Key takeaways
- Technology does not decide outcomes, leadership choices do.
- Automation decisions are design decisions. The structure of work defines success.
- Projects that replace people without redesigning roles rarely succeed.
- Deskilling corrodes trust and innovation.
- Augmentation makes roles richer and organisations more resilient.
- HR’s test is whether automation delivers a clear human win, raises role quality and involves employees in design.
AI is changing the language of strategy meetings. Behind every promise of speed or scale sits a quieter conversation about what happens to work itself. The technology may be new but the question isn’t: will automation create better jobs or fewer of them?
IMD strategy professor Howard Yu calls this the automation fork in the road. His argument is simple: if an AI plan demoralises the very people needed to make it succeed then it isn’t automation at all but erosion.
That framing cuts through the hype. It reminds HR and business leaders that automation isn’t a force of nature but a set of design choices. And those choices determine whether jobs become richer and more valuable or narrower and more disposable.
What a 1980s car plant teaches us about AI and work design
In a recent post on LinkedIn Yu reflected on a case from the 1980s that still defines this debate – GM’s notorious Fremont plant. In the early 1980s it was a byword for dysfunction: absenteeism, sabotage and the lowest-quality cars in the organisation. Two years later the site reopened under Toyota as the New United Motor Manufacturing Inc (NUMMI). The same workforce returned, yet within a year it had become one of the most productive plants in the United States.
What changed? Workers were given permission to stop the production line if they spotted problems. Toyota introduced the ‘andon’ cord that any worker could pull if they saw a fault. At GM halting production had been forbidden. At Toyota it became a sign of care and improvement. Toyota paired technology with human judgment. The factory’s success showed that people were never the problem. The system was.
Design choices decide whether automation builds trust or undermines it.
The automation fork: how organisations choose the future of work
Yu’s automation fork describes the point where every organisation decides whether technology will extend people’s capabilities or narrow them. The concept aligns closely with research by MIT economists Daron Acemoglu and Simon Johnson, whose work earned them the 2024 Nobel Memorial Prize in Economic Sciences. Their studies show that technology can either concentrate wealth and erode job quality or create broad-based prosperity, depending on how it is deployed. As Acemoglu says, progress is not automatic but depends on the choices we make about technology.
Over the past few years Acemoglu has warned of what he calls “so-so” technologies – systems that automate tasks but fail to raise overall productivity. Self-checkout machines are the classic case. After three decades some retailers are quietly removing them because they displaced workers without delivering meaningful efficiency gains. Acemoglu suggests that AI could follow a similar path if designed purely to replace human input rather than improve how people work.
In an article in MIT Sloan Management Review last year he described the difference through an industrial example: “In productivity revolutions of the past, like at the Ford Motor Company, automation was critical but only when combined with new products, new tasks, new ways of using machinery, new creativity. The Ford factory would not have done anything of note if it took exactly the cars that other companies were producing and made them with a bit more automation.”
That perspective reinforces Yu’s fork in the road. The real value of automation comes when technology complements human creativity. As Acemoglu says: “We should be using machines to make humans better.” Generative AI has this potential but only if used to enhance decision-making and insight instead of replacing the very tasks that give people purpose.
MIT Sloan’s Johnson makes the same point from a management lens: “The way we make progress with technology is by making machines useful to people, not displacing them.”
Their warning is practical as well as philosophical. Acemoglu estimates that only around 5% of jobs are at genuine risk of full automation within the next decade. The greater risk lies in wasted investment: projects that displace people without raising productivity or trust. “A lot of money is going to get wasted,” he told Bloomberg. “You’re not going to get an economic revolution out of that 5%.”
When automation falls short
We’ve seen more recent examples where removing people was positioned as progress but produced little real value. Amazon Go’s cashierless stores promised thousands of outlets but only a handful remain and, even working smoothly, the experience left customers uneasy. Tesla’s attempt to automate everything in its Model 3 factory led to breakdowns and delays before Elon Musk admitted: “Humans are underrated.”
The same dynamic applies inside organisations when AI is deployed without rethinking how value is created.
Automation lifts performance only when it creates new value for people as well as systems.
What good AI and automation design looks like
Where automation supports skill, culture and performance the results are consistently better.
In banking the arrival of ATMs reduced cash-handling tasks but expanded the teller role. Branches focused more on advice, problem-solving and customer relationships. Productivity rose because technology handled the transactions while humans handled the complexity.
In healthcare AI tools now scan medical images and highlight potential issues for radiologists. Rather than replacing specialists this triage helps them focus on interpretation and treatment planning. Multiple studies show improved accuracy and faster diagnosis when humans and machines work together.
In both these cases we can see a simple pattern: technology succeeds when it is used to remove drudgery and elevate human contribution.
Designing AI with worker voice
Recent research from MIT, Bringing Worker Voice into Generative AI , argues that technology design succeeds when those closest to the work shape how it is used. Based on interviews with more than 50 people including business leaders, AI engineers and labour leaders, the authors found that the more that stakeholders are involved in defining the problems and opportunities the technology can address the more likely it is that these tools will be used to augment how workers do their jobs rather than displace them.
As MIT professor Julie Shah, a co-author of the study, notes from her own work: “The most successful applications of automation were those where the engineers were working side by side with those on the shop floor. The opportunity here is to follow that same playbook so that generative AI can have the positive outcomes we hope it will have.”
The skills employees need to thrive with AI and automation
Automation only delivers sustained value when employees have the skills to adapt with it. A large-scale study in Nature Human Behaviour examined more than 70 million job transitions across nearly 1,000 occupations. It found that technical expertise now has a half-life of less than four years. The only skills that endure are foundational: maths, reasoning, communication, collaboration.
Workers with these skills adapt, learn and move into new specialisations. Those without them get stuck in low-mobility “skill traps” with lower wages and higher automation risk.
Augmentation only works if people have the right foundations. Otherwise automation simply exposes gaps. HR’s role is to ensure that those foundations are in place.
Meanwhile field research by Harvard Business School and MIT with consultants at Boston Consulting Group found that workers using GPT-4 were faster and produced higher-quality results on tasks where the technology could assist but performed worse on tasks beyond its capability. The findings underline a simple point: productivity gains depend on design. When roles are built to combine human judgment and machine support, people perform better. When systems try to replace judgment entirely, performance falls.
The half-life of technical skills is now under four years. Foundational skills last decades.
Why HR must lead on AI and automation choices
As AI spreads across white-collar and knowledge work every pilot, every HR tech investment, carries implicit choices. Rather than adopting any AI solution that promises efficiency HR leaders can test projects against three simple questions:
- What is the human win? What will people stop doing and what will they start doing instead?
- How will this raise role quality, not just output? Are we creating opportunities for judgment, creativity and collaboration?
- How are employees involved in the design? Were they part of shaping the pilot, or are they hearing about it only once it’s deployed?
The answers to these questions don’t just shape productivity but also culture, trust and HR’s credibility as a strategic voice.
Practical checkpoints for HR leaders
To keep AI projects aligned with people-centred design HR leaders can use five simple checkpoints before rollout:
- Clarify the purpose. Define how the project improves both business outcomes and job quality.
- Involve employees early. Invite those closest to the work into pilot design. Their insight prevents errors and builds buy-in.
- Measure experience as well as efficiency. Track indicators such as autonomy, collaboration and learning alongside productivity.
- Create safe escalation routes. Ensure that people can raise concerns about AI outputs or monitoring without fear of reprisal.
- Link automation with learning. Every system change should include time and budget for skill development.
These checkpoints turn automation from a technology initiative into a capability-building exercise.
The choice ahead: designing AI for better, not fewer, jobs
The shift to AI marks another stage in the long story of how people and technology shape one another. The outcome depends on everyday decisions about roles, skills and culture. History suggests that progress is broadest when tools amplify human potential.
Yu is right that we stand at a fork in the road. But it’s not an abstract fork for economists or technologists. It’s a very real set of choices facing HR leaders in every AI deployment.
Instead of asking what the technology can do we need to ask what we want our people to become. And the time to decide that is now.
FAQ
Q: What does Howard Yu mean by the “automation fork" in the road?
A: It’s the point where organisations choose whether technology will complement people or replace them. Each path leads to very different outcomes for jobs and performance.
Q: How can HR test whether an AI project is heading in the wrong direction?
A: Look for warning signs: no clear human win, little employee involvement or outcomes measured only in cost savings.
Q: What’s the difference between augmentation and deskilling?
A: Augmentation raises the value of remaining human tasks, while deskilling narrows and monitors them, eroding dignity and performance.
Q: What skills matter most in an AI-enabled workplace?
A: Reasoning, communication, collaboration and problem-solving – the foundations for learning new tools and adapting to change.
Q: What role should HR play in automation governance?
A: HR must set guardrails on monitoring, lead redeployment and reskilling and ensure that cultural mechanisms like “stop-the-line” exist for AI.
The “how” of building responsible automation – readiness audits, co-design frameworks, collaboration norms – is a bigger conversation than one article can cover. That’s why The People Space has developed its Essential Guide to HR’s Role in the Human-Machine Workplace. It provides practical tools and templates for HR leaders navigating these choices.
About the author
