What AI experimentation reveals about organisational culture

Experimentation reveals the real operating norms

When people begin using AI in everyday tasks, hidden norms quickly surface. Some teams test early and share what they learn. Others wait for permission, or quietly use unofficial tools because the approved options feel too slow. MIT’s research on generative AI management suggests this split between sanctioned use and shadow use is not just a technology issue; it is a culture signal about trust, clarity and appetite for coordinated change ³.

In practice, the most telling question is not whether staff are curious about AI. It is whether they believe experimentation will be supported, reviewed and improved. Where leaders frame AI as a learning process, experimentation can widen participation. Where they treat it as a compliance exercise, people may comply at surface level while exploring in private. From an organisational perspective, that gap matters because it often hides both useful innovation and unmanaged risk ³.

Good pilots need boundaries, not open-ended enthusiasm

Stanford’s 2026 AI Index shows rapid adoption and accelerating capability, alongside a rising count of documented incidents and lagging responsible AI reporting. That combination matters for culture because it means enthusiasm alone is not a reliable control. Organisations that experiment without a clear purpose, review process or risk ownership may learn quickly, but they also expose themselves to avoidable failure ¹.

MIT’s work points to a practical response: give employees sanctioned access to a small set of tools, define acceptable use, and explain what is and is not appropriate for work. That approach is not about slowing progress. It is about making experimentation legible. In cultural terms, the message is that curiosity is welcome, but it must operate within shared standards. For leaders, the strongest signal is consistency: people take their cues from how often those standards are applied, not how elegantly they are written ³.

Leadership style shapes what AI can teach you

Harvard Business School research on leaders working with AI and human teams suggests that strong leadership is visible in the quality of interaction rather than in volume of talk. The study found that effective leaders asked more questions, encouraged conversational back-and-forth and used inclusive language. That is relevant beyond the laboratory because AI trials can reveal whether managers still default to telling rather than listening when the environment is uncertain ².

For organisations, this means AI experimentation can become a mirror for leadership capability. If a team uses AI well, it may indicate that leaders are comfortable with inquiry, ambiguity and reassessment. If the same team struggles, the issue may be less about tool competence and more about habit. Our advisory view at Authenticate is that AI pilots are most useful when they are designed as disciplined conversations about work quality, decision rights and learning behaviour, not simply as demonstrations of novelty.

Use experimentation to test culture, not just use cases

A helpful way to structure AI experimentation is to ask what the pilot should reveal about the organisation. Is the aim to improve a process, to build confidence with a tool, or to surface barriers in decision-making? Those are different goals. Without a clear question, experimentation can generate activity without insight. MIT’s distinction between broadly applicable tools and tailored solutions is useful here because each requires different governance, skills and ownership ³.

Leaders should also watch who participates. If experimentation is confined to a small technical group, the organisation may learn little about how AI affects frontline judgment, collaboration or customer response. Wider involvement helps reveal whether people across functions can ask better questions, identify weak assumptions and challenge outputs responsibly. That broader participation is also useful for spotting where training, role clarity or process redesign is needed before scaling further ⁴.

Practical steps for leaders

Start by defining the decision the pilot is meant to inform. A vague ambition to “explore AI” usually produces vague learning. A sharper purpose, such as testing how AI supports drafting, research, or knowledge retrieval, makes it easier to evaluate both benefits and limits.

Set a small number of non-negotiables: approved tools, data handling rules, human review expectations and escalation points for questionable outputs. This gives people room to experiment without guessing the boundaries.

Ask teams to record not only what the tool produced, but what the team learned about its own working habits. Did people challenge outputs? Did they verify claims? Did they rely on one voice? These questions often reveal more about culture than the tool itself.

Finally, make review a shared practice. Regular reflection sessions help distinguish useful experimentation from noisy activity, and they create a safer route from trial to adoption. The aim is not to prove that everyone is ready for AI at once, but to understand what the organisation does when it is asked to learn in public ¹²⁴.