AI’s battle of the skills: upskilling vs deskilling

Laetitia@Work #78

Hi everyone,

Like a lot of people who write for a living, I often ask myself where my value lies now that generative AI can produce high-quality text, process more information in seconds than I could in a year, and enable anyone—regardless of writing skill—to create polished, expert-level content. The reality is, making a living from expert writing is already becoming increasingly difficult. The notion that less qualified individuals can now perform tasks once reserved for specialists is fueling considerable anxiety.

Aside from the debate about whether AI will enable human workers or replace them, THE question is: Does generative AI mostly amplify the skills of experienced workers, or does it level the playing field by enabling less experienced, less qualified workers to perform at higher levels? Will upskilling or deskilling prevail? (For those who’ve never heard these words: the upskilling strategy consists in pushing people to acquire more skills whereas the deskilling strategy relies on hiring less qualified, less paid people).

Both narratives coexist. Both the upskilling and deskilling phenomena coexist and probably always have throughout technological revolutions. The digital revolution has simultaneously empowered highly skilled professionals to achieve unprecedented productivity while enabling less qualified workers to perform complex tasks with technological assistance. Consider how specialised software has made elite architects, designers and engineers vastly more productive, while GPS navigation transformed transportation by allowing drivers without geographic knowledge to navigate efficiently (think of the Uber driver versus the London cabbie who spent years mastering "The Knowledge").

This dual pattern continues with generative AI, making any definitive claim about which trend will dominate inherently suspect. I'm highly sceptical of studies that claim X% of people will need upskilling and Y% of jobs will disappear by 2032 or any other arbitrary date but I do understand such predictions are a lucrative business in this uncertain world of ours.

Here are some thoughts on the upskilling vs deskilling battle💡👇

Two competing models

The upskilling model: AI as a skill multiplier

In this model, AI tools enhance the capabilities of already skilled workers, widening the gap between experienced and novice employees. In their academic paper, Crowston and Bolici (2025) describe what they call a "multiplier effect" in which AI enhances the capabilities of those with higher prior knowledge, leading to significantly better task performance compared to novices. The performance gap between experts and beginners is set to widen. Those who excel at prompting, grasp context, understand available tools, and apply critical thinking to distinguish fact from fiction will significantly outperform those who take information at face value.

One example comes from research on sales environments. When AI systems were implemented to handle routine lead qualification calls, investigators found that top sales agents still dramatically outperformed junior colleagues. The best agents were 2.8 times more likely to close sales than their less experienced counterparts because they could develop better sales scripts and answer unexpected questions—skills that junior agents couldn't match even with AI support.

This pattern is particularly obvious in roles requiring creative problem-solving or deep domain expertise. When routine tasks are automated, the remaining work often demands higher-level skills, giving experienced employees an advantage that AI can amplify rather than diminish.

The deskilling model: AI as a skill leveler

Conversely, many studies show AI dramatically improving the performance of junior workers, sometimes bringing them near the level of experts.

According to a new paper by MIT Sloan associate professor Danielle Li, MIT Sloan PhD candidate Lindsey Raymond, and Stanford University professor Erik Brynjolfsson, PhD ’91, inexperienced workers actually stand to benefit the most from generative AI.

The co-authors found that contact center agents with access to a conversational assistant saw a 14% boost in productivity, with the largest gains impacting new or low-skilled workers. In other words, the workers were upskilled, not replaced, thanks to the technology.

“Generative AI seems to be able to decrease inequality in productivity, helping lower-skilled workers significantly but with little effect on high-skilled workers,” Li said. “Without access to an AI tool, less-experienced workers would slowly get better at their jobs. Now they can get better faster.” (MIT Management Sloan School, 2023)

This pattern extends to software development as well. GitHub's CEO predicts that AI could soon write up to 90% of code created by corporate developers, potentially reducing the advantage that experienced programmers have traditionally held over newbies.

However, while AI might enable junior workers to match or even surpass seniors in technical capabilities, experience still helps with soft skills. The interpersonal abilities that make someone truly effective in an organisation—conflict resolution and emotional intelligence—can’t really be accelerated. These skills are developed through years of workplace interactions, failures, and adaptations. A junior employee might produce impressive AI-assisted deliverables but still struggle with reading a room, managing difficult conversations, or navigating office politics.

There may be a gender gap when it comes to AI's impact on skills

The upskilling versus deskilling debate takes on additional significance when viewed through the lens of gender. Current research suggests that AI's impacts on skills may affect men and women differently, potentially either reinforcing or helping to close existing gender gaps in the workforce.

In fields where women are underrepresented (software engineering or data science, for example), AI's levelling effect could theoretically provide opportunities to accelerate gender parity. Female newcomers could thus be empowered to enter male-dominated fields.

But there's a concerning flip side. Many of the roles at highest risk of deskilling through AI automation are in sectors with high female representation: administrative support, customer service… The deskilling of these roles could disproportionately affect women's economic opportunities and career advancement paths.

The demographic implications of the two models

These competing models create different challenges for managing workforce demographics, particularly in ageing societies with shrinking pools of young workers.

Scenario 1: Focus on senior workers while the lower rungs of the ladder disappear

If organisations automate entry-level work while retaining only experienced staff, there’s a fundamental challenge: how will new workers develop the expertise needed to eventually become experienced workers with critical skills and process knowledge? This creates a "missing ladder rung" problem.

When AI handles the routine tasks that traditionally served as training grounds, newcomers lose essential opportunities to develop skills and professional intuition. Without these formative experiences, the pipeline of talent development breaks.

And that may lead to a demographic cliff: as experienced workers retire, they leave behind a knowledge vacuum that can't be filled because younger workers haven't had the chance to accumulate equivalent expertise. The result could be a widening skills gap and reduced institutional knowledge. Senior workers could also be asked to never retire.

Scenario 2: Junior workers thrive while seniors face obsolescence

If AI primarily levels up junior workers, organisations may be tempted to hire less experienced, less expensive staff. It’s so tempting to save on human work! Also, AI's ability to act as a performance leveler could help companies address skilled worker shortages by making it possible for them to hire less experienced people in higher-value roles.

This approach makes sense in regions with strong demographics or open immigration providing abundant young workers. But it also comes with problems. The first one concerns the fate of senior workers who would then face increased precariousness. The second one concerns organisational learning: if you eliminate the need for experts, then who will develop solutions to the new problems that the AI hasn't encountered before? How will you generate the valuable primary data to feed future AI systems?

Finding a balance

Organisations need strategies to navigate these competing trends. Here are some interesting approaches:

1. Always preserve learning opportunities

If AI systems become more capable, workers risk becoming overly dependent on them. Rather than enhancing human judgment, AI can begin to replace it entirely, creating workplace environments where employees operate on "autopilot," merely implementing AI suggestions without critical evaluation. This dependency introduces significant organisational fragility – when systems inevitably develop bugs or fail (and they always do!!), no one retains the capabilities to diagnose or solve these issues, potentially leading to catastrophic failures (I don’t know why, but this makes me think of Elon Musk’s whiz kids and the federal payment systems they risk bringing to catastrophic failures).

To counter this effect, organisations should design AI implementations that deliberately preserve opportunities for human decision-making and skill development. This might include creating "training modes" where AI assistance is deliberately limited to encourage skill development, or systems that provide explanations alongside recommendations to foster deeper understanding rather than blind acceptance. These approaches ensure workers maintain their problem-solving capabilities. (Use pen and paper on a regular basis!)

2. Be cautious about overreliance on AI

Organisations that become critically dependent on AI systems will become more and more fragile. They could lose the capacity to function when AI systems become unavailable because workers no longer possess the necessary competencies to perform key tasks on their own.

Skill erosion represents a critical concern in AI-augmented workplaces, where workers gradually lose abilities they once possessed through lack of practice and overreliance on automated systems. Do you remember the humans in the film WALL-E? They relied on automated systems so much that all their skills (and muscles) had completely atrophied.

Unlike traditional tools that extend human capabilities, generative AI can replace entire cognitive processes, leaving skills to atrophy through disuse. For example, if an organisation's accounting department becomes entirely dependent on sophisticated software, accountants may lose their ability to perform calculations or follow procedures manually, creating significant disruption if systems fail or require replacement (which, again, they eventually do).

This creates a paradox: as organisations become more technologically sophisticated, they become more and more fragile. Likewise, the more individuals use AI, the more fragile they may become. (This completely contradicts the FOMO-fueled injunction to use AI more in order to avoid falling behind.)

3. Consider who benefits from AI productivity gains

Who really benefits when AI makes workers more productive? Does the extra value created go to workers through higher pay, or mostly to companies and the tech firms that build these AI tools? For example, if customer service staff can handle more calls per hour with AI help, will they earn more money, or will companies just hire fewer people? (I have a hunch, but I don’t want to spread pessimism).

Ultimately, how these benefits are shared determines how workers feel about AI. If companies use AI mainly to cut jobs or keep wages flat while demanding more output, workers will resist using these tools. Conversely if workers see their paychecks grow when they use AI to be more productive, they'll be happy to adopt the technology.

This distribution of gains will ultimately need to be negotiated between workers and employers. Unfortunately, without strong collective bargaining, individual workers may struggle to claim their fair share of AI-driven productivity gains.

A word of conclusion?

The debate between upskilling and deskilling will not have a single winner, as both models will persist across industries and roles. The most important idea is perhaps that organisations should always design AI implementation strategies with a long-term view on skill development, particularly in ageing societies where experienced workers will become increasingly scarce.

Biologist Olivier Hamant’s work on robustness in living systems further informs this perspective. He insists that resilience stems not from rigid optimisation but from flexibility, diversity, and the ability to absorb shocks. In nature, redundancy—often seen as inefficiency—enhances resilience, ensuring continuity even when individual components fail. Yet modern organisations, obsessed with efficiency and lean structures, often strip away these redundancies, leaving them more vulnerable. As you integrate AI into evolving skill ecosystems, you need to prioritise adaptability over pure optimisation! Accept that some “waste” (like training people whose work you don’t yet need) will actually make you more robust!

---

You could also read these other newsletters on related topics:

• Should seniority be a criterion for pay?

• Why AI can make work more painful

• What if generative AI was actually underwhelming?

• Are older people less innovative?

🎙️ For Nouveau Départ, I recorded several podcasts (in French):

• Fermentations : pourquoi le moisi nous fascine (with Anne-Sophie Moreau)

• Mal-entendus : les Français, les médias et la démocratie (with Nina Fasciaux)

• L'IA, le travail et les RH : au-delà des idées reçues (with Pierre Monclos)

• L’avenir de l’automobile (with Sarah Zitouni)

• Retour au pays et hybridation (with Pauline Rochart)

💡For Nouveau Départ, I wrote many new articles (in French): Vive le biais de normalité !, Fragiles progrès : allons-nous perdre les acquis féministes ?, Achats et ressources humaines : une frontière floue, Expulsions de sans papiers et marché du travail, La Silicon Valley en guerre contre ses travailleurs, La fin des programmes DEI dans les entreprises ?, Ceux qui reviennent … et créent du lien, Joyeux 50ème anniversaire, Jeanne Dielman !

→ Subscribe to receive my future podcasts and articles directly in your inbox!

Miscellaneous

🤔 Return-to-office won't actually make your employees more productive, McKinsey says, Ben Kesslen, Quartz, February 2025: “McKinsey said executives who think return-to-office will be a panacea to their productivity problems should instead focus on “collaboration, connectivity, innovation, mentorship, and skill development.” Those five qualities, the firm found, are essential to improving workplace performance and strengthening organizational health.”

Don’t be too efficient because it will make you less robust! 🤗