The realities of AI at work: Recap of the Work AI Index Report and why AI productivity gains are failing to reach the organisation
The AI productivity boom that isn’t showing up in business performance
Artificial Intelligence has arrived in the workplace at a remarkable speed. According to The Work AI Index: Global from the Work AI Institute, 87% of digital workers now use AI at work, and 75% believe it makes them more productive. Workers estimate that AI automation saves them around 11 hours per week.
This sounds great for the AI advocates out there, yet there is a striking contradiction at the heart of the research: despite widespread adoption and significant personal productivity gains, only 13% of employees believe their organisation is performing significantly better as a result.
This finding should give every business leader pause.
For the past two years, many organisations have focused their AI strategies on adoption metrics, licences, pilot programmes and tool deployment. The assumption has been relatively simple: more AI usage equals more business value.
The evidence suggests otherwise.
The report argues that the missing piece is not technology but work design. Organisations have underestimated the hidden human effort required to make AI productive, trustworthy and useful in real business environments. As a result, much of the value supposedly created by AI is being absorbed by an invisible layer of workt hat few leaders are measuring and almost nobody is managing.
The researchers give this hidden effort a name: "botsitting."
The hidden cost of AI: Botsitting
Botsitting refers to the labour involved in making AI outputs usable.
It includes:
- Providing missing context
- Correcting inaccurate outputs
- Re-running prompts
- Validating answers
- Debugging AI-generated work
- Cleaning up downstream errors
The average digital worker spends 6.4 hours every week performing this activity—almost an entire working day.
Perhaps more surprisingly, workers spend more time botsitting than they do using AI to generate work products.
The report found that:
- 37% of AI-related time is spent botsitting
- 36% is spent actually using AI
- 27% is spent learning AI tools and building agents
This finding challenges one of the most common assumptions about AI adoption: that time saved automatically translates into value created. In reality, many organisations have simply shifted effort rather than eliminated it.
Workers increasingly act as the integration layer between disconnected tools, fragmented datasets and incomplete AI outputs. The technology may produce an answer in seconds, but employees must still verify whether it is accurate, relevant and safe to use.
The result is a hidden productivity tax that rarely appears in business cases or ROI calculations.
When "Botsitting" becomes "Botshitting"
The report introduces a second concept that leaders should pay serious attention to: "botshitting"(you got that right – it’s not a typo).
Botshitting occurs when employees deliver AI-generated work that they:
- Have not properly verified
- Do not fully understand
- Cannot confidently defend
The research found that 69% of AI users admit to at least one botshitting behaviour.
Examples include:
- Delivering work they could not explain if challenged
- Using unapproved tools
- Violating AI usage policies
- Blaming AI for mistakes that were actually their own
- Knowing an output is flawed but using it anyway
This behaviour rarely arises from malicious intent. Instead, it emerges because workers become overwhelmed by the growing burden of botsitting. Eventually deadlines, workload pressures and fatigue encourage people to accept outputs that are merely "good enough." This creates a dangerous organisational cycle:
- More AI is introduced.
- More botsitting is required.
- Workers become fatigued.
- Verification decreases.
- Poor-quality outputs move downstream.
- Additional cleanup and rework become necessary.
- Organisations deploy even more AI in response.
Rather than creating efficiency, the organisation becomes trapped in a loop of hidden labour and downstream correction.
For professional services firms, financial services organisations, healthcare providers and other knowledge-intensive businesses, this should be a significant concern. The greatest risks are not necessarily AI hallucinations themselves, but the gradual erosion of human judgement and accountability around them.
The three AI paradoxes every executive should understand
The report identifies three fundamental paradoxes explaining why AI adoption is not automatically translating into organisational performance.
1. The Productivity Paradox
AI undoubtedly improves individual productivity.
However, organisations continue to underestimate the coordination effort required to convert individual gains into collective outcomes.
Work produced faster still needs reviewing, integrating, validating and coordinating across teams.
The result is that local efficiency gains are often neutralised by organisational friction.
2. The Judgement Paradox
Historically, poor-quality work often looked poor.
Bad writing, awkward phrasing and inconsistent thinking acted as warning signals that triggered review and challenge.
AI removes many of these signals.
Outputs now appear polished, coherent and professional—even when they contain factual errors, flawed assumptions or missing context.
As AI quality improves, human oversight becomes more important while simultaneously becoming less likely.
3. The Ownership Paradox
Workers increasingly fear that AI could replace parts of their role.
Ironically, those most concerned about automation are often among the heaviest users of AI.
Over time this can weaken ownership of work. Employees begin focusing on appearingAI-enabled rather than demonstrating expertise and judgement.
The report found that more than half of workers believe AI has already automated meaningful aspects of their job that they would have preferred to keep themselves.
For organisations dependent on expertise, creativity and professional judgement, this should raise important questions about long-term capability development.
What separates organisations that are winning with AI?
Perhaps the most valuable insight from the study is that leading organisations are not necessarily using more AI.
They are managing AI differently.
The organisations reporting transformational impact consistently focus on what the authors call the "human infrastructure of AI."
They focus on work design before tool selection
Rather than starting with vendor offerings, they begin by understanding where work gets stuck, where friction exists and where value can genuinely be created.
They prioritise context
One of the strongest findings in the report relates to context.
More than half of workers say critical information needed to perform their jobs remains inaccessible to AI systems.
The most successful organisations recognise that connecting AI to data is not the same as giving it meaningful context.
Employees working with context-rich AI systems were:
- Less likely to experience burnout
- Less likely to produce unexplainable outputs
- Less likely to use unauthorised tools
- Less likely to engage in botshitting behaviour
They measure quality, not just activity
Many organisations still measure AI success through adoption metrics:
- Token volume
- Tool usage
- Logins
- Generated outputs
The report strongly warns against this approach.
Transformative organisations measure:
- Work quality
- Productivity
- Time saved
- Employee engagement
- Skill development
- Business outcomes
In other words, they measure whether work is better – not simply whether more AI is being used.
They invest in people as much as technology
Successful organisations actively:
- Reward AI skills
- Recognise experimentation
- Provide training
- Improve AI literacy
- Encourage judgement and oversight
Most importantly, they help employees learn a skill that may become increasingly valuable in the AI era: knowing when not to use AI.
What this means for Business Leaders
The central message of this research is clear. AI transformation is not primarily a technology challenge. It is an operating model challenge.
Too many organisations are pursuing AI through procurement, licensing and deployment while neglecting the fundamental redesign of work required to unlock lasting value.
The businesses that pull ahead will not necessarily be those with the most AI tools. They will be those that:
- Design work around human-machine collaboration
- Preserve judgement and accountability
- Reduce context-switching and tool sprawl
- Ground AI in enterprise context
- Measure outcomes rather than activity
- Invest heavily in workforce capability
At Enfuse, we believe this reflects a broader truth about modern transformation. Technology rarely creates competitive advantage on its own. Advantage comes from how organisations redesign processes, decision-making and ways of working around that technology. The organisations creating sustainable value from AI are building operational intelligence, not simply deploying intelligent tools. And that may be the most important take away from this research.
The future belongs not to organisations with the most AI, but to those that build the human infrastructure required to make AI genuinely useful.
Related posts
Dive in more with our related insights.
