For many people, AI means a chat box.
You ask ChatGPT a question. It explains a topic, rewrites an email, or creates a first draft. That is useful—and it is only one way AI can fit into real work.
The bigger product question is not simply, “Can we add AI?” It is: What role should AI play here, and how much responsibility should it have?
I use five roles to make that conversation concrete.
1. Answer
AI can find, summarize, and explain information in response to a question.
This is the role most people already know. It can help an employee understand a policy, help a customer compare options, or help a manager make sense of a long document.
The key question is whether the answer needs to show where it came from. A restaurant suggestion and a medical explanation do not carry the same stakes. The more important the answer, the more important evidence, uncertainty, and expert review become.
2. Create
AI can produce a first draft: words, images, code, plans, presentations, or many other forms of content.
The value is rarely “no human needed.” More often, AI removes the blank page and gives a person something to react to. The person still brings context, taste, responsibility, and the final judgment.
A useful creation tool makes revision easy and keeps the user in control. A weak one creates more material but also more review work.
3. Recommend
AI can suggest a next step based on information it has been given.
It might recommend which customer to contact, which issue to investigate, which product to consider, or which task deserves attention first.
Recommendations become useful when people can understand the relevant reason behind them. “Do this” is less trustworthy than “Do this because these three things changed.” The product should also make it easy to disagree.
4. Help decide
AI can organize evidence, compare options, surface tradeoffs, and help a person make a better decision.
This role is close to recommendation, but the human decision remains explicit. The AI is there to improve judgment—not quietly replace it.
The product should make the alternatives visible, show what information may be missing, and avoid presenting uncertainty as certainty. A confident answer is not automatically a good decision.
5. Act within limits
AI can complete a task: update a record, route a request, schedule an action, call another tool, or carry out a sequence of steps.
This is where the excitement around “agents” comes from. It is also where boundaries matter most.
Before AI acts, the team needs to decide what it may change, what requires approval, what can be reversed, and what happens when the situation falls outside its instructions. The goal is not maximum independence. The goal is the right level of independence for the task.
A 60-second way to use the five roles
Bring one customer experience, business decision, or repetitive task to mind. Then ask:
- Which role would create the most value: answer, create, recommend, help decide, or act?
- What is the most damaging mistake the AI could make?
- Who should review, approve, or override it?
- What would be measurably better if it worked?
- What is the smallest safe test that could teach us something real?
These questions move the conversation beyond “we should use AI.” They turn a broad ambition into a product choice.
ChatGPT made AI visible. The next chapter is about deciding where AI belongs—and designing the human judgment, trust, and feedback around it.