Google analysed millions of anonymised AI usage records worldwide and concluded that AI is currently used primarily as a tool that supports and collaborates on work rather than replacing workers.
Most usage was "shallow" collaboration, confined to supporting roles:
The core figure: in non-routine cognitive work, fewer than 10% of AI conversations aimed at full automation.
An interesting contrast sits alongside it. That category represents 35% of all tasks but accounts for 65% of Gemini usage.
In other words, non-routine cognitive work attracts the majority of AI usage while over 90% of that usage is collaboration rather than automation. The area where AI is used most is also the area least automated.
That contrast provides empirical grounding for the replacement-versus-augmentation debate: heavy usage does not indicate replacement.
The researchers cautioned against optimistic readings.
1. Collaboration may convert into automation. As model performance improves, collaboration-centred usage may progressively shift toward automation.
2. Labour market effects may still appear. Productivity gains among experienced workers could reduce entry-level hiring.
This second caveat deserves particular attention. It argues that hiring can fall even without replacement occurring. If one experienced worker's output rises, the case for hiring a junior weakens — a different mechanism from individual-level substitution.
3. Productivity gains are not showing up in the statistics. Researchers warned that current productivity increases in households and industry are not well reflected in GDP and similar measures.
Unmeasured productivity growth complicates policy judgment — and helps explain why arguments about whether any of this is working continue unresolved.
Diagnose your current usage level honestly. If you are still at drafting and information retrieval, that is the normal range. Whether to stop there or build verification processes and go further is an organisational decision.
Account for the side effects of reducing junior hiring. Fewer entry-level hires means a smaller pipeline of experienced workers later. AI raising senior productivity also means senior people remain necessary.
Close the organisational support gap. Korean workforce surveys show the same pattern — near-universal usage with insufficient training. Those figures are covered in 86.9% of Korean Office Workers Use AI at Work.
Find ways to deepen shallow collaboration. Solidifying rules into documents rather than re-explaining context each session is covered in Marketers Who Re-Explain Their Brand Every Time vs. Those Who Locked It Into a Skill.
Calibrate automation expectations. Fewer than 10% aiming at full automation is consistent with agent pilots failing to reach production — the causes of which are covered in Four Myths About AI Agents.
From analysing millions of anonymised AI usage records worldwide, that AI is currently used primarily as a tool supporting and collaborating on work rather than replacing workers.
Usage confined to supporting roles such as drafting, reviewing, idea generation and information retrieval — which described most AI use in the data.
Fewer than 10% of AI conversations in non-routine cognitive work. That work represents 35% of all tasks but 65% of Gemini usage.
That improving model performance may convert collaboration into automation, that productivity gains among experienced workers could reduce entry-level hiring, and that current productivity increases are not well reflected in GDP.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
Opening an AI before a search engine is becoming the default. As the path back to original creators narrows, where does competitive advantage in the AI era actually come from?
Meta expanded its vibe coding app Pocket to all U.S. users. Describe what you want and it builds a small app or game — then that output circulates in a social feed.
Meta's CEO published a 6,500-word essay on personal superintelligence that doubles as an argument against AI regulation — while Meta cut 20% of staff earlier this year.