AI is very good at suggesting features.
Ask an AI model what is missing from your product, and it will usually give you a long list. Ask it how the product could be improved, and you will get another list. Continue the conversation, and the list will keep growing.
Some suggestions may be useful. Others may solve problems that your users do not have. Some may introduce new concepts, settings, workflows, roles, and dependencies that make the product significantly more complex.
Used naively, AI can become a feature-creep machine.
The problem is not that AI generates bad ideas. The problem is that generating ideas is much easier than deciding which ideas belong in the product.
More features do not necessarily create a better product
A product is not simply the sum of all the useful features we can think of.
Every feature has a cost:
- Users must understand it.
- The interface must accommodate it.
- Developers must implement and maintain it.
- It must be tested and validated.
- Support teams must explain it.
- Documentation must cover it.
- Future changes must remain compatible with it.
- Other features may need to integrate with it.
A feature can be valuable in isolation while still making the overall product worse. This is especially relevant when using AI. AI can identify many possible use cases, edge cases, configuration options, integrations, and improvements. It does not naturally experience the accumulated complexity of the resulting product.
The goal of product management is not to include as many useful features as possible. The goal is to find the right set of features:
- Not too many.
- Not too few.
- Simple enough to understand.
- Complete enough to solve the intended problem.
Simplicity should not mean removing everything. A product that is too limited may fail to solve the real problem. But complexity should have to justify itself.
The product manager as a curator
When AI is involved in product development, the curator role becomes increasingly important. A curator does not only decide what should be added. The curator also decides:
- What should not be added.
- What should be simplified.
- What should be postponed.
- What should be removed.
- Which ideas need more research.
- Which problems should be solved outside the product.
This requires active judgment.
AI can help generate and explore alternatives, but it should not decide the product direction simply by producing a convincing feature proposal. A well-written proposal is not evidence that a feature belongs in the product.
Adding a feature is only one possible outcome. A product process must also support rejecting ideas, simplifying proposals, and removing existing functionality. Otherwise, the product will become more complex over time.
Go deeper than the AI-generated proposal
When a change is considered, we need to go into the details. This means applying domain knowledge rather than accepting the first reasonable-looking proposal generated by AI.
When we already understand the domain, we should use that knowledge to challenge the suggestion:
- Is this how users actually work?
- Is this a frequent problem or an unusual edge case?
- Is the proposed workflow natural?
- Does the feature introduce new terminology or concepts?
- Could the problem be solved more simply?
- Will users understand when and why to use it?
When we lack domain knowledge, the answer should not be to let AI fill in the gaps. We should spend the necessary time and energy to understand the problem. That may include:
- Talking to users.
- Reading relevant standards and documentation.
- Studying existing workflows.
- Trying competing products.
- Building small prototypes.
- Testing the process ourselves.
- Looking at how similar problems are solved elsewhere.
The purpose is not only to collect information. It is also to build enough understanding to form our own judgment and gut feeling about the product. AI can support this research, but it cannot replace the experience and context required to make a good product decision.
Competitors are input, not specifications
Competitor research is useful, but it can also create feature creep. It is easy to create a comparison table, identify everything competitors support, and treat the missing items as gaps in our own product.
That is usually the wrong conclusion.
Competitors may:
- Target different customers.
- Support different workflows.
- Carry years of legacy functionality.
- Accept more complexity.
- Solve problems that are outside our product scope.
- Include features that their own users rarely use.
The question should not be:
Which competitor features are we missing?
A better question is:
Which user problems are we failing to solve, and what is the simplest suitable solution?
Sometimes the answer will be to add a feature. Sometimes it will be to improve an existing one. Sometimes the correct decision will be to do nothing.
AI should increase the number of options, not the number of features
AI gives product teams access to more ideas, alternatives, and possible solutions than before. That can lead to better products, but only when combined with strong curation.
The value of AI is not that every reasonable suggestion can now be implemented faster. The value is that we can explore more possibilities before deciding what the product should become. The product manager's responsibility is still to make that decision.
AI can propose.
The curator must decide.