Build Like a Lab: Josh Woodward on Finding, Testing and Killing Product Ideas in the AI Era

Host Lenny Rachitsky opens with a thesis: because the underlying technology is shifting so quickly, every company has to start operating like a labs team. That means trying a lot with the latest models, finding out what is newly possible before a competitor or startup does, and doing it without distracting everyone else in the company. Woodward, who oversees Google Labs, the Gemini app and AI Studio, says the premise is "spot on".

He adds a caveat that the devil is in the details. In some sense every startup is already a lab, so the real question is how to bring that methodology and energy into bigger or mid-sized teams. The goal is to cultivate both the ideas and the kind of people who thrive in that environment.

Asked where the best ideas at Labs came from, Woodward says there is no single answer, but the pattern is clear: they very rarely come from calendar-scheduled design sprints. They show up when people are swimming, in a hallway, or hacking over a weekend, and then someone comes back with a video saying "look at this amazing thing". Products like Notebook, Flow, AI Studio and Google Beam all started with very small groups who were curious and would not stop.

The team's main tool is a list of things that are "almost possible". Woodward compares the moment one crosses over to a phase shift, from solid to gas, and says he saw one two days earlier that he cannot stop thinking about. Once something crosses over, the team tries to form a group around it quickly, and then matches the new capability to a real user problem that people will pay to solve.

One area he is watching closely is immersive entertainment, where a model reaching a certain latency, language coverage and modality could unlock new experiences. He says Labs keeps a document of 82 predictions that all begin with "we believe the future is", fully expecting that nine out of ten, or more, will be wrong. In his view you have to hold a point of view about the future and live in it a little to chase it.

When Rachitsky points out that being excited by a demo is a low bar these days, Woodward agrees that dazzling demos arrive every other hour. He looks for universal user behavior or a shift on that side, and for a group of people who cannot live without the product and will pay for it. His examples of a product breaking through include the first NotebookLM podcast, which he heard as two AI hosts making British Parliament debates genuinely interesting, and Flow's predecessor Whisk.

On product-market fit itself, he says it is "way more art than science". In the early days it may be a couple of people with conviction who cannot fully explain it. What he does when showing early prototypes is watch people's eyes: "that is the metric", not a DAU-over-MAU ratio or a D7 retain. If their eyes light up and they lean in, you have hit a nerve with a pain point.

He also warns against falling in love with the product instead of the problem, because it takes three, four or five pivots to find something that works. When something does take off, as with Nano Banana in Gemini or the NotebookLM podcasts, he describes it as "swarming it and holding on for dear life", both exhausting and fun.

Woodward says people often expect the leader to make the kill decision, but his experience is the opposite: the team knows first. The signs are that passion starts to run out and you have tried everything you can think of without it working. People sense this the way they sense that a relationship is not working.

The leader's job is to create an environment where people can ask uncomfortable questions. He gives an example from the day before the interview, when a Gemini feature the team was excited about got early data back and the people who went out to look users in the eye saw it was not there. The PM said she did not think they should launch it, and Woodward replied-all to thank her. He says he trusts the team because they are the ones making it work every day.

Woodward admits it looks odd that one person oversees both zero-to-one Labs experiments and a Gemini app with over a billion users. He says they sit at opposite ends of the scale, and jokes that Labs gets excited about 10,000 users where most dashboards would not count that low. What unites them is the kind of builder they attract and a shared obsession with AI.

The team's priority order, he says, is "users first, Google second, and our product third". That applies whether it is Labs or Gemini in the third slot, and it keeps the teams from building for themselves. Labs is now entirely AI-driven, though it also explores adjacent things such as Google Beam, a kind of magic mirror that creates a 3D, hologram-like view of the person you are talking to.

On where a consumer breakthrough may come from, he points to entertainment and whether feeds will persist or technology can bring people closer together in real life. He also mentions messaging and chat, which no longer feel solved, and big scarce problems such as time, money and shared experiences. The right form factor, whether a chatbot or a personal agent, is in his view still to be invented.

Rachitsky asks a pointed question: why do other AI assistant apps connect to Gmail, Calendar and Docs and do magical things while Google has not? Woodward says parts of this exist today, for example in Gemini and Gemini Spark, where you can connect that data. He adds that the industry is converging on a UX simplification: forget modes and toggles and give users one prompt box where you say what you want and it just does it.

He says people regularly tell him they have all their stuff with Google and want the product to just work, and that there is reluctance to connect it to competitors' products. The vision of the last year or so is a Gemini that is personal, proactive and powerful, and he says more is coming soon. Rachitsky notes that Google's advantage is distribution, which makes growth easy once something works.

On what is over- and underhyped, Woodward calls model benchmarks overhyped, saying most people will never care about an Elo rating and the industry can lose sight of why a feature is useful to someone. The underhyped topic, he says, is principles and values: you speak through your products, which encode those principles, so the industry should talk more about what future it wants to build.

Instead of repeating the usual talk about taste and judgment, Woodward says he looks for mispriced signals. One is unlearning rate: not only how fast someone learns, but how fast they can learn something and then walk away from it when they see their product test differently. Another comes from a book about Roger Federer, who had "explosive endurance", the ability to go really hard over a long time while pacing yourself and taking care of the people around you.

The third is the ability to build and scale trust. Even as people direct swarms of agents and need fewer human collaborators, he thinks collaboration matters more, because lower execution costs mean more products and more people teaming up with different people. Labs teams used to be five to seven people and are now two to three, maybe four, like musicians who jam together and play different songs, with the songs being the products.

Asked whether PMs, designers or engineers are gaining, Woodward says PMs may adapt to other functions fastest because they are used to simulating a lawyer or a BD person in meetings. But the best people come from any function, including strategy, ops, BD and marketing. He pushes back on the idea that everyone is now just a builder: people still have a specialty, like a major and minor in college, and he tells newer PMs to code in service of becoming an exceptional PM.

To help people cope with a non-stop pace, Labs thinks in seasons. At any time 20 or 30 experiments run at once, and each team gets milestones and a rhythm. Woodward says the best lesson he has learned as a leader is to name the mode: say plainly that a team is in an all-out, explosive period, and be equally clear when it is not. After Google I/O, for instance, many Labs teams are told to just hack and rediscover what is new, because those weeks seed the next season.

Planning has shortened to a rolling six months, the window in which every lab is likely to ship some new pre-training model, with teams aiming to take a product to a meaningful milestone in 50 to 100 days. He says Gemini has a six-month roadmap with a multi-year vision, because a project can start from something seen 48 hours earlier. The key is to unblock people so they can "just let people cook".

Responding to a framework from Dan Shipper about dedicated frontier teams, Woodward says every company needs people whose job is to be at the frontier, and who have, as he puts it, eight $200-a-month subscriptions. The second ingredient is an environment where they can build. The third is the hardest: in history, large-company labs usually fizzle after three to four years because the parent cannot commercialize what they make. A lab needs people who can say that something looks like a toy now but will be essential in five years.

He describes how Labs now separates zero-to-one, one-to-ten and ten-to-a-hundred stages, because larger products such as Notebook, Flow and AI Studio risk crowding out early experiments. Zero-to-one is judged on eyeballs, while ten-to-a-hundred looks at conversion waterfalls and CPAs.

His three tips for setting up a lab are: give it independence, sometimes even reporting to the CEO; define whether you are graduating ideas into existing products or inventing new product lines, and be clear what winning looks like; and use the lab to challenge how the company works. Labs, he says, creates "good trouble", for example by piloting a builder job ladder before Google rolled it out.

Woodward says Labs looks for people who cannot stop building, are intellectually curious, and are obsessed with solving people's problems rather than their own ego. He emails new joiners a document called "Labs in a nutshell" in their first week. In interviews, someone saying "let me show you this thing I just built" gets instant credibility, as do people energized by unstructured, unknown situations where the way forward is unclear.

On org design, he says one common mistake is rushing to consolidate and make everything clean; sometimes you have to live in the messiness and see what emerges. The other is hiring too many people too fast. He recalls a project that reached 30 engineers with zero product-market fit but an inspiring vision, and says growth in headcount is mostly a vanity metric.

On Logan Kilpatrick's question of whether Google will have 10,000 products or five in five years, Woodward says he keeps changing his answer. He is increasingly unsure that we will think of products the same way, and imagines more bespoke experiences generated by models on common surfaces. Labs is running experiments to explore both consolidation into super apps and fragmentation.