We often think of AI as a technological revolution that will transform industries, disrupt jobs and change the nature of competitive advantage. Inside organisations, though, it is unfolding in a much more complex and less predictable way. In many firms, AI is still more narrative than operational reality, making it hard to move from discussion to meaningful action.

This tension was revealed by senior leaders of global firms at a recent workshop in Paris co-organised by INSEAD’s Stone Centre for the Study of Wealth Inequality, its counterpart at University College London, and the Centre for Economic Policy Research (CEPR). 

The senior executives, who came from professional services, retail, creative, consumer goods and other sectors, spoke on condition of anonymity. They said AI is reshaping how work gets done, how decisions are made, and how value gets created and distributed, but the real challenge is how firms make sense of the AI transformation. 

Underneath the surface of experimentation and investment are deeper questions around organisational design, human capability and the future of work. From what the participants said, four key tensions stood out.

1. What does AI adoption really mean?

Across organisations, the term “AI adoption” is widely used but understood differently. For some firms, AI is primarily a tool for incremental productivity gains, improving efficiency at the margins. For others, AI is empowering a more fundamental shift: Employees across functions are building solutions themselves, enabling innovation by many rather than a select few.

In other contexts, particularly in knowledge-intensive sectors, AI raises questions beyond productivity: think authorship, data origins and credibility. When outputs are generated by models trained on vast amounts of external data, it raises the question of who creates value and how that value should be recognised.

A key insight of the workshop, which had some 30 participants, is that shared language and AI literacy are closely connected. When organisations use the same term internally to describe fundamentally different phenomena, this can lead to misalignment. Leaders may see a coherent strategy, while employees experience a fragmented set of initiatives with no clear purpose, resulting in a gap between AI ambitions and their limited connection to everyday operations. 

People in the same organisation need both a common understanding of what AI is and the knowledge and confidence to use it effectively. Firms that begin with language and learning – that develop a shared, working understanding of what AI is, what it’s not and what it’s expected to achieve – tend to move more effectively from experimentation to impact.

The implication for leaders is simple but important: Before scaling AI, define it not just in technical terms but in strategic ones. What specific problems is the technology meant to solve? Where is it expected to create value and where is it not? What does success look like, and over what timeframe?

This isn’t just about semantics. One workshop participant from a not-for-profit described a split within their organisation between AI enthusiasts and sceptics, noting that there’s often “a mismatch between what AI does and the sentiment we give to it”. Without a shared understanding, the lack of common meaning becomes, in the words of the participant, “prohibitive to the discussion”, let alone to working together.

2. Overcoming organisational bottlenecks

In addition to addressing ambiguity, the issue of organisational readiness is equally important. Access to tools, data or technical expertise is often cited as a barrier to AI adoption. In practice, however, most firms can access powerful systems, but the bigger challenge is integrating them into existing structures and workflows.

This reflects the “J-curve” effect seen in previous waves of technological change. Initial adoption is typically followed by a period of adjustment, during which productivity may decline. One leader of a firm providing coaching and mentoring services augmented by AI offered a candid account of this transition. Their firm spent 18 months in a building phase that generated little visible output. “It was an unproductive time,” they acknowledged, “but gains take time to realise.” To its credit, the firm understood it as a necessary investment – and has since seen returns. The J-curve demands patience as much as capability.

Several participants described this phase as lasting 12-18 months, often with limited visible gains. Understanding the J-curve not as failure but as a necessary phase helps organisations manage expectations and sustain commitment. Firms that persevere are those that invest in complementary capabilities, including training, redesigning workflows and change management. 

As AI enables more widespread experimentation, employees across levels and functions can test, adapt and build. This calls for a shift in both structure and mindset, with leaders supporting a degree of decentralisation and ambiguity while staying aligned with strategic priorities.

3. Spreading AI’s benefits

Perhaps the most immediate effects of AI adoption are being felt in the human experience of work. AI is creating divisions, as employees who are able and willing to use these tools effectively often gain a disproportionate advantage. The rest risk being left behind due to lack of access, training or confidence.

These differences aren’t only technical but behavioural and cultural. Some people are more inclined to experiment, while others are more cautious, particularly when the technology challenges existing ways of working or raises concerns about quality and reliability. Firms that invest in broad-based AI literacy and create environments for experimentation tend to distribute benefits more evenly. One academic participant argued that leadership needs to take responsibility. “If we don’t do anything, it will increase inequality,” they argued. Indeed, the divergence in hiring trends suggests that a gap is already opening between the AI-savvy and the rest.

There’s also a deeper dimension related to meaning and identity. In many roles, especially those involving creativity or human interaction, AI raises questions about authorship, autonomy and the value of work. Some employees may find meaning in tasks that others are quick to automate, highlighting the need for careful and inclusive decisions about where AI is applied.

Additionally, AI is increasingly being used in people management processes such as hiring. Although this can improve efficiency, it also raises new challenges, such as making it harder to distinguish among candidates when many now use AI tools to craft their CVs. This places greater emphasis on human judgement and the interpersonal aspects of recruitment. 

For leaders, the challenge is to ensure that AI adoption does not create a tiered or hierarchical organisation. This is not only a question of fairness but also of performance.

4. The nature of human work and developing talent 

As AI systems become more capable, they are increasingly performing tasks that were previously central to many roles. Activities such as drafting reports, writing code and conducting analysis are now automated or significantly accelerated. Roles are evolving, with a shift from execution to oversight and judgement. 

A useful way to think about this shift is in terms of “semi-autonomy”. Rather than fully delegating decisions to AI, organisations are designing systems that augment human judgement while preserving human responsibility for consequential decisions. The role of the human is moving from being inside the loop to being on top of it.

This has implications for organisational design and talent development. In professional services firms, junior employees were traditionally trained in detailed analytical work under supervision. If AI reduces the need for such work, how will expertise be developed and assessed? At the same time, the skills that differentiate individuals are changing. The ability to think critically, connect distinct pieces of information and exercise judgement in uncertain situations is becoming increasingly valuable. Ditto interpersonal skills such as empathy, communication and leadership. Several leaders expressed concern about “intellectual offloading”, where individuals defer to AI outputs without fully engaging with the underlying reasoning. 

The risk of skipping foundational steps was captured vividly in the discussion. If the pace of AI-assisted work is prioritised over depth of understanding, organisations may accumulate blind spots. Panellists noted that new graduates may still need fluency with underlying data even if they no longer produce it directly. As one participant put it, some young people today can’t change a light bulb because the task has always been handled for them. 

By the same logic, if detailed knowledge is never acquired, it can’t be drawn on when things go wrong. The challenge for firms is to ensure that speed doesn’t come at the cost of the judgement that only arises from getting one’s hands dirty. In other words, integrate AI in ways that continue to develop human capability. This may require different training, redesigning career pathways and creating new opportunities for experiential learning. 

Leading through the transition 

Beyond internal organisation, AI is pushing firms to rethink how they create and capture value. In professional services, the traditional model based on billable hours is increasingly under pressure and gradually being replaced by value-based pricing.

At the same time, sources of competitive advantage are evolving – proprietary data, strong client relationships and the ability to build ecosystems are becoming critical. Beyond generating solutions, firms must have the data and context to create something that scales and delivers long-term value.

Leading through this transition requires deliberate, not just fast, action. Organisations should be clear about what they are trying to achieve, thoughtful about which decisions should remain human, and intentional about how AI aligns with organisational design and purpose.