Your Job Is Not Just to Use AI. It’s to Work Well With It.

You can lead your team's AI work by making clear who owns the result.

Consider a team preparing a customer proposal. Someone uses AI to pull the notes together, draft the recommendation, and suggest options the team had missed. It is useful work, and there is something encouraging about having a draft ready while there is still time to improve it. Then a reviewer asks whether the proposed delivery date has been checked with the people doing the work.

The person who prepared the draft assumed the reviewer would check. The reviewer assumed the date came from an agreed plan. Now someone has to trace it back before the proposal can go out, and a colleague who could have helped shape the recommendation is being asked to approve a promise they never made.

If AI can do part of your job, what exactly is your job now? In that situation, you can help the team agree on how a draft becomes a recommendation someone can stand behind. You need enough technical fluency to understand what the tool can contribute, and enough judgment to define what people still need to decide. You can start that conversation without managing everyone involved.

That is where work design becomes practical. Trust is harder to maintain when nobody knows what has been checked. Functional silos persist when a proposal crosses into another team's work only at approval time. Changing the process also means talking with the people whose responsibilities will change, including who will have time to review it. Adding AI to the task leaves those agreements for people to make.

Those agreements also need to give people time and responsibility to question the work, even when an AI-generated answer looks ready to use. In the survey behind The Impact of Generative AI on Critical Thinking, 319 knowledge workers described thinking that shifted toward verifying information, integrating responses, and overseeing the task. Higher confidence in AI was associated with less reported critical thinking. These are self-reported patterns, but they give a reason to make review responsibilities explicit before a polished draft makes the work feel finished.

Before your next AI-assisted task, ask the people involved, “Can we agree on what AI will help us do, who will check the work, and who will decide whether it is ready to use?” Capture the answers in a short task card, using these five sections:

1. Outcome: Describe the result you need, who will use it, and what it must meet to be useful.
2. AI role: List the parts of the task AI will help with, the inputs it will use, and what it should leave to people.
3. Human role: Name who owns the result, who checks the work, and who decides it is ready to use.
4. Evidence: List the sources, tests, or people you will use to check important claims and assumptions before relying on the result.
5. Stop rule: Specify what would make you pause the work, who will resolve the issue, and what must be checked before continuing.

Keep the review proportional to the stakes. An internal recap may need a quick check for accuracy and fit. A customer commitment deserves a check of every material claim. Work involving safety, employment, legal exposure, or significant money should go through the relevant qualified reviewer and approval process. Asking AI to reconsider its answer does not provide independent evidence.

The card is a starting point to test. Ethan Mollick makes the case for learning from the work itself:

Measure whether AI is improving the whole task by tracking total time, corrections, and rework across the next few comparable tasks. Include the time other people spend checking or fixing the work, and compare the final quality with your usual standard. If your part gets faster but someone else spends longer untangling assumptions, change the inputs or give AI a narrower role. You are helping people use the tool well without leaving responsibility to whoever happens to catch the mistake.

Which part of your team's AI-assisted work is everyone assuming someone else has checked?

If this resonates, share with your network to help more teams agree on who checks and owns their AI-assisted work.

References

Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, Article 1121, 1–22. https://doi.org/10.1145/3706598.3713778

Stanford Seed. (2024, June 11). Co-intelligence: An AI masterclass with Ethan Mollick. Stanford Graduate School of Business. https://www.gsb.stanford.edu/insights/co-intelligence-ai-masterclass-ethan-mollick