Lenny's Podcast · Oct 11, 2026 · 1h 2m · 43m saved
With Josh Woodward, Vice President, Google Labs, who also runs the Gemini app and AI Studio
Google Labs now staffs a new project with two or three people, Josh Woodward said. It used to assign five to seven.
Most large companies that start a lab see it fade within three or four years, he said, or never manage to sell what it invents. His group has turned NotebookLM, Flow and AI Studio into products with large and growing user bases, and he also runs the Gemini app, which has more than 1 billion users.
"We used to say like it's about five to seven people. Now it's about two to three people." — Josh Woodward
Josh Woodward, Vice President at Google Labs, on Lenny's Podcast, oversees what Lenny Rachitsky, who hosts the show, called the longest-running and largest labs team in the world. His job is to experiment with new AI products and scale the ones that work, and in the interview he described how his teams find ideas, judge them and shut them down.
The full interview is covered here so you can skip it. 1 hour 2 minutes of audio, 19 minutes of reading.
Here are the 19 lessons that matter.
Key Takeaways
Good ideas rarely come from design sprints or brainstorms set on a calendar
Labs keeps a written list of what is almost possible and acts when a model crosses the line
One document holds 82 predictions about the future, most of which he expects to be wrong
Early product-market fit shows in people's eyes, not in daily active users or retention
The team usually knows to kill a project before the leader does
New lab teams are 2–3 people instead of 5–7, because building a product costs less
Josh hires for an "unlearning rate": how fast someone drops what they learned once a product test shows it is wrong
Not everyone should become a generic builder; people still need a specialty, like a college major
Most corporate labs fade in 3–4 years or never commercialize their work
His fix: independence from existing business units, and a clear goal of graduating features or inventing new products
Model benchmarks are overhyped, he said, because most people will never care about a model's ranking
His team plans six months ahead, and the best teams reach a real milestone within 50 to 100 days
1. Every Company Needs a Lab
Lenny, who also writes Lenny's Newsletter, opened with his premise. AI models change so quickly that every company now has to work like a labs team: try many things with the latest models and find what has become possible before a competitor or a startup does.
Josh agreed, and said the hard part is doing it inside a larger company
He said every startup is already a lab. The question is how to bring the same methods and energy into medium-sized and large teams, and how to build an environment where both the ideas and the people who generate them can thrive.
