If You Sell AI to Your Staff as a Productivity Tool, Don’t Be Surprised When They Use It Like One
Stanford’s new research finds adoption rises sharply when leaders frame AI as job enrichment, not as a way to do more with fewer people.
Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Most AI programmes I see in South African boardrooms open with a slide of productivity numbers: hours saved, cost reduced, headcount freed. The slide is usually accurate, and it is also the exact line of language that destroys the adoption the business case depends on. New research argues that the quickest way to kill an AI rollout is to pitch it as a productivity tool, and the quickest way to make it work is to pitch it as a way to make the work itself better, backed by manager support, protected time and new metrics.
CONTEXT AND BACKGROUND
The best new evidence sits in a Stanford HAI piece summarising research by Arvind Karunakaran, Roshni Raveendhran and Tami Kim. A study of 184 workers at three firms found that required AI use made staff feel their work had lost meaning, which lowered their motivation to learn. A two-year study of a law firm compared two divisions. In the division where managers framed AI around enrichment and backed it with training, protected time and knowledge-sharing, paralegals used the tool 58% more and experimented with it 70% more than peers in the other division. The authors argue that performance metrics must change too, because rewarding volume or AI tokens used pushes staff towards low-value activity.
Gallup finds US employee engagement flat at 31% in the first half of 2026, unchanged for the second year, even as AI adoption rises. At organisations where managers actively support AI use, 48% of staff are engaged; without that support, the figure is 30%. Access is not adoption.
INSIGHT AND ANALYSIS
Staff behaviour already shows the cost. In a 14-country WalkMe study, 54% of workers had bypassed company AI tools in the past 30 days, 45% had used unsanctioned shadow AI, and only 9% trusted AI for business-critical decisions against 61% of executives. That is a trust gap, not a tooling gap, and it reads like staff who do not believe their employers' story about what the tool is for.
Researchers at the University of Oxford's Wellbeing Research Centre, writing for Harvard Business Review, argue that leaders face a choice between using AI to cut costs through automation and using it to grow revenue through augmentation. Automation delivers early gains; augmentation requires deeper investment and depends on employees who engage with the tools.
PwC's 2026 AI Jobs Barometer, analysing more than one billion job postings, found that AI postings grew about eight times faster than the overall labour market, and the largest gains sit at firms that restructure work rather than bolt AI on. Restructuring means deciding which tasks AI should take and which humans should do more of, a conversation that runs well only when staff believe the answer is not "all of them".
I have previously written about this when arguing that AI is becoming a second brain, and that when memory becomes a service, thinking can start to atrophy. The same lesson applies to adoption. If staff treat AI only as a way to produce output faster, they will produce polished work they understand less well, and the organisation will lose the judgement it depends on. Framing the tool as enrichment is not a kinder story. It is a more accurate one about what the tool should do.
IMPLICATIONS
South Africa has the highest working-age generative AI adoption on the continent, at 23.1% in the first quarter of 2026, up from 19.3% in the first half of 2025. That lead is real and fragile. With youth unemployment above 60%, "AI will cut your headcount" framing lands harder here than almost anywhere, and the shadow AI figures suggest many staff have drawn their own conclusions.
Boards should check four things. What language appears in the AI business case, and does the same language reach staff? Are managers trained before executives, or still last in the queue? Do performance metrics reward judgement, or token counts and output volume? Is there protected time to experiment, or is AI simply added to the existing workload? The strongest counter-argument, that productivity numbers are honest and necessary, is correct; those numbers improve when enrichment framing lifts adoption, and decay when "do more with less" drives adoption underground.
CLOSING TAKEAWAY
The quietest failure in corporate AI is not a model that gets things wrong, but a workforce that uses it defensively, in secret, and only enough to keep up. That failure is a leadership choice disguised as a technology problem. Boards that want AI to earn its business case should start by changing the business case itself, so that it describes what AI will let people do, not what it will let the company stop paying for. The firms that get this right will be the ones whose staff still have reasons to care about the work.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence's impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan's work at https://www.aiforbusiness.net
Comments