AI-powered support steals more than 6 hours per week from employees

AI now automates more than a quarter of digital work. However, employees spend hours managing the technology, creating new bottlenecks.

As AI is increasingly integrated into everyday work life, a new productivity paradox is emerging: Although the technology speeds up processes, it also places additional burdens on employees: they must provide context, verify results, and often repeat the process multiple times across different tools.

Botsitting and botshitting

According to a new survey by Glean's Work AI Institute of 6,000 full-time employees, this leads to...digital Section on two new behaviors:

“Botsitting”, which is all the mostly invisible work to make AI usable in the first place;

“Botshitting” is the delivery of AI-generated results that are not verified, not well understood, or perhaps even unreliable.

The report was jointly authored by experts from the Work AI Institute, Emory University, Stanford University, UC Berkeley, UC Santa Barbara, UNC Charlotte, University College London, and the University of Notre Dame.

More AI use – but also more frustration

There is little doubt that AI is rapidly becoming a central team member in the workplace. According to the study, 87 percent of [unclear] already use it.digitalAI for employees. The technology already automates more than a quarter of their work, saving them about eleven hours per week.

Nevertheless, only 13 percent of respondents stated that AI use has significantly improved their company's performance. Instead, the time saved is negated by the very technology that makes it possible in the first place.

According to this, employees lose an average of about a third of their working week (6.4 hours) due to "botsitting".

Regarding the input of contextual data, large language models (LLMs) are trained on the vast corpus of the internet, but not always on company-specific data. Therefore, employees often need to provide additional information about their company's products, customers, services, or other details.

Many people get frustrated when the tools do not understand their daily work well enough to be truly helpful. Since multiple systems are often used, the same inputs have to be reformulated repeatedly.

It is exhausting for employees to do this work and, at the same time, experience that it is often neither recognized nor rewarded.

When control becomes a burden

Another problem is that employees often have to correct AI-generated results that appear professional and convincing but, in reality, contain errors, are incomplete, or lack important context. Debugging is particularly burdensome because it is often performed by people who were not involved in the original creation. These individuals then have to familiarize themselves with the system painstakingly.

The situation becomes problematic when the workload becomes so great that it leads to "botshitting": due to a lack of time, employees deliver AI-generated results without adequately checking them.

As part of the study

• 69 percent of users admitted to having already done this;
• 41 percent stated that they sometimes submit work that they cannot explain when asked;
• Another 28 percent blamed AI errors, even though they themselves were responsible.

Botshitting means outsourcing your critical human thinking, judgment, and understanding. These are precisely the tasks that must remain with humans.

According to the researcher, this behavior occurs particularly frequently among individuals who work with multiple AI agents. Due to their high scalability, errors could quickly spiral out of control if appropriate controls and permissions are lacking.

The negative effects often only become apparent after three, four, or five steps. Then, extensive cleanup and detective work are needed to find out where the agent went wrong.

Use AI… but not too much

Interestingly, more than half of the respondents report receiving more support from AI in their daily work than from their superiors. Furthermore, they often find collaborating with AI easier than collaborating with people.

At the same time, many face a dilemma about disclosing their use of AI. 54 percent of AI users who consider themselves particularly successful use unauthorized tools or misuse approved tools. 36 percent conceal the extent to which AI contributes to their performance.

According to Hinds, this depends heavily on company culture and the psychological sense of security. "In many organizations, there is massive pressure to demonstrate AI competence and show that you are a power user," she explains. On the other hand, some employees want to avoid their performance appearing too dependent on AI and having their own value questioned.

What successful organizations do differently

According to the report, leading companies are not distinguished by spending more time on AI itself. Rather, they invest more time in AI-related work. This includes the fact that they

• define the context,
• define quality standards;
• develop judgment, and
• decide which tasks should not be delegated to a model in the first place.

The companies with the greatest transformation potential are proactively addressing AI challenges: They offer training and support, view AI as an opportunity to redesign work, and officially reward AI skills.

One of the most difficult skills is knowing when not to use AI. It is not just about clicks or tokens consumed, but about real skills and real learning.

Furthermore, successful companies clearly communicate their AI strategy and explain its rationale. It should not be a static set of rules, but must be continuously developed further.

This needs to happen at every level, including top management. It is about seeing how leaders are using the technology and sharing both the success stories and the failures.

Successful companies also actively use key performance indicators. They measure quality, efficiency, and employee engagement in various ways and provide employees with valuable feedback to help them assess their own acceptance and success.

Employees are increasingly using AI as a learning tool themselves and preferring it to other learning channels. This underscores the importance of low-code and no-code tools with a flat learning curve and organizational context, which are directly embedded in workflows. This is significantly different from what we have seen with previous technologies.

Share this post