Judgment Is the Craft: Robby Stein's Three-Chapter Playbook for Product Managers in the AI Era

Stein speaks at the Lenny and Friends Summit in San Francisco, in September 2026, after almost 20 years of building products. He has founded startups and led teams through big changes at Instagram and Google Search, where billions of people now use products his teams created. He also admits to building things that zero people use today.

At Instagram his team shipped Stories and built the experiences around it, including Reels, feed ranking and direct messaging. At Google Search he now works on what search means in the AI era, where people can ask anything and get answers across video, audio and text. He says his talk is about the craft of product, not that transformation, and about how anyone can now do with a few AI agents what once took years and whole teams.

He says the PM's value used to include getting things done, creating energy and momentum, and acting like a project manager. In a world where almost anything can be built, he argues that the true value is judging and taste. Above all, the PM should be the absolute best at decision making, and Stein stresses that many decisions turn out to be wrong, which makes it a humbling craft.

Chapter one of Stein's three-part process is understanding people deeply. He draws on Clayton Christensen's *Competing Against Luck* and the idea that people hire products to do jobs for them. His Instagram team once held an offsite with the creators of the framework, where they practiced extracting insight through a story about a recent major purchase.

His own purchase was a bed. The method is to put yourself back in the moment, asking where you were, who you were with and what you were doing. In the store, his wife asked him to flop onto the bed, and she chose it because it wouldn't wake her if he moved. Stein notes a bed PM might have built a roadmap around cooling, eco or price, and would never have learned that the deciding factor was motion transfer.

He gives two examples from Google. People came to early AI chatbots for inspiration, such as choosing a couch, but chatbots were text-only, so the team invested early in multimodal understanding and retrieval so users could see couches and refine them across turns. In AI Mode, food and neighborhood questions lacked maps, ratings, opening hours and trustworthy descriptions. Bringing those in produced what he calls a magical product that is still among the most popular, judging by thumbs-up ratings.

On AI's role, he says he is still thinking it through. He is excited about sourcing jobs to be done from interview transcripts at scale, and about an agent that interviews users with the same methodology to learn why they use a product.

Chapter two starts from a seed idea and a first version that Stein says is usually not great. He compares the next stage to a training process: build a clear analytical view of the problems, rank them by priority, understand them deeply, fix them and ask again. Each change list is like an epoch that moves the product closer to market fit.

On Stories, adoption was uneven because many people didn't feel comfortable sharing. The team inverted the earlier question, asking why people weren't doing the thing, and grouped the answers. The top theme was audience, such as an ex, a teacher or an aunt being on the app, and a survey of thousands of people quantified the weights.

Stein says that without overcoming that burden, no creative tool would work. The team spent two years cycling through ideas, including confusing versions with favorites and a private profile, before finding Close Friends. They then cut the old product and relaunched it as a Stories experience, and it worked.

Reels first launched in Brazil as an ephemeral format, and it failed because people who put effort into a dance didn't want it to vanish in a day. The bigger insight was that creators wanted to be entertainers and build businesses, so the team made Reels a permanent format that can go viral. At Google, users who opt in can give feedback on responses, and in one shopping example a backpack answer should have asked about the user's son and his height. Making the experience collaborative became one of the biggest drivers of engagement and helpfulness.

Chapter three is fit and finish, which Stein calls craft and says is often overlooked. People can sense whether creators cared about what they made, he says, and two things drive that: whether the product works perfectly with no user pain, and how it makes people feel. Obsessing over the product and being a power user is part of it, but AI makes it possible to scale this.

He built an internal tool, using Google's Antigravity system, that uses the product as an agent. It goes to Google, types a question, asks a follow-up in the AI experience, records and screenshots the experience, and scores it against a comprehensive rubric. It can tell whether a homework answer renders math properly or emits odd characters, and whether a question about bioluminescence gets a visual tray instead of plain text.

The team has invested in agents that find broken or off-spec behavior and increasingly fix it themselves. Stein says that where you once needed a QA team, or a PM obsessively using every part of the product, that era is largely over.

For delight, Stein points to the redesigned Google search box, reinvented after 20 years and announced at I/O. It had to grow with the user and handle files, videos and photos, while still feeling like Google. The team added a gradient of color to signal the AI behind it, and a cursor that cycles through Google's colors, which he says people keep spotting and posting about on X.

Tapping into the box makes it jump out and load a ring, with haptics, so the whole thing feels as if it is getting ready. Through motion, color, feel and touch, Stein says, a product communicates the humanity behind it and gets a soul. He adds that many people overlook this part of the creative process.

He wraps up with the playbook. Understand people deeply and find their core needs, be rigorous and analytical about what must be fixed for the product to work, and then build it to be simple, great and a pleasure to use. Much of it is not new, but AI makes it possible in a new way, and he urges builders to think less about features, tech or distribution and more about the people on the other side of the screen.