AI makes it easier to imagine what a product could do. But who decides what it should do?
Artificial intelligence has made an extraordinary difference to the way we approach product development.
We can explore ideas faster, analyse requirements, identify potential risks, generate user journeys and evaluate technical alternatives in minutes rather than days.
AI can take an enormous amount of information and reduce it into something understandable and actionable.
That is one of its greatest strengths.
But I think we are beginning to encounter an equally important problem.
AI is exceptionally good at reducing noise. It is also exceptionally good at creating it.
And in product development, that distinction matters enormously.
1. When Possibilities Become Noise
Imagine asking AI to help design a relatively simple product.
You explain the problem, describe the intended users and ask it to identify functionality worth considering.
Within seconds, AI produces a comprehensive response.
It suggests the core features you expected. Then it adds a few more.
Perhaps you should consider personalisation. Perhaps predictive analytics would be useful. Perhaps users would benefit from automated recommendations, advanced notifications, integrations or an intelligent assistant.
Every suggestion sounds reasonable.
Some may even be genuinely valuable.
But here is the problem.
Something being potentially relevant does not make it important.
AI can generate an almost unlimited number of plausible possibilities. It does not automatically know which possibilities deserve a place in your product.
That decision requires context, evidence and judgement.
Without those, we risk mistaking a comprehensive list of suggestions for a well-considered product strategy.
2. AI Can Turn Uncertainty Into Apparent Requirements
One of the subtler risks of AI-assisted product development is how easily a suggestion can begin to look like a requirement.
Consider the language AI often uses:
- You may also want to consider…
- It could be beneficial to include…
- A potential enhancement would be…
- For a more comprehensive solution, you might add…
- Another important consideration is…
These are not necessarily bad suggestions.
In fact, exploring alternatives is one of the most useful ways to work with AI.
The difficulty arises when we stop distinguishing between an idea, an assumption, a validated requirement and a genuine product need.
A feature suggested by AI can quickly find its way into a backlog. Once it appears in a backlog, it can acquire an estimate, a priority, acceptance criteria and eventually a development commitment.
At that point, something that began as a possibility has become part of the product's expected scope.
Yet nobody may have established whether users actually need it.
AI-generated possibilities are inputs to product discovery. They are not evidence of product demand.
That distinction needs to remain explicit.
3. More Functionality Does Not Necessarily Mean a Better Product
Every additional feature has consequences.
It requires design decisions, implementation, testing, documentation, maintenance and support.
It may introduce new dependencies, additional user interactions and unexpected behaviours.
Even when the feature works perfectly, it can make the overall product harder to understand.
Consider an appointment-booking application.
Its fundamental purpose might be to allow someone to find an available appointment, book it, reschedule it or cancel it.
AI could suggest intelligent scheduling, predictive recommendations, loyalty programmes, personalised dashboards and automated engagement campaigns.
Some of those capabilities might make sense for particular businesses.
But if the primary user simply wants to book an appointment quickly, additional functionality may create friction rather than value.
The product becomes more sophisticated without necessarily becoming more useful.
And that is a quality problem.
Quality is not measured by how much functionality a product contains. It is measured, in part, by how effectively that functionality serves its purpose.
Sometimes the highest-quality product decision is to leave something out.
4. Feature Noise Begins With Decision Noise
I think there are two distinct forms of noise worth recognising.
Feature noise is unnecessary or poorly justified functionality that makes its way into a product.
Decision noise is the accumulation of suggestions, possibilities, assumptions and considerations that makes it harder to identify what actually matters.
The second often creates the first.
When AI presents 25 plausible ideas, a product team must evaluate them.
But if every idea receives equal attention, the team can spend considerable time investigating things that were never particularly important.
Worse, teams may begin prioritising ideas because they sound innovative, are technically achievable or appear repeatedly in AI-generated recommendations.
This creates an illusion of progress.
The backlog grows. The product roadmap becomes more ambitious. More functionality is delivered.
But the underlying product may not be improving at the same rate.
We have increased activity without necessarily increasing value.
The challenge is no longer simply generating good ideas.
It is preventing weakly justified ideas from becoming commitments.
5. How Do We Decide What Actually Matters?
This is where product judgement becomes essential.
Before accepting an AI-generated recommendation, I believe we should ask five fundamental questions.
Does it support the product's purpose?
Every product should have a clear reason for existing.
A feature should contribute meaningfully to that purpose, rather than simply expanding what the product can do.
Is there evidence of a genuine need?
A plausible use case is not the same as demonstrated demand.
User research, observed behaviour, customer feedback, operational evidence and business objectives should inform the decision.
AI can help analyse that evidence, but it should not substitute invented assumptions for it.
What happens if we don't build it?
This is one of the most useful questions in product development.
If the answer is that very little changes for the user or the business, the feature may not deserve immediate attention.
There are exceptions, particularly where security, accessibility, safety or regulatory obligations apply.
But the question forces us to consider consequence rather than possibility.
Does the value justify the complexity?
Every feature has a cost beyond its initial implementation.
We need to consider its effect on usability, maintainability, reliability, testing and future product decisions.
The cost of complexity is cumulative.
Is this important now?
Something can be a good idea without being the right idea today.
Product judgement includes sequencing, restraint and the willingness to defer functionality until there is sufficient evidence to justify it.
These questions are not intended to discourage innovation.
They help ensure that innovation serves a purpose.
6. Using AI to Reduce Noise Rather Than Amplify It
The solution is not to stop asking AI for ideas.
It is to change how we use it.
Instead of asking:
"What other features should I add?"
We can ask:
"Which of these features directly supports the product's primary user outcome, and what evidence would justify including each one?"
Instead of:
"What else should I consider?"
We can ask:
"Which considerations could materially affect safety, usability, business value or delivery risk? Separate established concerns from speculative possibilities."
And instead of:
"How can I make this product more comprehensive?"
We can ask:
"What can I remove or simplify without compromising the product's essential purpose?"
This is a fundamental change in approach.
We are no longer using AI primarily to expand the solution.
We are using it to challenge, refine and simplify the solution.
AI becomes a tool for disciplined product thinking rather than an engine for generating additional scope.
7. The Relationship Between Product Judgement and Engineering Judgement
In my earlier discussion about engineering judgement, I explored an important distinction:
A piece of code can be technically excellent while implementing functionality that should never have existed.
That principle extends directly into product development.
Engineering judgement helps determine whether a solution is technically appropriate, reliable, maintainable and safe.
Product judgement helps determine whether the solution is necessary, valuable and consistent with the intended user experience.
Quality judgement brings these perspectives together.
It asks whether we are solving the right problem, whether we are solving it appropriately and whether we have sufficient evidence to trust the outcome.
AI can support all three disciplines.
But generating a technically feasible solution is not the same as validating the need for that solution.
And generating a list of potential improvements is not the same as making a good product decision.
8. Simplicity Is an Outcome of Good Judgement
Simplicity is often misunderstood as the absence of sophistication.
I see it differently.
A simple, effective product can be the result of considerable analysis, experimentation and difficult decisions.
It may take more judgement to remove five unnecessary features than to implement them.
It may require more experience to recognise that a complicated workflow should be redesigned rather than automated.
And it may take more confidence to reject an impressive AI-generated capability because it adds little meaningful value.
Good product development is not about implementing every possibility.
It is about making deliberate choices.
The ability to say no is as important as the ability to build.
9. The Real Opportunity for AI-Assisted Product Development
AI gives us the opportunity to work differently.
We can spend less time generating documentation, exploring basic alternatives and performing repetitive analysis.
That should create more capacity for the activities that require deeper judgement: understanding users, challenging assumptions, evaluating consequences and making informed decisions.
But that benefit is not automatic.
If we use every efficiency gain to generate more requirements, more features and more complexity, we may simply create a larger volume of work.
The opportunity is to use AI to improve the quality of our decisions, not just increase the quantity of our output.
That means asking AI to identify uncertainty, expose assumptions, compare trade-offs and challenge unnecessary complexity.
It also means retaining accountability for what ultimately enters the product.
Final Thought: Just Because We Could Doesn't Mean We Should
AI is making it increasingly easy to build things.
It is also making it increasingly easy to imagine more things to build.
Neither capability automatically creates product value.
The real challenge is deciding what deserves our attention, what deserves investment and what deserves a place in the product.
A good product is not the sum of every useful idea anyone could imagine.
It is a deliberate collection of capabilities that work together to solve meaningful problems.
AI can generate possibilities. Product judgement creates focus. Quality comes from knowing the difference.
And perhaps one of the most valuable things we can ask AI to do is not tell us what else we could add, but help us understand what we can confidently leave out.
Unicorn Inspira | Strategy Meets Quality
Perspectives on product strategy, quality governance and responsible AI-enabled delivery.