What Happened
The Work AI Index, published by Glean's Work AI Institute - a research collaborative involving academics from Stanford University, UC Berkeley and Harvard - surveyed 6,000 full-time digital workers across the US, UK and Australia, including 1,500 in Australia. The findings were reported by SMBtech on 13 June 2026.
The headline finding is stark: 90 per cent of Australian digital workers now use AI at work, compared with 84 per cent in the US. But only 10 per cent say AI has significantly improved their organisation's performance, behind both the US at 12 per cent and the UK at 18 per cent.
Australian workers report that AI already automates 27 per cent of their work output and expect that figure to rise to 34 per cent over the next year. The gap between those two numbers - between what AI is doing now and what workers expect it to do soon - points to real optimism. The performance data suggests that optimism is running well ahead of reality.
Why It Matters
The report identifies what it calls a new productivity paradox. AI is making individual tasks feel faster, but the cumulative burden of supplying context, checking quality and stitching together disconnected tools is falling on employees in ways that are not being recognised or accounted for.
Workers spend an average of 6.5 hours per week on what the report terms "botsitting" - the unrecognised effort of prompting, correcting and redoing AI-generated work. Roughly four in ten say AI sessions fail outright, requiring a restart, substantial rework or a reset back to zero.
That is not a minor friction cost. At 6.5 hours per week, botsitting consumes close to a full working day. For organisations paying for AI tool licences and expecting a return, that hidden overhead is eating the margin.
The fatigue numbers compound the picture. Forty-three per cent of Australian workers report feeling worn out by AI tools, compared with 33 per cent in the US. Australian workers are using AI more, trusting it more uncritically, and burning out on it faster - a combination that should concern any business leader who assumed adoption alone would drive results.
Key Details
The quality-control data is where the report gets uncomfortable. Seventy-seven per cent of Australian AI users admit to at least one unchecked AI-output behaviour - a higher rate than both the UK at 70 per cent and the US at 64 per cent. And 36 per cent have blamed AI for a mistake that was actually their own.
Seventy-three per cent of Australian workers have corrected or redone AI-assisted work in the past month, with 30 per cent doing so at least weekly. That is not a sign of AI working well - it is a sign of AI creating a secondary review workload that sits on top of the original task.
The report also notes that 58 per cent of Australian workers have sent a digital twin to attend in their place - a figure that raises its own questions about presence, accountability and what "attending" a meeting actually means when AI is the one showing up.
Security risks add another layer. Separate research from SecurityBrief Australia found that Australian small and medium businesses are increasingly relying on free or consumer-grade AI tools at work, which introduces data handling and confidentiality risks that most organisations have not formally addressed. [1][2]
Background and Context
Australian workers are adopting AI at a higher rate than their counterparts in the United States and United Kingdom, but most organisations are failing to convert that usage into measurable business outcomes. The Work AI Index quantifies the hidden labour involved in making AI tools usable - and the numbers suggest the gap between adoption and value is structural, not incidental.
The research is consistent with what Mindiam's AI strategy practice sees in client engagements: organisations that deploy AI tools without a supporting framework for governance, training and workflow integration tend to generate activity rather than outcomes. Workers become proficient at using AI in isolation but struggle to connect that usage to business results that anyone can measure.
The AI training gap is a significant part of the story. When workers are not trained on how to evaluate AI output critically, the default behaviour is to trust it - or to spend untracked time fixing it. Both outcomes are costly. The first creates quality risk; the second creates a hidden labour burden that never appears in the productivity figures organisations use to justify their AI spend.
For Australian small and medium businesses in particular, the combination of high adoption rates, low governance maturity and heavy reliance on free tools creates a specific kind of exposure. The tools are being used; the frameworks to use them well are not in place. That is a gap AI automations and structured implementation can close, but only if organisations treat it as a design problem rather than a training problem.
What Comes Next
The Work AI Index data will likely sharpen the conversation Australian business leaders are already having about return on AI investment. For most of 2024 and 2025, the dominant question was whether to adopt AI. That question has been answered - 90 per cent adoption settles it. The question now is whether adoption translates into anything that shows up in the numbers that matter.
The botsitting finding is the most actionable part of the report. If workers are spending 6.5 hours a week correcting and redoing AI output, that is a workflow design problem. It means AI is being inserted into processes that were not built to accommodate it, and the gap is being filled by human effort that nobody is measuring or managing.
Organisations that want to close the performance gap - to move from the 10 per cent who report meaningful gains toward something closer to the UK's 18 per cent - will need to treat AI integration as a deliberate strategic exercise, not a procurement decision. That means auditing where botsitting is happening, redesigning the workflows that generate it, and building the AI capability in their teams to evaluate output rather than just produce it.
The professional services sector faces particular pressure here. Knowledge work is where AI adoption is highest and where unchecked AI output carries the greatest reputational and compliance risk. The 77 per cent of Australian AI users admitting to unchecked output behaviours is not an abstract statistic in that context - it is a liability.
Frequently Asked Questions
What is the Work AI Index and who conducted it?
The Work AI Index was published by Glean's Work AI Institute, a research collaborative involving academics from Stanford University, UC Berkeley and Harvard. The survey covered 6,000 full-time digital workers across the US, UK and Australia, with 1,500 respondents drawn from Australia specifically. It was designed to measure not just AI adoption rates but the quality and organisational impact of that adoption.
What does "botsitting" mean and why does it matter for Australian businesses?
Botsitting is the term the Work AI Index uses to describe the unrecognised effort workers put into prompting, checking, correcting and redoing AI-generated output. The report found workers spend an average of 6.5 hours per week on this activity - close to a full working day. For Australian businesses, this matters because it means a significant portion of the productivity AI is supposed to create is being consumed by the overhead of managing AI, rather than flowing through to measurable business outcomes.
Why are Australian workers more fatigued by AI than their US counterparts?
The report does not offer a single causal explanation, but the data points to a combination of higher adoption rates, higher rates of unchecked AI-output behaviour, and more frequent rework. Forty-three per cent of Australian workers report feeling worn out by AI tools, compared with 33 per cent in the US. Workers who are using AI more intensively, trusting it more uncritically, and spending more time fixing its mistakes are likely to experience more fatigue - regardless of whether the tools themselves are more or less capable.
What should Australian businesses do differently based on this research?
The research suggests three practical priorities. First, audit where botsitting is actually happening in your organisation - which workflows generate the most rework and why. Second, invest in structured AI training that builds critical evaluation skills, not just tool proficiency. Third, treat AI integration as a workflow design problem: the tools need to fit the process, not the other way around. Organisations that approach AI as a procurement decision and stop there are likely to stay in the 90 per cent adoption, 10 per cent results category.
Is the security risk from free AI tools a separate issue?
It is related but distinct. The botsitting and performance data comes from the Work AI Index. The security risk associated with free and consumer-grade AI tools in Australian SMBs is documented separately by SecurityBrief Australia, which found that reliance on those tools is increasing even as the data handling risks they introduce remain largely unaddressed. Both issues point to the same underlying gap: high adoption without the governance frameworks to make that adoption safe or effective.