AI Adoption Is Being Built Around Researchers, Not With Them

Performing Instead of Building

Most research organisations are still going through the motions of AI adoption rather than building. An unarticulated anxiety runs through the profession. This kind of transformation has no precedent, for those experiencing it or those expected to lead it. Experienced practitioners are watching expertise built over years be simultaneously celebrated as irreplaceable and steadily eroded. The tools doing the eroding can approximate aspects of that expertise at considerable speed. The reluctance to engage with new tools, the overcritical assessment of AI outputs, the disengagement from conversations about the future of the discipline: these are not technophobia. They are a rational response to a genuine threat to professional identity many organisations are yet to address directly.

The human dimension is real, but the operational impact tends to go equally unmeasured. Teams that are psychologically disengaged produce fragmented, inconsistent usage patterns. The tools get adopted in name, the underlying capability does not develop. The organisation falls further behind at each inflection point. What is consistently underestimated is that AI adoption is not primarily a technology problem. The strategy can be sound, the tools can be available, but adoption still stalls. People are not aligned with the direction, and processes have not been redesigned for new ways of working. Anxiety and disengagement, left unaddressed, become a structural drag on the very transformation leadership is trying to achieve.

Understanding why requires looking at how the industry has settled into performing adoption rather than doing it.

How Performing Becomes the Default

Drawing on responses from 3,000+ senior leaders across 24 countries, Deloitte's 2026 State of AI in the Enterprise report identified the AI skills gap as the single biggest barrier to integrating AI into existing workflows (Deloitte, 2026).

The research industry is no exception. The response varies by organisation size, but the outcome is largely the same: activity without capability.

Larger organisations tend to move faster. The budget, dedicated resource, and structural support exist to make transformation possible at pace. But speed without alignment creates its own problems. Meanwhile, researchers expected to change how they work are rarely consulted early enough about what that change should look like.

Mid-size organisations face a more constrained set of options. The ambition is present but the resource to act on it properly is not. This results in two recognisable patterns.

The first is studied passivity. Teams acknowledge the technology but wait for a clearer signal from senior leadership, clients, or the industry at large before committing to anything substantive. It reads as ignorance from the outside but is a reasoned reluctance to invest in a moving target.

The second is individual effort without organisational strategy. The champion exists, interest is real, but there is no framework, dedicated time, or collective direction. This pattern is the most precarious because it depends entirely on a few individuals' bandwidth. In practice it produces isolated experiments that do not accumulate, leaving individuals to carry responsibility for what is structurally a shared problem. In my own experience, the AI champion role existed as exactly this kind of designation without the conditions to make it real. Full delivery workloads left no protected time for building. The response was straightforward: carving out half a day per week specifically for experimentation. A small structural change, but the first one where intention is matched by the conditions to act on it.

What Building Actually Takes

The organisations making sustainable progress are not the ones with the largest AI budgets. They are the ones that have invested in people and process before expecting the tools to deliver.

Give researchers agency over how they learn

Not everyone on a research team is starting from the same place. Some are comfortable experimenting independently. Others need more structured support before they can engage meaningfully with new tools. A blanket training programme treats a diverse team as a uniform one and produces uniform disengagement. Individual conversations about starting points and learning preferences are both more respectful and effective at producing actual capability.

Share the uncertainty rather than manage it from a distance

Researchers who do not know what their role will look like in two years do not become more productive in the interim. They become more guarded. A direct conversation about which tasks AI will absorb, augment, and cannot plausibly touch is not a difficult one to have. Leaders often hesitate because they are not entirely certain of the answers themselves. They do not need to have the answers before starting the conversation. Shared uncertainty is more useful than the appearance of clarity. The discipline of research is fundamentally about tolerating ambiguity long enough to produce something useful. For a research team, naming the uncertainty is already the first step toward resolving it.

Treat protected time and external objectivity as non-negotiable

Half a day each week of hands-on experimentation on real problems produces more than any workshop. AI adoption rarely works as a one-time technical rollout, and what makes the difference is the capacity to test, adjust, and refine over time. Organisations navigating this transition often also need a degree of objectivity that is difficult to maintain from the inside. Existing hierarchies, delivery pressures, and professional identities steer what gets prioritised. External input from experienced practitioners can compress months of internal trial and error into a focused engagement of a few days.

Consult researchers before the plan is finalised

Plans are designed by people with strategic responsibility and handed down as directives. It is a reasonable way to manage change at scale. However, researchers who will determine whether adoption succeeds or fails understand their own work with a granularity that no strategy document can capture. Involving them at this stage is not a consultation exercise. It is how the adoption becomes realistic.

The Missing Piece

There are tool subscriptions, occasional workshops, and an expectation that AI adoption will happen alongside client work. In most research organisations, that is the extent of it. Research is built on working through complexity and making sense of what resists easy answers. That thinking should shape how adoption is designed. The piece that is missing from most adoption strategies is present in the same organisation that wrote them. The question is whether organisations are willing to ask early enough for it to matter. Those that do build lasting capability, not just the appearance of progress.