AI, behaviour and work.

What AI delivers, and where it goes wrong

• In an experiment with realistic writing tasks, people using ChatGPT finished roughly 40 percent faster, and their work was rated about 18 percent higher.^[1]
• Among customer support agents, issues resolved per hour rose by 15 percent on average. Less experienced agents gained the most.^[2]
• Consultants who used AI on tasks it handles well completed more tasks, faster and to a higher standard. On a task outside that range, they were more likely to get it wrong than colleagues without AI.^[3]

for your team AI pays off most when people know which tasks it helps with and which it doesn’t. That isn’t technical knowledge, it’s judgement, and judgement can be trained.

Trust: too much or too little

• People abandon an algorithm faster after seeing it make a mistake, even when it outperforms humans on average. This is known as algorithm aversion.^[4]
• In other situations the opposite happens: people are more likely to follow advice when they believe it comes from an algorithm.^[5]
• People working with automated systems check less carefully the more reliable the system seems. This is known as automation bias.^[6]

for your team The goal isn’t more trust or less trust, it’s the right amount for each task. And that starts with understanding how a language model arrives at an answer.

When people actually start using a tool

• Whether people adopt a new technology depends heavily on two things: do they find it useful, and do they find it easy.^[7]
• Later research adds social influence and organisational support: what colleagues and managers do and expect matters.^[8]

for your team One enthusiastic colleague isn’t enough. Adoption happens when the value shows up in people’s own work, and when the team agrees on it together.

Technostress

• New technology can be a source of stress: pressure to work faster, always being reachable, or the sense of never keeping up. This technostress is linked to more role stress and lower productivity.^[9]

for your team Pushback on AI is often uncertainty rather than unwillingness. More tool explanations won’t fix that. A proper conversation and clear agreements will.

Learning as a group

• Teams learn better when people feel safe to ask questions, make mistakes and voice doubts. This is known as psychological safety.^[10]

for your team That’s why the training runs on group discussion and your own examples, not just explanation.

Our own numbers

From the first AI sessions onwards, I measure before the training and four weeks after how often participants use AI and how confident they feel doing it. The results will appear here once there’s enough data.