AI trolley problems: leaders
Twelve calls about people, pace and where the time AI saves should go.
For executives, heads of service and managers who lead teams through change.
Twelve calls, then your report
Pull the lever or do nothing. Some dilemmas have no lever: they're open thought experiments where you pick a side. After each call you see what happened and what the other choice would have done.
Your answers stay in your browser. Nothing is sent or stored.
Dilemma
Result
Key takeaways for leaders
Twelve takeaways (best read after you play)
- Tell people early when AI will change their work, with a timeline and a support offer.
- Budget for skills alongside licences. Access without training gives shallow use and more risk.
- Decide openly what happens to time AI saves. If people never see any benefit, they stop reporting savings.
- Consult people before AI monitors or scores them, and check the measure tells you something useful.
- Give every automation a named owner and enough documentation for someone else to run it.
- Agree where AI drafting is welcome and where people expect a manager's own words, and say when AI helped.
- Measure the tasks and decisions AI changes. Active users show activity, which is only a starting point.
- When you stop an unapproved AI tool, offer an approved one quickly, or the workaround comes back.
- Allocate scarce licences by the work they'd change, and add a few enthusiasts to spread what works.
- The person who sends AI-assisted work answers for it, so check client-facing output before it goes.
- Say when AI helped with your own work. It sets the norm for everyone else.
- Run AI work as small experiments with measures and stop rules, then scale the ones that work.
All twelve dilemmas as thought experiments
The announcement
Trolley problemPart of your team's work will be automated next year. Pull the lever to tell them now, before the details are settled. Do nothing and wait until the plan is final.
- Pull the lever: A hard conversation today
- Do nothing: Twelve months of rumours
Ask your team: Which roles will AI change in the next year, and when will the people in them hear it from us?
Licences or lessons
Trolley problemYou have £100,000 for AI this year and the plan spends all of it on licences. Pull the lever to spend half on training, which halves the number of licences. Do nothing and give everyone access.
- Pull the lever: Half the licences, real training
- Do nothing: £100,000 of licences
Ask your team: What share of our AI budget goes on skills, and how would we know it's working?
The time dividend
Thought experimentAI saves each person in your team about five hours a week. You can bank that as capacity and take on more work, or give part of it back as time for learning and fixing the processes that annoy people.
- Option A: Bank it as capacity
- Option B: Share it with the team
Ask your team: When AI saves time in our team, where does that time go, and did we agree it together?
The activity score
Trolley problemA supplier offers AI that scores staff productivity from keyboard and mouse activity. Pull the lever to consult staff and assess the data protection impact first, which delays it a quarter. Do nothing and switch it on.
- Pull the lever: A quarter's delay
- Do nothing: Monitoring switched on
Ask your team: Have we asked the people affected before using AI to watch or score their work?
The champion leaves
Trolley problemOne person built all 15 of your team's automations. They leave on Friday and nothing is written down. Pull the lever to pause every automation until it's documented. Do nothing and hope they keep running.
- Pull the lever: 15 automations paused
- Do nothing: 15 automations, no owner
Ask your team: Who owns each automation and agent in our team, and is any of it written down?
The AI-written review
Thought experimentAnnual reviews are due and a manager in your area wants AI to draft them. You can allow it, as long as the manager edits and signs each one, or ask managers to write reviews themselves.
- Option A: Allow it, with sign-off
- Option B: Keep reviews human
Ask your team: Where is AI help fine in managing people, and where do people expect a manager's own words?
The green dashboard
Trolley problemYour director wants weekly active users of Copilot as the measure of success. Pull the lever to report which tasks got faster or better instead, which takes real work. Do nothing and send the usage figures.
- Pull the lever: A harder report, a real answer
- Do nothing: A green usage dashboard
Ask your team: What would we measure if we wanted to know whether AI made our work better?
The unapproved note-taker
Trolley problemYour team uses a free AI note-taker in client meetings. IT hasn't approved it and nobody knows where the recordings go. Pull the lever to stop it today. Do nothing and keep the good notes.
- Pull the lever: Back to manual notes
- Do nothing: Client calls on an unknown server
Ask your team: Which AI tools does our team use that nobody approved, and what would make the approved ones good enough?
Who goes first?
Thought experimentYou have 20 licences for a team of 80. You can give them to the enthusiasts who've been asking, or to the roles with the most repeated writing and decision work, whether or not those people asked.
- Option A: The enthusiasts
- Option B: The best-fit roles
Ask your team: Who would gain most from AI in our team, judged by the work they do?
The wrong figure
Trolley problemA team member sent a client a report with an AI-drafted section that contains a wrong figure. Pull the lever to add a peer check on client-facing work that AI helped with. Do nothing and ask people to be careful.
- Pull the lever: A peer check on client work
- Do nothing: Errors in front of clients
Ask your team: Which of our outputs go to clients or regulators, and who checks the parts AI helped with?
Say it out loud
Trolley problemYou used AI to help write this month's all-staff update. Pull the lever to add a line saying so. Do nothing and send it as it is.
- Pull the lever: A line saying AI helped
- Do nothing: Quiet use at the top
Ask your team: Do we say when AI helped with our work, and does the team feel able to do the same?
One big bet
Thought experimentYou're planning next year. You can run one ambitious AI programme across your whole area, or three small experiments, each with a measure and a decision to scale or stop every quarter.
- Option A: One big programme
- Option B: Three small experiments
Ask your team: What would we stop if this quarter's AI experiment misses its measure?
How the scoring works
Each answer moves one or more of the six areas below up or down by a point. Your score in an area is where your total lands between the lowest and highest totals the twelve dilemmas allow. 75% or more is strong, 45% to 74% is developing, under 45% is a gap. Your archetype comes from two totals. One is pace, measured as momentum over the following months rather than speed today, so a shortcut that causes an incident later counts as slow. The other is the remaining five areas combined (care), where the top half starts at 70%. It's one practitioner's view of good practice, written down so you can argue with it.
- Direction and value. Every AI use in your area has a purpose and a measure someone reviews.
- People and trust. People hear about changes to their work from their manager first.
- Accountability. Someone named answers for every AI tool your team uses.
- Openness. Leaders are open about their own AI use.
- Pace. Good ideas reach a decision within weeks.
- Skills. The training budget sits alongside the licence budget.
Question to take awayWhich roles will AI change in the next year, and when will the people in them hear it from us?
Inspired by Neal Agarwal's Absurd Trolley Problems. The dilemmas, scoring and reports here are new, written for AI at work.