Source: raw/Why_every_company_now_needs_to_think_and_operate_like_a_lab_team_Josh_Woodward_VP_Google_Labs.md — Creator: Lenny’s Podcast (Lenny Rachitsky), guest Josh Woodward, head of Google Labs, the Gemini app and AI Studio, 16 years at Google · URL: https://www.youtube.com/watch?v=3I6QYkccd3o · Platform: YouTube (fetched 2026-10-21).

Woodward runs one of the largest corporate labs teams, the group behind NotebookLM, Flow and AI Studio, and gives specific operating rules rather than general enthusiasm: keep a list of what is “almost possible”, judge early prototypes by people’s reactions rather than retention metrics, let the team call when to kill an idea, and staff new projects with two or three people. He also explains why most corporate labs fail within a few years and gives three tips for setting one up. It complements Dan Shipper’s labs-versus-product talk in Lenny’s Summit 2026. These are one leader’s practices, told in interview.


Key Takeaways

  • Ideas come from small obsessed groups. “You can’t microwave good ideas”; Labs products “get started by very small groups of people who are just obsessed and usually very curious about something.”
  • Keep an “almost possible” list. The team tracks capabilities that are nearly there, and acts when one crosses over: “it’s almost like a phase shift.” Labs also keeps a doc of “82 predictions that all begin with we believe the future is”, knowing most will be wrong.
  • Early product-market fit is in people’s eyes. “You’re looking at people’s eyes. That is the metric … it’s not a DAU over a MAU, it’s not a D7 retain.” Expect “three, four, five pivots” before you have something, so fall in love with the problem, not the first solution.
  • The team knows first when to kill. “The team usually knows before the leader knows.” He publicly thanked a Gemini PM who argued against launching a feature after weak early data.
  • Teams got smaller. A new Labs team used to be “five to seven people. Now, it’s about two to three people. Uh maybe four”, with roles blurring so one person can do PM, code and design.
  • Short roadmaps, fast milestones. He keeps a “6-month road map” with a multi-year vision; the best teams go “from like idea to some meaningful milestone in like 50 to 100 days.”
  • Benchmarks are overhyped; principles are underhyped. He wrote an internal doc of Labs principles over the summer: “why are we building what we build?”

Details

Hiring signals.

  • “Unlearning rate”: “how fast can they learn something and then walk away from it?”
  • “Explosive endurance”, a phrase from a book about Roger Federer: work that is “both a marathon and a sprint”.
  • Building trust: people who “can build trust, that can scale trust”, because that “underpins any kind of great collaboration”. He raises it while noting that people now collaborate with “a swarm of agents, but also with people”.

Roles. PMs adapted fastest to building, but specialists still matter, like a college major and minor; “there’s a risk if everybody [be]comes a builder that you lose a lot of respect and appreciation and obviously the expertise that comes from those specialties.”

Name the mode a team is in. Some periods are “explosive time”, for example five products launching to showcase a new model. After Google I/O, “June is like, hey, just go hack”, which he defends as where “the new seeds are coming from for like the next season”. Being clear when a team is not in sprint mode matters as much as declaring it.

Why corporate labs fail. “A large company who starts a lab, the lab is usually ineffective and fizzles out after about 3 to 4 years”, or the company can’t commercialize what the lab makes. A lab needs people who live at the frontier (“the people who have like eight $200 a month subscriptions”), an environment where they can build, and someone who bridges to the core business so the lab is not dismissed as “some weird R&D group always coming up with like gimmicks and toys”.

Lifecycle stages. Google Labs distinguishes zero-to-one, one-to-ten and ten-to-a-hundred stages. Zero-to-one looks at “eyeballs”; ten-to-a-hundred looks at conversion funnels and CPAs. Bigger graduated products such as NotebookLM, Flow and AI Studio risk crowding out the zero-to-one work.

Three tips for setting up a lab.

  1. Independence: report into enough distance from the core business, in some companies to the CEO.
  2. Decide whether the lab graduates features into existing products or invents new product lines (Google Labs started as a graduation lab).
  3. Use the lab to change how the company works (“good trouble”): Labs piloted a “builder job ladder” before it rolled out across Google.

Mistake: hiring too fast. On one past project “we got to like 30 engineers … and we had zero product market fit but a very inspiring vision.”

Where Gemini is going. “Forget the modes and toggles. Just like one prompt box. Tell it what you want and it just does stuff for you.” Gemini Spark can already connect Google data, and Google is working on “personal intelligence”: a Gemini that is “personal, proactive, powerful”.

Rituals. A miniature rake and a cash award for people who free up TPUs; a golden band-aid for fixing paper cuts; giant ears for the “LLM whisperer” who finds surprising model behaviour; and an unannounced meeting where four or five leaders tell someone why they are appreciated (“This is not a bad meeting”).

Try It

  1. Start an “almost possible” list for your field and review it when a new model ships.
  2. Demo prototypes in person and note reactions before you look at usage dashboards.
  3. Ask each project team directly whether it should continue, and reward the person who says no.
  4. Staff the next experiment with two or three people and set a milestone within 50-100 days.
  5. Label each team’s current mode (sprint or exploration) so slower periods are deliberate.
  6. For agencies: a small internal labs pair can test new AI tools for client work, with someone responsible for bringing what works into the delivery team.

Open Questions

  • The “3 to 4 years” lab lifespan comes from his reading of history; no data source is named.
  • Product names in the captions (NotebookLM as “notebook”, Gemini Spark) were not checked against Google’s announcements.