In 2025, Yan and colleagues ran an unusual study. They placed nine hundred and five participants into small teams to complete analytical, creative and ethical tasks. Some of the teams contained an AI persona. Some did not. The participants in the AI persona teams were not told about its presence.
The teams behaved differently. The presence of the AI persona shifted how teams generated ideas, weighted evidence, deferred to the most plausible suggestion, and resolved disagreement. The shift held across all three task types. And the participants did not detect the change.
That study describes, with unusual clarity, the phenomenon we have come to call identity drift. The team is not the team it was last quarter, and nobody has noticed.
What drift looks like
Drift is rarely dramatic. It is rarely the result of a deliberate decision. It looks like this:
The team starts using a model to generate first drafts. The first drafts are good. Within three months, the team is editing rather than writing. The voice of the work, which used to be recognisably the team's, becomes recognisably the model's.
The team starts using an AI scheduling tool to find meeting times. Within six months, the team is meeting on the days the tool prefers, with the people the tool has prioritised, in the configurations the tool has optimised for. The team's rhythm has changed. Nobody chose the new rhythm.
The team starts using a model to summarise customer feedback. The summaries are useful. Within nine months, the team is acting on the summaries rather than the source. The team's relationship with its customers has been mediated by a system that has its own preferences for what counts as signal and what counts as noise.
None of these is a scandal. Each of them, on its own, is sensible. The aggregate is identity drift, and identity drift is what happens to teams that do not have an explicit conversation about who they are and what AI is for inside that.
Why drift is the default
Three forces conspire to make drift the natural state.
The first is convenience. AI tools reduce the cost of doing things. Cost reduction generates use. Use generates dependence. Dependence generates change in how the team works, which generates change in how the team thinks about what it does.
The second is invisibility. The Yan et al. finding is the important one here. Teams cannot detect the shifts as they happen. The behavioural change runs ahead of the cognitive awareness of it. By the time the team notices the shift, the shift has already produced a new normal.
The third is incentive. Most organisations measure adoption. They count seats licensed, prompts run, hours saved. They do not measure whether the team's identity, its character, its standards, its voice, has held its shape. The metrics push for more drift, not less.
Build this capability in your organisation
Growth Performance designs bespoke programmes that turn insight into measurable behaviour change.
What leaders can do
The work is not to resist AI adoption. The teams that resist adoption fall behind on every dimension that matters. The work is to make the identity conversation explicit, repeated and consequential.
This means three practices.
Name the silent work. Once a quarter, ask the team a direct question. "What is AI doing in our work that it was not doing six months ago, and what has changed about us as a result?" Capture the answers. The exercise is not therapeutic. It is a navigation aid.
Decide what to keep. Not every shift is bad. Some shifts are improvements. The work is to distinguish the shifts the team welcomes from the shifts the team would not have chosen if asked. Decide explicitly. Document the decision. Tell the team.
Install a check. Drift compounds in the absence of friction. A fifteen minute monthly ritual, where the team reviews one workflow and asks "is this still us?", produces enough friction to prevent the worst outcomes without slowing the team to a crawl.
The leadership move
The deeper move, behind all three practices, is the one that distinguishes leadership in augmented teams from management of augmented work. The leader has to hold the question of who we are open, on purpose, against the gravitational pull of convenience, invisibility and metric design.
This is not a comfortable position. It is easier to delegate the question to the tool vendors, the productivity dashboards and the consultants. The teams that do that are the teams the Yan et al. study is describing. The shift happens. They do not notice. By the time they notice, the conversation has already been had on their behalf.
Try this
In your next leadership team meeting, take ten minutes. Ask each member to write, on paper, the answer to one question. "What is one thing that has changed in how we work in the last six months that I would not have agreed to if asked at the start?" Read the answers aloud. Do not debate them. The exercise is the start of holding the line.
Get the full framework. This article draws on The Augmented Us white paper, the evidence based primer with the full six stage practice, the four practitioner artefacts and the research base. Download the white paper or take the team diagnostic.