Craig Stanley
inXS
Home / AI trolley problems / Users

AI trolley problems: users

Twelve everyday calls about checking, privacy, saying when AI helped and keeping your own skills.

For anyone using Copilot, ChatGPT or another AI assistant in their day-to-day work.

Read without playing

Key takeaways for users

Twelve takeaways (best read after you play)
  1. Check every source, figure and quote in AI output before it goes out under your name.
  2. Only put work or personal data into AI tools your organisation has approved for it.
  3. Say when AI helped with your work, especially where it matters to the reader.
  4. Keep practising the skills you need to check AI output.
  5. Tell people when an AI is recording or summarising a meeting, and let them say no.
  6. Decide what saved time is for. More output isn't always the most useful choice.
  7. When AI ranks or filters people, look at who it put at the bottom.
  8. Use AI drafts for structure, and write the parts that need care yourself.
  9. Use AI for a first answer, and check policy, legal and money questions with the person who owns them.
  10. Take personal details out of prompts unless the task needs them and the tool is approved for them.
  11. Correct AI errors kindly and in the open, so others don't act on them.
  12. Try new AI features on a real task, and keep what helps.
All twelve dilemmas as thought experiments

  1. The made-up source

    Trolley problem

    Your AI-drafted report cites five sources and you can't find one of them. Pull the lever to spend 20 minutes checking every source. Do nothing and send it.

    • Pull the lever: Twenty minutes checking
    • Do nothing: Made-up sources, your name

    Ask yourself: What will I check myself before I send AI-assisted work?

  2. The free chatbot

    Trolley problem

    You need a quick summary of a client contract. The approved tool is slow today. Pull the lever to paste it into a free chatbot instead. Do nothing and wait for the approved one.

    • Pull the lever: A client contract in a free chatbot
    • Do nothing: The slower approved tool

    Ask yourself: Do I know which AI tools I'm allowed to use for which kinds of information?

  3. Say so?

    Thought experiment

    AI helped you write a proposal your manager is pleased with. You can mention that it helped, or say nothing.

    • Option A: Mention it
    • Option B: Say nothing

    Ask yourself: Would I be comfortable if my manager knew how much AI helped with this?

  4. Skill fade

    Trolley problem

    AI writes all your spreadsheet formulas now, and you've stopped following how they work. Pull the lever to write one yourself each week. Do nothing and keep the time.

    • Pull the lever: Slower Fridays
    • Do nothing: Formulas you can't read

    Ask yourself: Which skills do I need to keep so I can check the AI's work?

  5. The note-taker

    Trolley problem

    The AI note-taker joins your meetings automatically. One guest hasn't been told. Pull the lever to pause and ask everyone. Do nothing and carry on.

    • Pull the lever: An awkward pause
    • Do nothing: Recorded without knowing

    Ask yourself: Does everyone in the meeting know the AI is listening, and what happens to the notes?

  6. Twice as fast

    Thought experiment

    AI halves the time your weekly reports take. You can use the time to write more reports, or spend it making the existing ones better.

    • Option A: Write more reports
    • Option B: Make them better

    Ask yourself: If AI saved me an hour a day, what would I want to do with it?

  7. The bottom of the pile

    Trolley problem

    You're shortlisting for a role and AI has ranked 200 applications. Pull the lever to read the bottom 20 yourself before they're rejected. Do nothing and reject them.

    • Pull the lever: An hour on the bottom 20
    • Do nothing: The bottom 20, rejected unseen

    Ask yourself: When AI ranks people, do I check who it ranked low and why?

  8. The upset customer

    Trolley problem

    AI has drafted your reply to an upset customer. It's polished and a little cold. Pull the lever to rewrite the opening yourself. Do nothing and send the draft.

    • Pull the lever: Five minutes rewriting
    • Do nothing: Polished and cold

    Ask yourself: Would the person reading this feel it was written for them?

  9. Who do you ask?

    Thought experiment

    You're stuck on a question about an internal policy. You can ask the AI assistant, or ask the colleague who owns the policy.

    • Option A: Ask the AI
    • Option B: Ask the owner

    Ask yourself: When does a quick AI answer need checking with the person who owns the topic?

  10. Names in the prompt

    Trolley problem

    You want AI to draft a letter about a named colleague's sickness absence. Pull the lever to take out the name and health details first. Do nothing and paste it all in.

    • Pull the lever: Anonymise first
    • Do nothing: Health details in the prompt

    Ask yourself: Does this prompt contain anything about a person that it doesn't need?

  11. The wrong answer, shared

    Trolley problem

    A colleague posts an AI answer about holiday entitlement in the team chat. It's wrong. Pull the lever to correct it, politely, in the same chat. Do nothing and avoid the awkwardness.

    • Pull the lever: A polite correction
    • Do nothing: A wrong answer spreading

    Ask yourself: If I see a wrong AI answer being shared, do I feel able to say so?

  12. The new feature

    Trolley problem

    A new AI feature has appeared in a tool you use every day. Pull the lever to spend an hour this week trying it on a real task. Do nothing and leave it for later.

    • Pull the lever: An hour this week
    • Do nothing: Leave it for later

    Ask yourself: What's one task I could try AI on this week?

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 speed, 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.

  • Checking. Anything sent under your name has been checked by you.
  • Privacy. Personal and client data only goes into tools approved for it.
  • Openness. Your team knows how you use AI.
  • Learning. You're learning the tools and keeping your own skills.
  • Productivity. AI saves you time on the tasks where it helps.
  • Fairness and care. People affected by your AI-assisted work are treated fairly.

Question to take awayWhat will I check myself before I send AI-assisted work?

Inspired by Neal Agarwal's Absurd Trolley Problems. The dilemmas, scoring and reports here are new, written for AI at work.

About me

Craig Stanley

Microsoft AI consultant and technical architect, based in Whitley Bay. Over the last few years I've delivered Microsoft 365 Copilot, Copilot Studio agents, Azure AI Foundry work and governance for UK public sector and financial services organisations.

What interests me is the decision underneath the tool: what it costs, what it risks, and whether a small, transparent model can make it better. I write the methods up here and on Substack so anyone can use them.

Find me
inXS