This is the first of two posts on how WRI’s Product Studio uses AI deliberately across its design and development process. Part 2 will cover how WRI’s Design System has been made AI-native to accelerate development while keeping quality consistent.When a digital product team can generate a working prototype in a single afternoon, the temptation to skip straight to building can be hard to resist.But we’ve seen how a quickly produced polished prototype can narrow thinking. It can create a false sense of momentum before the right questions have been asked, sending teams in a direction no one has consciously chosen and raising the cost of pivoting.The stakes are particularly high at WRI, which builds tools to help decision-makers understand and act on complex environmental and socioeconomic data. Building the wrong thing can mean far more than wasted time and money: it can shape how decision-makers interpret the evidence, which problems they prioritize and, ultimately, what they choose to do.Using AI to Make Discovery More Iterative Product Studio is the team that leads WRI’s digital product development, ensuring that the products in our portfolio are consistent and reliable in meeting users’ needs. Our approach has always been grounded in discovery first: understand the problem and form a hypothesis. Get a rough prototype in front of real users and stakeholders as early as possible. Learn, adjust. Then build. This isn’t a methodology we adopted in response to AI, but the foundation of WRI’s approach to product design and development. AI hasn’t changed that logic so much as raised the stakes of skipping it. To protect against the pitfalls of rushing into building, Product Studio has established lightweight prototyping practices that support the discovery phase. One of them is using AI to create many low-fidelity prototypes rather than a single realistic one, with the goal of sparking the right kind of conversation. Prototypes are worth creating when they help us determine whether: The idea reveals a real user need with a credible path to impact, rather than being just a technically interesting capability.Our assumptions about the user, data, or workflow hold up, and whether the product would break if one of them turned out to be wrong.The concept is still rough enough that users’ feedback will target the idea itself, rather than the execution.There are differences worth surfacing early, whether between team members’ views or between what users say and what they actually do. While Product Studio exercises restraint when it comes to using AI for functional prototyping, we rely on it to enhance early discovery. We use AI to create many low-fidelity prototypes rather than a single realistic one. We turn stakeholder conversations into sketch-like wireframes in hours, deliberately unpolished, so people critique the idea rather than the execution. We keep running summaries of the thinking behind a project (for example, concept description, problem solved, approach to concept testing and concept test results) so our reasoning stays visible and easy to revisit when an assumption changes. Using AI in this way widens the options under consideration before the team commits to any of them. This approach improves the quality of our conversations. Domain experts who would struggle to react to a written technical specification engage faster with something they can see. Identifying differences early means they are often quicker and cheaper to address. AI-generated sketch used with Cool Cities Lab users to test the idea of a concise summary for a selected area. Testing showed that participants clearly understood the concept as intended Our recent project on AI for Urban Cooling is a good example of our approach. We used AI to generate dozens of rough concepts for how it could support heat-adaptation planning, then homed in on five worth putting in front of real users. We took those ideas, which were deliberately early-stage because the goal wasn’t to validate a design but to learn which problems users most wanted AI to solve, to 46 practitioners. The concepts all tested well, so we asked participants to rank them from most to least valuable. A clear winner emerged — one concept drew 11 of the ‘most valuable’ votes and none of the ‘least valuable’ votes. That gave us the confidence to move it into prototyping, knowing it addressed needs that users prioritized rather than ones that sounded or looked good. High-fidelity prototype of the Selected Area Summary AI concept shown earlier as a sketch AI Doesn’t Change the Logic. It Raises the Stakes. WRI’s approach to responsible AI rests on three principles: be curious, be user-centered and be accountable. In the discovery phase, that means exploring where AI can be useful while staying grounded in evidence and user needs.That discipline matters even more now that teams can build convincing products before they have established whether they are solving the right problem, making it easier to commit to the wrong direction.AI can play different roles at different phases of product development. Once the direction is set and a product enters development, we use AI to build faster without drifting from what consistently works. Read the second part of this series, focusing on how WRI’s Design System is improving development, now with AI support.