The leadership team has reached agreement.
Finance supports the AI initiative because it expects lower operating costs. Marketing sees a faster content engine. Operations expects better forecasts. HR hears a change to roles and skills. Technology is already thinking about data, architecture, access, and all the ways this could go sideways at 2:00 on a Sunday morning.
Everyone says yes.
To five different plans.
The problem usually stays hidden because the word *AI* feels specific enough to move the conversation forward. It sounds like a category. In practice, it can hold several very different capabilities, changes to work, investments, and risks.
That makes agreement surprisingly easy.
It also makes execution surprisingly hard.
Listen for the strategy hiding behind the word
When Finance says AI, the conversation may be about automation and cost. Which repeatable steps could happen with less manual effort? How much time or expense might that remove?
When Marketing says AI, the conversation may be about generation. How could a team produce drafts, concepts, variations, and analysis faster while protecting quality and brand judgment?
When Operations says AI, the conversation may be about prediction. Where could patterns in existing data improve a forecast, surface an exception, or support a decision?
When HR says AI, the conversation may be about work. Which tasks change? Which judgment becomes more important? What will managers need to evaluate that they have never had to evaluate before?
When Technology says AI, the conversation may be about the system underneath all of it. What data does it need? What can it access? How reliably does it perform? Who maintains it when the business changes?
Each interpretation can be reasonable. Each points toward different spending, ownership, measures, and consequences when something fails.
Yet the meeting may record one line in the plan: “Launch the AI initiative.”
That line can survive several rounds of approvals because everyone gets to keep their own version of the future inside it.
Finance approved a cost case.
Marketing approved a production engine.
Operations approved a decision system.
HR approved a change to work.
Technology approved an architecture problem.
The disagreement may simply have been scheduled for later.
Watch where the hidden meanings reappear
The first clue often shows up in the budget.
One leader expects a software purchase. Another expects a data project. Someone else assumes training is the main expense. The person leading the initiative receives a budget built for one interpretation and a mandate built from several.
Then ownership gets fuzzy. Technology may own the platform, while a functional leader owns the business result. HR may own the role changes, while Legal and Risk set boundaries on use. Those can be sensible divisions of responsibility, but only when the handoffs are visible. A broad instruction to “partner cross-functionally” leaves everyone supportive and nobody clear about the decision they hold.
Success measures drift too.
A generative system might be judged on quality, review time, and rework. A predictive system may need to beat the current forecasting baseline. An automation may need to reduce cycle time without raising exception rates. An agent with permission to act brings questions about escalation, supervision, and the consequence of a wrong action.
One AI metric rarely carries all of that.
Usage may show that people opened a tool. It says little about whether the intended work changed, whether the output was good, or whether the business outcome improved.
This is why some apparent execution problems begin as language problems. The team starts building before it has made its different assumptions visible.
Give the ambiguity a name
Computer scientist Marvin Minsky used the phrase “suitcase word” for familiar labels that carry a jumble of different meanings. In his writing on consciousness, he argued that one ordinary word can appear to name one thing while packing many different processes inside it.
AI has become exactly this kind of word in leadership conversations. Leaders can use the same label while referring to different technologies, risks, workflows, and desired outcomes.
The metaphor is useful because it turns a vague communication concern into something a team can inspect.
If “AI” is the suitcase, what has each person packed inside?
A perfect definition that settles every technical debate would ask too much of this exercise. The useful move is smaller: make the meaning precise enough for this decision.
That means the next question after “Should we invest in AI?” becomes:
What exact capability are we considering, what work will it change, and what business outcome should that change produce?
The wording creates room for several answers. It also stops those answers from hiding inside one label.
Try this in your next leadership meeting
Choose one active or proposed AI initiative. Give each leader the prompts below and ask them to answer independently before anyone discusses them. Five quiet minutes can be more revealing than another hour of general conversation.
The AI suitcase exercise
1. The business outcome
What should become observably better if this initiative works?
2. The work
Which workflow, task, decision, or customer moment will change?
3. The capability
Are we asking the system to automate steps, predict an outcome, interpret images or video, generate new material, take actions as an agent, or combine several of these?
4. The system action
What will the AI actually produce, recommend, decide, or do?
5. The human contribution
Where will people add context, judgment, approval, correction, or escalation?
6. The consequence of error
What happens when the output is incomplete, wrong, biased, stale, or delivered at the wrong moment?
7. The accountable owner
Who holds the business outcome, distinct from the executive sponsor and the technical owner?
8. The evidence
What would we compare with today’s process to decide whether the change helped?
Once everyone has written an answer, compare the pages. Circle the places where answers differ. Those differences are the meeting.
You may find that the team has one initiative with a few wording gaps. Good. Clarify them and carry on.
You may find that two or three separate initiatives have been packed into one. That gives you a choice: separate them, sequence them, or state which one has priority.
You may also find agreement on the technology and disagreement on the outcome. That deserves attention before a vendor, pilot, or project plan gives the ambiguity a more expensive form.
Leave the room with a sentence you can test
The exercise can end with one shared sentence:
We are using [specific capability] to change [named work] so that [business outcome] improves, measured against [current baseline]. [named person] is accountable for that outcome, with [named human judgment or control] protecting the work when the system reaches its limits.
It may feel almost painfully specific. That is part of its value.
Finance can inspect the outcome and baseline. Technology can test the capability and controls. HR can see the work and human contribution. The functional leader can confirm the operational change. The CEO can see who holds the result.
Now the team has something more useful than agreement on AI. It has an initiative people can explain the same way after they leave the room.
A shared word can start the conversation. A shared definition gives the work somewhere to go.
If your leadership team is stuck inside broad AI conversations and needs a practical way to turn them into decisions about real work, subscribe to the Work Redesigned Lab at lab.workredesigned.co.
Source note
Marvin Minsky, “Suitcase words in Psychology,” a draft chapter from *The Emotion Machine. https://www.mit.edu/~dxh/marvin/web.media.mit.edu/~minsky/eb4.html, July 28, 2005.