I was planning the MVP for my first app with AI.
My original plan was simple: release an Android app first.
AI gave me a bigger plan. It suggested building Android and iOS in parallel, then launching both together.
At first, that looked like a normal suggestion. It was written clearly. It had a plan behind it. It even sounded ambitious in a good way.
But I had not decided to build both platforms.
That was the problem.
A suggestion can quietly become a decision
When AI helps us plan, it does not only write down what we say. It can fill gaps, connect ideas, add features, choose an approach, and make a plan look complete.
Sometimes that is exactly what we need.
But if we do not notice what AI has added, a suggestion can quietly become part of the project. Then we are no longer just reviewing an output. We are living with a decision that nobody clearly made.
For an app, that decision could change the platform, the MVP scope, the architecture, the cost, or the user experience. For another piece of work, it could change the recommendation, the message, the priority, or the next action.
The output may look helpful. The impact can still be real.
So who owns the decision?
AI can generate the idea, but it cannot own the consequence.
If an unapproved feature makes the first version harder to use, the end user does not experience it as “an AI suggestion.” They experience it as part of the product.
That is why I think ownership stays with the person or team using AI. If an organization uses AI in its work, it also needs to set the boundaries: what AI can suggest, what it can change, and what needs a human approval before moving forward.
The formal guidance says the same thing in a bigger way: human roles and responsibilities should be clear when AI is part of a decision. NIST’s human-AI interaction guidance makes this point for organizations building and using AI systems.
In normal work, I would say it more simply:
AI can bring an idea. A person still needs to say yes, no, or not yet.
The three questions that helped me take control
When I was creating the MVP and later the architecture, I started using three questions before allowing AI’s output to move forward.
1. Did I ask AI to make this decision?
This was the question behind the Android and iOS plan.
I had asked AI to help me build an MVP. I had not asked it to change the platform decision. Building both platforms might be useful later, but it was not part of the first release.
So I began asking AI to show me everything it had added beyond my original request.
Did you include anything I did not mention? If yes, list it separately.
That gave me space to look at an idea before it became a commitment.
2. What does this change for the user and the project?
The next pause came when AI started suggesting architecture and product details.
AI can create a detailed solution very quickly. But if I do not understand why something is included, I may have a problem later when I need to explain it, change it, or support it.
So I ask: Why is this here? What problem does it solve? What happens if we do not add it? What will the user actually see or experience because of it?
I do not need to learn every part of coding to ask those questions.
I need to understand what a decision means for the product. What will this button do? What happens after the user taps it? Does this make the journey easier or more complicated?
AI is very good at explaining those things in simple language. That is where it helps me most.
3. Who is approving this before it becomes work?
This is the question that stops a suggestion from becoming a hidden decision.
For my app, I am the person who needs to decide whether a feature, architecture choice, or change belongs in the MVP. In a team, the owner may be a product manager, a lead, or the person responsible for the outcome.
The important thing is that the owner is visible before AI starts building the next step.
Each iteration, Git push, and version should move towards the user goal we have agreed. If a new suggestion is useful, I can save it for later. If it is needed now, I can approve it properly. If it is unclear, I can ask AI to explain it before I accept it.
I now separate decisions from suggestions
I use a small decision record when I am building with AI:
| Label | What it means |
|---|---|
| Decided | This is part of the current plan and can be built. |
| Suggested | AI has raised it; I will consider it, but it is not part of the work yet. |
| Needs understanding | I need to know why it matters and what it changes before deciding. |
It is simple, but it has made a big difference for me.
AI is still free to give ideas. In fact, I want it to. Some of its suggestions can be useful later or make me see a problem from another angle.
But I do not want AI to make the product bigger, change the goal, or add work without a clear decision from me.
I had a similar experience when I asked AI to challenge an app idea I was excited about. The questions were useful, but I still had to decide what belonged in the product. You can read that experience here.
The line I try to keep clear
I do not think AI should be limited to only following orders. Its suggestions are part of the value.
But there is a difference between AI helping me think and AI deciding what happens next.
The first can make the work better.
The second needs an owner.
Before I ask AI to move forward, I now ask one final question:
Is this my decision, or is this an AI suggestion that I have not reviewed yet?
That question helps me keep control without losing the useful ideas AI can bring.
