The Financial Times recently reported an interview with Rebecca Hinds, head of the Work AI Institute. Hinds was discussing a new survey of 6,000 digital workers, which included the following arresting statistic: although respondents claimed that AI saved them 11 hours a week on average, only 13% reported any improvement in company performance.
Hinds offers three explanations for this paradoxical result:
- When calculating the 11 hours saved, workers aren’t counting all of the time they spend waiting for AI agents to complete tasks (an activity some are now calling “botsitting”).
- The workers often ignore the cost of toggling between multiple AI tools as they attempt to get a usable response (60% of the sample reported running queries across several tools in search of better outputs).
- Workers may also be participating in what Hinds calls “workplace theater,” in which they are “visibly performing work for bosses and colleagues, rather than focusing on the actual grind of getting things done.”
These findings are probably not what companies like OpenAI and Anthropic want to hear just months before their planned IPOs. But what interests me here is not the shortcomings of AI technology.
What caught my attention about this interview is that essentially all of these issues are discussed in my book Slow Productivity, which came out in early 2024, and doesn’t mention AI at all. An earlier generation of digital tools – like email, Slack, video conferencing, and mobile computing – also led workers to vastly underestimate the time wasted wrangling diverse devices, applications, and rapidly toggling back and forth between different tasks and channels. Workplace theater isn’t new either (in my book, I called it “pseudo-productivity”).
In this way, AI isn’t so much creating new problems as it is magnifying the types of problems that have long existed. Here I find a potential silver lining. This technology is sufficiently new and exciting that business leaders are paying more attention to its impacts. In seeking to understand how to make AI effective in the workplace, they might finally recognize what has long been broken.
I broadly agree with your argument. The findings in The Work AI Index suggest that AI is not creating a new problem so much as exposing an old one: organizations have long underestimated the coordination, context-sharing, and judgment required to turn individual productivity gains into system-level performance improvements.
What struck me most is the gap between perceived individual productivity and organizational outcomes. From a systems perspective, that’s exactly what we would expect when local optimization outpaces improvements in the overall workflow. Faster production of outputs does not automatically translate into better results if verification, coordination, and rework increase at the same time.
That said, I would be cautious about treating the report as definitive evidence. Much of the data is self-reported, which makes it vulnerable to recall bias and perception effects. The concepts of “botsitting” and especially “botshitting” are compelling, but they are also broad constructs measured through survey responses rather than objective observations of work. In addition, the report is sponsored by a company whose business is centered on solving context and knowledge-access problems, so there is a natural incentive to frame “lack of context” as the central bottleneck.
Even with those limitations, I think the core insight holds: AI is less a replacement for organizational capability than a multiplier of it. In well-designed systems it amplifies effectiveness. In poorly designed systems it amplifies confusion, rework, and coordination costs. The technology may be new, but the underlying challenge is an old one: designing work that allows information, judgment, and responsibility to flow effectively across the system.
I believe it’s naive to think people act rationally, analyze and recognize wherein lies the problem. Instead they are rather be guided by emotions, dopamine-driven reactions and herd psychology.
Re: “workplace theater” see the late David Graeber’s 2018 book “Bullsh*t Jobs: A Theory”.
Office work is broken because workers must spend 8-9 hours at their desk regardless of efficiency. We see this a lot with people who still don’t know how to efficiently use desktop productivity suites (Microsoft Office); one worker who knows all the keyboard shortcuts and tools to save time and avoid manually reproducing things may finish a similar task two to three times faster than a colleague that doesn’t know how to do these things.
Worker A (efficient) doesn’t get to go home sooner if the task is finished faster. They may get promoted faster if the workplace is meritocratic. Worker B (inefficient) would likely see the largest potential time gains from leveraging AI agents, but it’s unlikely that a worker who uses the right-click context menu to copy and paste in 2026 will effectively implement AI, and they might spend even more time trying to fight the AI to get it to do what they want than just doing the task themselves. Economically, the company cannot be more productive if they only have one task per day for each worker to do, and they could have fired Worker B years ago and just had Worker A do two tasks per day if they actually cared about efficiency, but they didn’t, and now we’re supposed to believe *both* workers will be fired in a couple short years and replaced with AI.
Little discussed is the inefficiencies added, let’s call it “productivity drag” by the way some workers are using AI. Example: a coworker uses AI to write an email. They input 4 bullet points into the AI. The AI outputs 4 paragraphs. The recipient now has to spend more time hunting through textual fluff for the actual information and action items, than they would have had to spend had the coworker just emailed them the bulleted list they fed the AI.
Honestly I’ve found the vast majority of places where AI was used at work, at best its impact is neutral, but far more often creates this type of productivity drag or even resentment. Is it fair that I should have to spend my human effort reading and replying to an AI-generated email?
The productivity drag doesn’t get mentioned as much as it should! It’s everywhere – half-baked ideas and POCs come across people’s desks every day. Besides the increased volume, engaging with AI generated text that has not been reviewed by the original author, creates an extra cognitive load on the reader (as it has been shown by studies looking at readibility scores of AI text).
I interpret this as a snapshot of our current moment.
AI productivity reporting generally does not mention the time horizon for how AI-skill attainment intersects with AI-product innovation. What happens when the longer task that requires multi-AI tools and waiting for outputs, becomes one tool with near-zero latency: will the same skills the workforce is employing with low productivity reports suddenly become productivity multipliers?
One could hypothesize that as AI-product innovation increases (reduced latency, increased accuracy, more agentic workflows), the firms who currently do not report improvements in company performance may skyrocket, as their workforce is AI-enabled, experienced, and skilled. Even the productivity floor of “workplace theatre” will likely increase relative to pre-AI levels. In this way, these practices *would be* highly correlated with long-run increased firm performance.
I think that “botsitting” and multi-AI platform toggling is not so different from waiting on an email response to take action on a particular task. I think it would be interesting to look at these same organizations and take a look at the change in overhead tax or end-to-end project completion timelines to capture systemic drag on organizational performance.
This reminds me of the WWII bomber survivorship bias example.* We shouldn’t reinforce the damaged areas of planes that return, we should reinforce the areas that are not damaged.
These AI productivity snapshots are similar to the damaged areas. I think we should look, holding all else constant, at how workplace performance will be impacted when the “undamaged” areas – the invisible, critical drivers – are addressed, such as increased AI systems’ effectiveness, hyper-reduced latency, and upskilling optimization.
What I am wondering is how this AI innovation interacts with deep work from your perspective. At an enterprise level, how do we intentionally carve out space to increase deep work enablement in the workforce when AI takes on high volume, admin heavy tasks? Will someone employing AI in deep work effectively accomplish more or less deep work than someone without employing AI?
*See Wikipedia page if unfamiliar, the website would not let me hyperlink