I walked in expecting a night about speed. I walked out thinking about judgment. Nobody in the room argued that AI isn't making teams faster. It is. Every speaker treated speed as a given and moved on to the harder questions: what's worth building, who holds the quality bar, and what has to be true before something ships.
A polished prototype can make a weak idea look ready
This came up all night. It has never been easier to make something that looks finished. That's great for sharing an idea. It gets risky when a convincing mockup starts sliding toward production before anyone asks whether it should.
David Eisner, VP of Product Design at Datadog, gave us language I want every team to borrow. Some prototypes are for discussion. Others are designs ready for build. They can look the same, and teams get into trouble when they mix them up.
Learn what's worth building before you build it
Michelle Parsons, founder and CEO of Lora, kicked things off. Her starting point was that social products optimize for attention and shrink connection down to reactions, comments, and swipes. She wanted to make room for deeper connection, knowing people aren't going to stop spending time online.
What I loved was how her team tested that idea. They didn't start with the app. They made small, problem-led experiences, using astrology as a way in because it sparks curiosity whether or not you believe in it. They shared those through social content about careers, relationships, and family. Then they brought interested people into more experiences, a Discord, and virtual and in-person events.
It worked. She shared that the team saw about 7,000 engaged users in a week and a half, with 12% paying.
She was also refreshingly honest about AI. It helped a small team build and automate feedback loops, and it created pressure to move faster than felt wise. Guardrails, agent review, and humans in the loop kept them grounded.
Her talk was titled "From Community Signal to Product in Eight Weeks." The part I keep thinking about is what came before those weeks. The vision was on a whiteboard in December, and the team didn't start building the actual product until May or June. They spent that time learning.
The line that stuck with me: "We are humans building for real humans. No matter what tools we have in front of us."
Designers are steering now
Ethan Ye, design lead at OpenAI, drew on his work on ChatGPT Images to show how far image generation has come. Early models gave us obvious defects. Today's models handle realistic imagery, readable text, complex layouts, and precise edits, which makes them useful for real creative work. He was clear about what's still hard: staying consistent across repeated edits, and making output feel less generic and more like one person's taste.
Then he walked through a designer's day now. Coding agents, prototypes, simulators, and code sit right next to Figma. The designer sets goals and context, judges what comes back, and guides teammates and their agents. He compared it to a design critique that keeps going.
"We went from a lot of designing to a lot of prompting and now to a lot of steering."
His talk was called "When Everyone Has a Designer," and his point landed. When more people can make a design quickly, shared principles and thoughtful review matter more. Design judgment has to go up, not down.
Governance belongs earlier than you think
Our VP of Product, Robin Cannon, moderated the closing panel, "Speed without governance is a liability," with Rania Svoronou, Head of Consumer Experience & AI Design at Citi, and David. Robin framed the tension for enterprise and regulated teams in one line: "AI is also very good at creating things that are plausible but wrong."
From there he pushed past "does it work?" He asked where the checks belong, who's accountable, whether faster individual work leaves debt for the next person, and how teams tell an appealing prototype from something ready to ship.
Rania brought the view from a financial institution, where risk management runs through everything. Even a non-AI design change may need legal and compliance review. AI adds new questions about which tools to use, where a human should step in, and who owns a decision made with an agent. She's seeing stakeholders show up with AI-made mockups and ask designers to just build them. A mockup can explain a concept, she said, but it doesn't replace the work of looking at the user journey, the design system, and the required process.
"Everybody's just trying to go faster, but are we checking the debt we're leaving behind?"
David came at it from another angle: "We're putting out things faster but not the right things." His team is working on explicit design-quality checks and a shared understanding of goals before work moves forward. He also pushed back on the idea that AI makes every task faster. For him, a bigger win may be less friction when handing off context, because one person can take a piece of work further on their own.
Where this leaves us
Put the three talks side by side and you get a sequence I'd hand to any product team. Michelle's experiments ask what's worth building. Ethan asks who holds design judgment when everyone can make something. The panel asks what has to be true before that something ships.
That last question is the one we spend our days on at Knapsack. A design system and the checks around it are how a team agrees on what "right" looks like, so faster output still lands on the right side of the line.
So here's what I'd take back to your team this week: what do you want to decide before AI helps you build faster?
A big thank you to UXDX for programming this with us, to Datadog for opening up their office and Ryan Leffel from UXDX NYC for hosting, and to every speaker for sharing so openly.
If you missed New York, come join us at one of our upcoming events!