, warned that “botshitting is offloading your critical human thinking, judgment, and understanding.” That’s a problem. The human layer is still the last checkpoint before anything reaches a client, a regulator, or the market. Letting that checkpoint fail turns AI from a strength into a liability.

Executives should build operational standards that define when AI output is acceptable and how it must be reviewed. Setting these boundaries protects both brand and worker trust. Verification must be seen as part of productivity. Quality control in the AI era depends on disciplined human oversight, measured.

Psychological and organizational factors shape AI usage disclosure

AI’s cultural impact inside companies isn’t just technical, it’s social. Many employees now rely more on AI tools for day-to-day support than they do on their managers. That’s not necessarily negative, but it creates tension around visibility. Some workers hide their AI use, afraid it might signal laziness or make their role seem replaceable. Others overstate how much they use AI to seem more advanced. Both behaviors point to one thing: uncertainty about how organizations truly value AI adoption.

Data from Glean’s survey shows that more than half of workers receive more daily help from AI than from their managers. Among high AI performers, 54% admit to using unapproved or noncompliant tools, while 36% hide how much they depend on AI. These trends reveal an undercurrent of anxiety about acceptance and safety in AI use. Psychological safety, knowing that using AI transparently won’t harm one’s standing, is still missing in many organizations.

Rebecca Hinds observed that organizational context determines how open people are about AI usage. Without clear support from leadership, workers will continue to operate in the gray area between innovation and compliance.

For executives, the solution is transparency. Make clear that AI fluency is valued, but accountability still matters. Encourage open discussion around experimentation, failure, and success. When leaders set this tone, employees don’t need to hide how they use AI, they’ll share it, refine it, and drive meaningful progress across the company.

Leading organizations redesign work around AI

The top-performing companies know that success with AI comes from design. They don’t focus on using AI more; they focus on using it better. That means redefining what good work looks like and training teams to make thoughtful decisions about when, and when not, to use AI. This approach creates durable performance gains instead of short-term boosts.

The most successful organizations in Glean’s study are spending more of their AI time on contextual tasks: defining quality, guiding judgment, and structuring decisions before delegating them to systems. They’re not trying to replace human insight; they’re enhancing it with precision tools. These companies invest in training programs and transparent governance, ensuring employees know how to use AI responsibly and effectively.

Rebecca Hinds, Head of the Work AI Institute at Glean, pointed out that real impact doesn’t come from “clicks of the tool, not just tokens used, but real skills, real learning.” That mindset multiplies capability. It’s visible in organizations where executives set the tone, using AI themselves, addressing failures openly, and normalizing iteration as a growth process.

From a leadership perspective, the takeaway is straightforward. Build governance that evolves with the technology. Track the impact of AI using established KPIs such as quality, efficiency, and engagement rather than vanity metrics like usage rates. Make data accessible to the people doing the work, not just the people monitoring it. The fastest path to value is clarity of purpose, supported by transparency and consistent feedback.

AI as a preferred learning and collaboration partner

Something new is happening in how people use AI at work. It’s becoming a learning partner, something employees turn to for guidance before they approach formal training sources. This behavior signals a major shift in workplace development philosophy: employees now expect AI to help them improve as they work.

Glean’s research found that workers increasingly prefer to use AI for coaching and skill-building tasks rather than traditional learning channels. They value its immediacy and the low friction it provides. Tools with low-code or no-code interfaces are accelerating this trend, giving employees direct access to problem solving and structured learning within their everyday workflows.

Rebecca Hinds highlighted that this shift is “starkly different from what we’ve seen with previous technologies.” It shows a new level of comfort with machine-guided growth and an increased expectation that digital systems should teach as well as execute.

For executives, this represents both opportunity and responsibility. Integrating learning directly into workflow environments can drive massive upskilling at scale, but it requires intentional design. AI should be developed and deployed as part of a long-term human development ecosystem. When companies invest in that integration, they don’t just get more efficient teams, they get smarter ones that continuously evolve alongside the technology.

Key takeaways for decision-makers

• Hidden labor undermines AI efficiency: AI boosts speed on paper but adds invisible human labor through constant oversight and correction. Leaders should quantify this hidden cost and redesign workflows to capture true productivity gains.
• Tool fragmentation drains focus: Workers lose time switching between unconnected AI tools and re-entering context. Executives should consolidate platforms and ensure AI systems are seamlessly integrated with enterprise data.
• Unverified outputs risk credibility: Many employees ship AI-generated work without proper checks, threatening accuracy and trust. Leaders should enforce clear review standards and position verification as a core productivity step.
• Cultural dynamics influence AI use: Employees often fear showing or hiding AI use due to unclear expectations. Executives must build transparency and psychological safety around experimentation to drive confident adoption.
• Redesigning work drives real AI value: Top organizations use AI to redefine processes. Leadership should prioritize training, adaptable governance, and executive modeling to scale sustainable innovation.
• AI becomes a continuous learning partner: Workers are turning to AI for education and support more than traditional channels. Businesses should embed AI-driven learning tools into workflows to accelerate skill development and adaptability.