Why every company now needs to think and operate like a lab team | Josh Woodward (Google Labs, Gemini, AI Studio)

I’m talking with Josh Woodward, who leads Google Labs alongside the Gemini app and AI Studio. As the technology moves this fast, every company has to play with new models, run experiments, and spot opportunities before a startup eats its lunch. Does that resonate?

Absolutely. Every startup already behaves like a lab; the harder question is how larger teams borrow that energy without losing it. You need an environment that grows unusual ideas, with people who thrive when the path isn’t obvious.

You’re not pointing people toward a brainstorm or scheduled design sprint. Where do good ideas begin?

There’s no answer key. The best ideas tend to appear when curious people are messing around, then come back saying, “You have to see this.” Stay near the frontier, keep a list of nearly feasible things, and notice when one crosses over. Then match that shift to a real user problem and something people might pay for.

We form views of possible futures across creativity, software, work, knowledge, and entertainment. You can track delay, languages, and modalities, but you also have to leave the Bay Area, look at classrooms and other cities, and see behavior changing. Most predictions will miss; you still have to put yourself out there and inhabit the future.

NotebookLM clicked when two teammates played me an AI conversation about British parliamentary proceedings, and it was genuinely compelling. An earlier image project that became part of Flow showed people combining images and animating them with accessible creative control. Other exciting things become products that just aren’t good. It’s a numbers game.

So deciding whether a bet is real is more art than the clean, data-driven product-market-fit story people want.

Very much so. Early on, I want prototypes outside the building, and I watch faces: do people lean in? That matters more than a dashboard. Love the problem, not your first solution; it may take three or five turns. When something catches, you swarm it. And teams usually recognize a failing project before the leader does. We held back a Gemini feature because the early reactions weren’t there, and I was proud the PM called it.

You also run Gemini at enormous scale. Where do you see room for consumer products?

Gemini is huge while a Labs project can celebrate ten thousand users, but both attract that early-builder heartbeat. Consumer feels wildly alive: entertainment, messaging, and personal help are all reopening. There are big constraints around time, money, memories, and real-world experiences. The final form—chatbot, agent, something else—hasn’t been invented yet. We’re working toward a more personal, proactive Gemini that can act across your information with fewer toggles.

Values are under-discussed. Products speak for us, whether we talk about it or not. Benchmarks are useful proxies, but most people don’t care about an ELO score. I’d rather ask: can we turn these capabilities into something somebody really wants?

For people, I look at unlearning rate: can you pick something up, test it, and let it go when reality says you should? I also like explosive endurance—going hard without burning out the people around you. Small teams may be two or three people directing agents, but I don’t buy that everybody becomes a generic builder. Keep your major; use new tools to become exceptional at it.

Give teams seasons: a full-on stretch around a model launch, then time to hack, rediscover, and build for fun. It can look unproductive, but that’s where next season’s seeds come from. Planning is rolling, maybe six months out, with meaningful movement in fifty to a hundred days. People need enough freedom to just cook.

The challenge is keeping a lab independent enough to be weird while making sure the company can use what it discovers.

Exactly. Give it room—sometimes close enough to the CEO that it isn’t buried in a business unit—but be clear whether it graduates work into products or creates new categories. Don’t make the org too clean or hire huge teams because headcount feels good. I’ve had thirty engineers and no product-market fit.

We look for people who can’t stop building, stay curious, care about users over glory, and leave you with more energy than they took. Keep it a little dorky, too: tiny rakes for reclaimed compute, gold Band-Aids for fixing irritating details, LLM Whisperers with giant ears. Those symbols make the work visible. Thanks, Lenny—this was great.

I think this will help a lot of people keep building at the edge. Josh, thank you so much for being here.