Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)
In a Nutshell
Every company needs an independent lab team positioned at the technological frontier to experiment with emerging AI capabilities before competitors do, as traditional corporate structures kill innovation within 3-4 years. The best ideas emerge from obsessed small teams working on side projects rather than formal brainstorming, with success measured by whether prototypes make people's eyes sparkle rather than traditional metrics. Labs must maintain independence, focus on user-first outcomes over lab glory, and use their position to challenge how the entire organization operates.
These notes were generated by AI and may contain inaccuracies.
Every company needs a dedicated team at the forefront of technology. These teams maintain eight $200-a-month subscriptions to stay current with emerging tools. The core mandate is experimenting with the latest models to discover what's now possible before competitors or startups claim market share. Companies must create space where strange ideas can grow.
Historical patterns show that laboratories within large companies typically become ineffective and fade away after three to four years. The critical failure is placing labs under existing business units rather than granting them special independence.
Josh Woodward, VP of Google Labs, has witnessed more ideas proposed, tested, and prototyped than most professionals. Good ideas rarely emerge from quick design sessions or scheduled brainstorming meetings. They don't originate from traditional corporate processes.
The best ideas emerge from diverse sources—often while swimming, walking corridors, during weekends, or on breaks. Technical teams at Labs build experiments and share video clips saying "Look at this amazing thing." The pattern is to remain alert to what catches attention.
Ideas come from interesting people and their side projects rather than the ideas themselves. Notebook, Flow, AI Studio, and Google Beam all originated from small groups obsessed and curious about specific problems who refused to stop working on them.
Teams track what is "almost possible"—maintaining a list of capabilities that haven't quite crossed the threshold. When something transitions from impossible to possible, it resembles a phase change. The approach involves positioning yourself at the technological frontier, monitoring experiments, and recognizing when something becomes viable.
Labs maintains 82 predictions beginning with "We believe that the future is" or "We predict that." Making 82 predictions is acknowledged as dangerous since most will be wrong, but having a view of the future is essential. The team pursues the future by positioning themselves at the center of the experience.
Current focus areas include the future of entertainment and immersive experiences. Teams monitor technical capabilities (response time, language support, media handling) alongside user behavior trends. Research extends beyond the San Francisco Bay Area to cities and classrooms to observe behavioral changes.
When evaluating whether an idea merits pursuit, the signal is surprise and personal excitement. The question shifts from technical capability to user behavior: can you reach people who cannot live without it and will pay for the value?
The Notebook LM podcast concept crystallized when two team members approached Josh outside a room around 6:30 PM on a Tuesday. They played a recording of British Parliament debates—content not known for excitement—yet two AI hosts discussed it entertainingly. This triggered the decision to pursue the product.
Google Flow emerged from an earlier project called Wisk, which allowed mixing multiple images and animating them. The level of creative control and accessibility made it seem newly possible.
Product-market fit assessment is more art than science. In early stages, often just two team members hold conviction that "the future will look like this" without being able to fully explain why.
The practical method involves showing early prototypes to people and observing their eyes. The true measure is whether eyes sparkle and people lean forward—not daily active users, daily-to-monthly ratios, or seven-day retention metrics.
Teams focus on falling in love with the problem rather than the product solution. Products typically require three, four, or five substantial iterations before reaching the right form. Loving the product first leads to bad outcomes.
When the "Nana Banana" clip went viral on Gemini and Notebook LM podcasts gained traction, teams rushed toward these moments and held onto them tightly.
Teams typically recognize when projects aren't working before leaders do. Signs include waning enthusiasm and exhausting all conceivable approaches without success.
The leadership approach creates environments where teams can voice uncomfortable questions. A recent Gemini feature that the team was excited to launch received initial data showing no enthusiasm. The product manager stated they shouldn't release it, and leadership supported this decision.
Gemini has over one billion users globally. Labs projects might celebrate 10,000 users—a number other product dashboards wouldn't register. Despite the scale difference, commonalities exist: both attract builders drawn to early-stage work, both share obsession with AI, and both operate under the priority framework of user first, Google second, product third.
All current Labs work depends on AI. Four to five years ago, other areas were explored, but the decision was made to focus entirely on artificial intelligence going forward. However, work extends beyond software—Google Beam is a hardware project creating 3D holographic versions of people for remote conversations.
AI is opening significant consumer opportunities despite previous assumptions that consumer markets were saturated. Key questions include whether entertainment will continue as different types of news feeds or whether technology can bring people together in the real world for shared experiences.
The messaging and chat area appears to have resolved, with many different experiments taking place both inside and outside of Google. This represents a completely different field that may no longer be as stable as it once seemed. When considering major problems people face, the discussion centers on rare things like time on Earth, how much money people have and how they can save it, and experiences people can have together in real life. There are probably new consumer products that could be built to address these issues, whether in the form of a chatbot, personal agent, or other formats.
Every AI assistant app launched pulls data from Gmail, Google Calendar, Docs, and all Google data to perform magical functions. Within Google, building such integrations has been very difficult. Gemini and Gemini Spark can connect these data sources, which is very powerful. The direction being pursued is moving away from different situations and toward a single command box where users tell the system what they want and it carries out tasks. This represents a broader user interface simplification happening across the entire industry.
People frequently ask Google to make their data work together, but there's hesitation in associating it with competitors' products. The vision centers on personal intelligence, where Gemini aims to be personal, proactive, strong, and work for users. This vision has been worked toward for almost the past year, with many things currently being tested for release.
The debate about principles and values doesn't receive enough attention in the AI sector. Whether at Google, a startup, or anywhere in between, products express and embody principles. There's a need for more discussion about what kind of things should be built and what kind of future is aspired to. This represents an area where more work could be done.
Model performance metrics are considered overrated. Most people in the world would never care about evaluating "Ello" and probably don't know how to pronounce it. The sector may be over-focusing on speeds, capabilities, and metrics, which are indicators of capability development, while overlooking the real reason why technology benefits someone and what product or feature will offer real value that helps people.
Every time a new model launches, there's discussion of artificial general intelligence. The metrics attempt to show aspects of superiority, but if users don't experience meaningful differences themselves, the improvements may not matter. The focus should be on whether something can be built that people want.
There's a lot of opinions about where AI is going next, but the key question is how the world should look and what vision for the future is desired. This doesn't mean all possible things need to be built, and perhaps not all things should be built.
The rate of "learning cancellation" is being researched - how quickly people can learn something and then abandon it based on product testing observations. Explosive stamina refers to how people take care of themselves and others around them, working very hard while sustaining it for a long time and finding ways to regulate work pace.
Collaboration and how people work with others is becoming increasingly important. The ability to blend worlds - being an amazing collaborator with agents and with humans - is valued. People who can build and expand trust are the cornerstone of great collaboration. The rate at which someone builds trust with others and creates that environment is a key indicator.
As costs of implementing products decrease, more products can be made with the same people. Team formation has shifted from needing five to seven people to two to three people, or maybe four. The approach is like two talented musicians collaborating with different people to play different pieces of music together, where the pieces are the products.
The Product Manager role is by nature more all-encompassing, making it able to adapt to other roles more quickly. PMs are more used to thinking about playing different roles like lawyer, business development expert, or salesman in meetings. However, the best employees can come from any job, including strategy, operations, business development, and marketing.
There's a danger if everyone becomes a "developer" - the loss of respect, appreciation, and expertise that comes from different disciplines. When forming teams, even when called "developer groups," the locations of specializations are still considered. The advice given to new product managers is that while coding and making prototypes is great, it should be done to become an exceptional product manager without losing the main specialization.
Seasons and cycles are used to manage intense work periods. For each small team, milestones or rhythms are created. The best lesson learned as a leader is to name and announce intense work states, making it clear when teams are not in that pattern. After Google's annual conference in June, lab teams are told to go innovate, rediscover what's new, build, and enjoy their time. This period, though it may seem unproductive with no releases, is where all new seeds for the next season come from.
The concept of work-life balance has evolved. Explosive endurance is valued because teams need to work together for many years without depleting energy completely. There's always something happening, whether it's Google I/O conference, new model releases, or other events. Planning has shifted to 6-month roadmaps with visions for several years, but the world changes so rapidly that things seen two days ago can become projects in 48 hours.
Planning typically uses about six months as the possible window, with individual points between pre-training model releases. Teams typically envision around 100 days, moving from idea to significant milestone achievement within 50 to 100 days. The culture and environment design removes obstacles to let people be creative.
Every company needs a group of people whose job is to be at the forefront, connected to the collective mindset of how people think and all versions of models. These are people with eight subscriptions at $200 a month for everything. The environment in which these people can build is equally important. Historically, large companies starting labs often see them become ineffective after 3 to 4 years, or the company cannot market things coming out of the lab.
There's an interesting balance between having teams at the forefront of innovation building cutting-edge things while connecting across the larger company so it's not perceived as just a weird R&D group always coming up with gimmicks. It takes certain people who can say that something sounding like a game now will be fundamental to the future in 5 years. This is where avant-garde ideas sometimes fail historically.
With lab growth and successful projects gaining users, there's risk of larger projects crowding out zero-to-one projects. Terms like "Zero-to-One" labs, "One-to-Ten" labs, and "Ten-to-One Hundred" labs denote different stages of the laboratory lifecycle. Each stage has different challenges and improvement priorities. In the zero-to-one phase, the focus is on user numbers.
In the "ten-to-one hundred" stage, conversion rates and acquisition costs become key metrics to evaluate. The focus shifts to driving people toward "magical moments" in the product, representing an expansion game in every sense of the word.
Laboratory teams have evolved from their previous goal of building large new business units. The current approach focuses on identifying what new technology will open up new horizons for the core business, rather than creating standalone ventures.
The first piece of advice is the need to create a space that allows unconventional things to grow. The team cannot simply be placed under the umbrella of an existing business unit. It must have some kind of independence, which is essential. In some companies, this means the team must report directly to the CEO or have sufficient distance from day-to-day operations.
The second point relates to the nature of people and adopting a "user first, then Google, then labs" mentality, where the lab's mission is the success of users and the success of Google. The goal is not simply for the laboratory to achieve success of its own kind, requiring humility and cooperation.
The critical question becomes: Is the goal of your labs to transfer innovations to existing products, or is your goal to create new product lines and categories? Labs started as a laboratory for transferring innovations, but this has expanded as artificial intelligence is changing and blurring many categories.
The third piece of advice involves using labs not just to develop products or explore technology, but to challenge how the company operates and the way things are done. Labs sometimes create "good problems" - problems that are beneficial to have, unlike bad problems that should be avoided.
An example includes being one of the first teams to recognize that functional roles are overlapping, necessitating a career ladder for entrepreneurs. This approach uses labs as a place to test new concepts before rolling them out across the entire organization.
The team is obsessed with finding people who can't stop building. A complete document about the team entitled "Laboratories in Brief" is sent to anyone joining the team via email in their first week. Key characteristics include:
- Building constantly
- Possessing intellectual curiosity
- Reading extensively
- Following relevant platforms and podcasts
- Being obsessed with solving people's problems rather than focusing on vanity or personal glory
- Adding energy when others are around them
There are approximately 17 or 18 such team members currently.
People in the "zero-to-one" stage are slightly different from those in the "one-to-ten" and "ten-to-one hundred" stages. When interviewing candidates, instant credibility comes from statements like "Hey, let me show you this thing I just built" or when candidates demonstrate how they approach unknown, unstructured, and vague matters where they don't know the way forward.
A common mistake is rushing to collect people or having a tendency toward merging and making things organized too quickly. Sometimes chaos must be lived with for a while to see what emerges. The willingness to make changes remains essential.
Another common mistake is hiring too many people too quickly, which happens in both large companies and startups. It can be tempting to focus on flashy metrics like team size doubling or tripling, but these are mostly worthless. An example involves reaching about 30 engineers on a project with an inspiring vision but failing to achieve market alignment.
The question of whether Google will have 10,000 products or only five in five years remains an ongoing discussion. One framework considers 10,000 Google products as 10,000 interfaces with their own brands, but this complexity may not be feasible.
Another consideration involves experiments that are more customized and made by prototypes, potentially meaning laboratory teams create agents that perform specific tasks through a common interface. Universal applications have been excellent in some use cases, but for many users they can be too complex, making it difficult to motivate value-adding behavior.
There is significant room for improvement in how feedback is gathered on products and experiences, whether at Google Labs, as a startup owner, or otherwise. This represents another area of focus on how to listen, gather, and collect ideas and inspiration from feedback loops. Artificial intelligence applications to feedback collection are creating interesting developments, with agents working to summarize ongoing activities.
The question of what makes a product manager great involves having a vision about certain types of people and the kinds of things that will please them or solve their problems. The challenge becomes how to exploit feedback loops using artificial intelligence and new types of products being built.
Another aspect involves exploring models of cooperation between humans and artificial intelligence that allow for more natural flow of insights and ideas, similar to productive human collaborations.
Several unique awards recognize different contributions:
- TPU Unit Harvester: Rewards people who efficiently recover and harvest TPU units, providing them with a pitchfork and cash prize
- Gold Bandage: Given to people who fix small, annoying flaws that accumulate in products and annoy users
- Large Language Model Whisperer: Awarded to those who push models to their limits in surprising ways, receiving a huge pair of ears as recognition
A thank-you event functions like a secret society, where invitations appear on calendars with four or five leaders present. There is no agenda, and participants take turns explaining why they value each person's presence on the team. The meeting always begins with the statement "This is not a bad meeting. This is a good meeting."
Listeners are encouraged to send feedback through various channels. The relationship between the laboratory and users should be reciprocal, with appreciation expressed for those who use the products being built.
Keep Lenny's Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
