Vishal Kapoor’s 10 rules for building honest products with AI

Rules for building products people can trust with AI

For Affirm SVP of Product Vishal Kapoor, building with AI is as much a trust problem as a technical one. His team builds financial products that sit directly between customers and their money, in a category where the emotional stakes are high and the incumbent players have trained people to expect obfuscation. That context shapes how Kapoor thinks about AI adoption: it has to make teams faster and more creative without eroding the honesty and clarity that keep customers coming back.

Kapoor brings an engineer's instincts to the challenge, starting from first principles on every problem rather than accepting defaults. When he pushed his team on why customers are offered three payment plan options instead of five or one, the point wasn't to change the number—it was to force a re-examination of assumptions. AI can accelerate exploring alternatives and test variations quickly, but Kapoor argues that the critical insight still comes from human judgment and the pushback that happens when people with different perspectives disagree.

That grounding in customer reality carries through everything else. Kapoor tracks social media and app store reviews directly and makes a point of talking to strangers who recognize the Affirm logo on his shirt. Internally, the team has Pluto, an AI-powered tool that Kapoor says he can query with questions like "How have our customers been disappointed with us in the last 30 days?" But he's clear that dashboards only point the way. Finance is deeply emotional—anxiety, relief, and frustration all surface at checkout—and teams need to experience those emotions firsthand to build products that respect them.

Prototype early, audit everything

When Affirm first piloted AI tools, Kapoor says there was understandable hesitation across the team. The reality that emerged was more practical than either of the extreme narratives: AI didn't make anyone 10x faster overnight, and it didn't replace anyone either. Instead, it worked best as another teammate—one that helps turn customer insights into concrete prototypes much faster.

Because Affirm's checkout flows span web, mobile, and desktop, the team runs different flights of experiments across modalities. Auditing every screen for every use case could take months, so the team uses tools including Figma Make to catch broken patterns early. When they need to update an interaction design, AI helps identify outdated patterns across all surfaces in days rather than weeks. That shift moves cycles away from engineers and back to designers and product managers, which Kapoor says unlocks both velocity and creativity.

The hard part is moving beyond the happy path. Kapoor's team found this when working on helping customers pick between 6-week, 6-month, and 12-month payment terms. Research showed customers fall into three distinct groups: people optimizing for 0% APR or minimal interest, people who want the shortest plan, and people managing cash flow with the smallest monthly payment. The solution was clear badging for each objective—but the question of which badge to show, to which person, on which checkout, on which device, was a complicated problem at scale.

Rather than taking six weeks to validate an idea in production, Kapoor says his team can now experiment in days with realistic prototypes. The key is validating before writing production code. During reviews, the team doesn't just watch the success metrics—they track counter metrics too. In the badging example, that meant watching for customers who completed a purchase out of confusion rather than intent. An increase in conversions alone would look like a win; complaints and refunds would reveal the real story. Kapoor says tracking customer dissatisfaction metrics during product reviews gives them a truer picture of how features perform in customers' hands.

Blurring roles, keeping guardrails

The traditional product development waterfall—PM writes a PRD, designer creates mockups, engineer codes, QA validates—is breaking apart at Affirm. Kapoor sees it as a deliberate strategy, not an accident. AI is democratizing skills across all stages: engineers can modify design artifacts, product managers can push code into pre-production to understand the feel of a feature, and designers can author requirements or even production-ready code.

This shows up in concrete ways. Product managers at Affirm used to review designs in Figma, not create them. After the design team showcased Figma Make, PMs started producing their own prototypes. Kapoor recounts one example involving a Shopify collaboration where a disclosure needed to move and a preset payment term needed to change in the Shop Pay Installments flow. The next day, a PM had a prototype in his inbox—a visual foundation that became the basis for the product requirements document. The team still performs full due diligence, but AI accelerates drafting, critique, and spotting errors and missed states before launch.

Kapoor treats tinkering as something to be actively encouraged and managed. Affirm founded an AI enablement steering committee to keep teams informed of the latest developments and triage requests for new tools. Security and procurement set up pilots for teams or individuals who want to try something new, with the most advanced users helping identify which tools are ready for broader rollout. Once a tool is generally available, the company tracks active usage, publishes reports on which teams are using what, and runs qualitative surveys to understand impact.

Kapoor acknowledges the shift is still in its "messy middle." Tools like Claude and Cursor are producing production-ready prototypes that reduce front-end development time, particularly because Affirm's design system maps directly to its codebase, preserving design intent from concept through launch. But he's also clear that while AI dramatically expands the number of ideas the team can explore, the joy of building still comes from the human end-to-end process—uncovering customer needs, finding a differentiated way to address them, and navigating the iteration cycles that follow.

Rule 10: Keep “AI slop” out of the product

Affirm’s core mission is built on trust, which Kapoor says comes from simplicity and transparency. The current AI environment makes it easy to generate a flood of features and experiments. The real challenge is using AI to cut through that noise, delivering something that is simple, transparent, and directly tied to real customer problems and original perspectives.

In practice, this means resisting the urge to add complexity. The competitive edge now lies in distillation—taking the raw capabilities of AI and shaping them into a focused experience. Simplicity is not the default state of AI-powered products; it is the result of disciplined, ongoing effort.