How AI flattery is eroding workplace culture

Leaders want humans in control

Leaders say employees should be in the loop, at the helm, and as co-pilots in AI-powered workplaces. But organizations are not testing whether AI design actually supports that vision.

AI sets the default globally

Just a few leading AI companies are setting the default for AI-employee interactions across enterprises globally, with scant testing and differentiation across organizations.

AI shapes critical work moments

Employees now use AI to develop team strategies, make core business decisions, and prepare for pay discussions, performance reviews, one-to-ones, and project kick-offs.

Three risks from current AI design

AI risks reducing collective diversity of ideas, encouraging cognitive surrender where employees defer to AI instead of using judgment, and embedding sycophancy that discourages challenge.

Sycophancy echoes corporate collapse

Corporate history shows sycophancy, unchallenged leaders, and uninterrogated ideas led to business collapse at Enron, Lehman Brothers, and Theranos. Organizations worked hard to root out sycophancy in human interactions.

BT Group tests an alternative

BT Group, the UK's largest telecommunications group with 77,000 employees, teamed up with MoreThanNow to understand how default AI design impacts performance and find a better alternative.

Goal setting is the test case

BT identified the annual goal-setting process as a high-stakes moment where sycophancy might undermine their organizational value of being Challengers who set big goals and stay curious.

The experiment design

BT ran a randomized controlled trial with two thousand four hundred participants. Three groups: control with no AI, default AI that mirrored existing tools and tended towards sycophancy, and challenger AI designed to be challenging…

How default AI responds

Default AI told employees: "That's great! What does that look like in practice for you?" Employees elaborated, finalized, and submitted. The tool did not push them to improve their goals.

“That's great! What does that look like in practice for you?”

How challenger AI responds

Challenger AI asked: "That's a solid goal.What does it look like in practice?What specific actions would you take?"Then: "Can you specify what you will do?"Then: "Can you push further on scale or timeline?"

“Can you push further on scale or timeline here?”

Challenger AI produced better goals

Challenger AI rated three and a half times more employee goals as excellent compared to default AI and control groups. It also reduced the likelihood of weaker goals being set.

Users still enjoyed the experience

Although BT dialed down sycophancy in the challenger group, participants reported no differences in enjoyment, psychological safety, and excitement about using the technology again across all three groups.

Measure outcomes, not just inputs

Leaders should move beyond measuring active users, license requests, token consumption, and training completion. Instead measure outputs and outcomes that reveal how AI-employee interactions influence performance and achievement.

Test before rolling out

Organizations should not adopt a wait and see approach. Embed organizational objectives and values in AI-employee interactions through randomized controlled trials before rollout, not after.

Invest in human solutions

Leaders should resist solving AI challenges with more AI. Instead invest in equipping managers with capability to give constructive feedback, create psychologically safe environments, and build collaborative teams.

BT moves forward with testing

BT is integrating challenger AI into goal-setting and will investigate how feedback interactions with AI affect work-output quality, feedback culture, and human relationships at work.