Augmentation or Automation? Tracking AI adoption and its human consequences.
One of this year’s significant trade-offs is that between balancing rapid AI adoption with governance, trust, and integration. In this new series of LEADINGThought we will share evidence as it emerges from AI adoption and its impact on the workplace and the workforce. Evidence of Boards and CEOs pushing for rapid adoption to achieve efficiencies and meet ROI commitments is slowly emerging.
But according to researchers at Stanford, cycles of general-purpose technologies, such as AI, require only 1/11th of tech investment [4]. The rest is the investment required to rewire the organisation and put processes in place between people and technology.
So, what are the patterns of early AI adoption and the workplace consequences from pushing through with rapid adoption and automation, instead of rethinking processes and putting in place the necessary governance to achieve trust and integrate AI to support augmentation. Should the focus be automation or augmentation, efficiency or innovation and is there a middle-ground option emerging? How do these different paths impact the humans in the equation?
AI Adoption: Status report
Data on AI adoption –albeit at times contradictory – is slowly emerging. Early studies have been focusing on patterns of early adoption. A distinction emerges between the levels of exposure of jobs, and automation vs. augmentation of tasks, as well as the level of business transformation that coincides with AI adoption.
Anthropic’s fourth Economic Index report [1], tracking AI adoption through its own LLM Claude, shows that the proportion of jobs that have seen AI usage has risen from 36% a year ago, to fifty per cent with AI usage for at least a quarter of their tasks. However, Claude usage remains concentrated among certain tasks, most of them related to coding and writing so far, and the most complex tasks are those where it is struggling most. Claude tends to also be used more, and appears to provide greater productivity boosts, on tasks that require higher education. This suggests that if these tasks shrink the net effect could be to deskill jobs. However, this depends on whether the productivity effects would give rise to new tasks or whether a human worker may need to validate AI output.
But how AI will affect the economy depends not just on the tasks Claude is used for, but the way users will access and engage with the model. The patterns of automation (where the model is directed to complete tasks) vs. augmentation (where users iterate with the model to complete tasks or ask it to explain concepts) has varied over the last year with initially higher augmentation and then increased automation, while we are now back to augmentation.
According to MIT NANDA’s report [9] on business adoption, only 42%% of organisations are transforming their business alongside deployment. As a result, just 5% of integrated AI pilots are extracting value with no measurable P&L impact for the vast majority. But behind the disappointing enterprise deployment numbers lies a surprising reality: AI is already transforming work, just not through official channels. According to the report, while only 40% of companies reported purchasing an official subscription, 90% of workers from the companies surveyed reported regular use of personal AI tools for work tasks.
Early evidence suggests that while usage of AI tools is high, the enterprise level approach is mixed, and the benefits are not yet being realised. There is still time to consider how best to deploy the technology to transform and augment than to automate and replace. Researchers and academics at the World Economic Forum meeting at Davos called out the need for business leaders to make a choice now between ‘automating the past’ vs. ‘augmenting for the future’. While the former may be great for the bottom line, there is no real potential in automating the past.
Automation vs Augmentation: the human consequences
According to research by Better Up 62% of desk workers believe that their organisations want to augment their capabilities which is good news. However, at the same time 30-40% report overreliance of AI tools. This emerging trend reflects people half-heartedly using the technology to cope with being under pressure, rather than engaging with it to develop new ways of working and engaging with customers.
These two potentially divergent paths have consequences, for performance, productivity, and wellbeing. According to research by Better Up’s Chief Scientist Kate Niederhoffer, there’s three behavioural dynamics that result from focusing on automating and ‘replacement’ vs. empowering and augmenting [2].
1. Wellbeing – replacing people has a massive impact on wellbeing, as job insecurity decreases motivation.
2. Workflow integration – the way in which organisations have introduced and framed AI tools can lead to divergent outcomes if AI deployment lacks staff involvement and co-creation.
3. Talent pipeline – the impact of new employees, where decreased hiring means that it’s not possible to build the talent pipeline that will receive the generational knowledge
Upcoming research by Better Up and Oxford university is expected to show the consequences of these two divergent paths, with the automation path leading to lower productivity, inability to attract and retain talent, with impact on performance, while the augmentation path achieves positive benefits across the board and compounding performance gains in the longer-term. The augmentation path takes longer but will pay off because AI adoption will increase.
Research at Harvard and Stanford [5,6] has highlighted the drop in junior staff hiring into AI-exposed occupations, as well as within AI adopter firms as the reason for declining employment of junior staff rather than exits. Data from the Harvard study also demonstrates that while junior employment in adopting firms declined sharply, senior employment continued to rise, alongside declining rates of promotion amongst juniors post-2022.
AI adoption data will continue to emerge as the technology also improves at pace. There remain several unanswered questions on the human impact. While early evidence demonstrates impact on junior employment in the case of larger adopter firms and jobs with high exposure to AI, there is as yet little clarity on how this may be reshaping careers more broadly, how both junior and senior staff will adapt depending on the signals received about the approach to adoption, and the impact on wellbeing already at an all-time low.
Where to Next?
The two paths of augmentation and automation have significant human consequences.
Recognising the benefits of ‘test-and-learn’ and experimenting with the tools, while also releasing capacity through efficiencies, has the potential for significant productivity and performance gains. Solely focusing on efficiencies neglects the value from transforming the business and impacts innovation.
Organisations need to be actively thinking about the optimal ratio between agents and humans and preparing people for that.
But whether AI leads to broad-based empowerment or rigid centralization isn't a technological question; it is a societal one. By 2050, the most important question about AI will not be what it can do, but who gets to decide what it does.
=============================================================
Our LEADINGThought series will continue to track trends and research on AI adoption and its impact on the workplace for the foreseeable future!
Resources
Anthropic Economic Index report: economic primitives, 15 Jan 2026. https://www.anthropic.com/research/anthropic-economic-index-january-2026-report
Better Up (Jan 27,2026). The braver path: Why AI’s biggest ROI comes from investing in people.
Brynjolfsson, E. AI Changed Work Forever in 2025. Time. https://time.com/7342494/ai-changed-work-forever/
Brynjolfsson, E., Rock, d., and Syverson, C. (2021). The Productivity J-Curve: How Intangibles Complement General Purpose Technologies. American Economic Journal: Macroeconomics, vol. 13 (1), pp.333-72.
Brynjolfsson, E., Chandar, B., Chen, R. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence (November 13, 2025). https://digitaleconomy.stanford.edu/app/uploads/2025/12/CanariesintheCoalMine_Nov25.pdf
Hosseini Maasoum, Seyed Mahdi and Lichtinger, Guy, Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data (August 31, 2025). Available at http://dx.doi.org/10.2139/ssrn.5425555
Lichtenberg, N. First of –its-kind Stanford study says AI is starting to have a ‘significant and disproportionate impact’ on entry-level workers in the U.S. Fortune (August 26, 2025). https://fortune.com/2025/08/26/stanford-ai-entry-level-jobs-gen-z-erik-brynjolfsson/