Who are your staff asking now?

• New staff now ask a chatbot the questions they used to ask a supervisor. They get faster. The team loses the conversations where ideas get argued over and where “how we do things here” gets passed on.
• AI makes each person’s ideas better and the team’s ideas more alike. In one early trial, teams with AI agreed about eight minutes sooner and looked at fewer options.
• In a small firm, the chat between an old hand and a new hire is the training. Keep it going on purpose.

From a Business Times report on the INSEAD AI Forum Asia, held in Singapore on 30 October.

• the spiral

  Ask the bot, not the boss

  Martin Gonzalez, who leads organisational AI research at Google DeepMind, calls it a “self-sufficiency spiral”. New hires take their questions to AI instead of a manager. They learn to work on their own, and the back-and-forth between colleagues fades. In his words, “Innovation emerges from a clashing of ideas.”

• better, but alike

  Each idea improves, the pile shrinks

  Phanish Puranam, a strategy professor at INSEAD, says studies find AI lifts the quality of each person’s ideas but narrows the range across the group. Each person gains today. The team pays later, so nobody has a reason to hold back.

• the trial

  Eight minutes faster, fewer options

  In a brainstorming experiment, teams working with AI agents settled on ideas about eight minutes sooner. They explored a narrower range and felt less ownership of what they came up with. The researchers are still collecting data.

• their fixes

  Write first, and don’t share one bot

  Gonzalez writes his own first drafts and only then asks AI to sharpen or challenge them. Puranam suggests each person uses their own personalised AI tool, not one agent shared by the team.

• no training team

  A big company has courses and buddies to fall back on. In a 15-person firm, the supervisor answering questions is the training. When the questions stop, so does the learning.

• fewer heads

  A small team has fewer ideas to start with. If everyone asks the same tool, the range shrinks faster.

• nothing looks wrong

  Output goes up, so the boss sees no problem. The cost shows up a year later, when an odd job comes in and nobody has learned how to handle it.

• 1 · order

  Alone, then together, then AI

  do this

  Each person writes three ideas on paper first. Put them together as a team. Only then ask AI to test, add to or knock down the shortlist.

  why

  Starting with AI pulls everyone towards the same answer.

• 2 · check in

  Bring the AI answer to a person

  do this

  When a new hire gets an answer from AI, they show it to someone senior with one line: “where I’d disagree”.

  why

  The AI answer becomes the start of a conversation, not the end of one.

• 3 · say why

  Seniors think out loud

  do this

  When your experienced staff make a call, they say why, even if nobody asked.

  why

  New staff who have stopped asking still hear the reasoning. AI can’t give them that part.

• 4 · count

  Count the options before you choose

  do this

  Before the team picks, count how many different options were on the table. Name one person to argue against the favourite.

  why

  If you only reward speed, speed wins.

• 5 · ask wider

  Ask AI for the odd one out

  do this

  Ask “what would nobody here suggest?” or “argue against what we’ve agreed”. Skip “what’s the best answer?”.

  why

  That way AI widens the list instead of closing it.

• 6 · by hand

  Some jobs without AI, at first

  do this

  For their first few months, new staff do a few judgement jobs by hand: pricing a job, answering a complaint, planning a delivery.

  why

  That is how they learn to spot when AI is wrong.

1. Pick a real question: how do we win back a customer we lost?
2. Split into two groups. Group A asks AI straight away. Group B writes alone for five minutes, then pools ideas, then asks AI.
3. After ten minutes, count the different ideas each group has. Then ask each group whose idea it feels like.

If group B has more ideas and feels more ownership, you have your answer for how meetings should run.

• The trial is early and not yet published. The article doesn’t name the studies behind the “ideas get more alike” claim.
• INSEAD’s new lab on this topic is funded with US$3 million from Google.org and Google DeepMind, and one panellist works at Google DeepMind. That doesn’t make them wrong. It’s worth knowing.
• Only the losses were described. Some teams may spend the time AI saves talking to each other more. Nobody has numbers on that yet.
• Personal AI tools mostly run on the same few models underneath, so giving each person their own may add less variety than it sounds. The variety comes from people.

Source: “AI may boost individual productivity, undermine workplace innovation: panellists”, The Business Times ↗ (INSEAD AI Forum Asia, 30 Oct 2026).

My own view. I use AI every day and shaped this page with Claude. Not affiliated with or endorsed by INSEAD, Google or The Business Times.