Lenny's PodcastWhy every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)
CHAPTERS
- 0:00 – 5:34
Why every company must operate like a “labs team” (and what that really means)
Lenny proposes that rapid AI progress forces every company to behave like an experimentation lab: constantly testing what’s newly possible before competitors do. Josh agrees, adding that the challenge isn’t believing the premise—it’s building the environment and teams that can sustain frontier experimentation inside larger orgs.
- •AI capability changes are forcing faster experimentation cycles across industries
- •A “labs mindset” is natural for startups; harder to import into medium/large companies
- •The real work is designing the environment, incentives, and people systems that let frontier work thrive
- •Success depends on nurturing both ideas and the builders who generate them
- 5:34 – 8:00
Where great product ideas actually come from (and why you can’t schedule them)
Josh describes patterns he’s seen across many Google Labs experiments: great ideas rarely emerge from formal brainstorms or design sprints. They tend to come from highly curious people hacking constantly, often in unplanned moments, with leaders staying alert to recognize and channel the spark.
- •Great ideas rarely come from design sprints or calendarized ideation sessions
- •You “can’t microwave” good ideas—conditions matter more than process
- •Strong signals often come from small groups obsessively hacking on side projects
- •Leaders must create space for weird ideas to grow and be ready to catch them early
- 8:00 – 9:42
The “almost possible” framework: tracking frontier capability shifts
Josh explains a core Labs tactic: maintain a running list of experiences that are “almost possible,” then watch for phase-shift moments when they become viable. The team combines frontier tech exploration with user problems and monetization potential to decide what to pursue.
- •Keep a live list of what’s “almost possible” and monitor for capability breakthroughs
- •Look for phase shifts—when latency, modality, reliability, or cost crosses a threshold
- •Pair new capability with a real user problem and a plausible willingness-to-pay
- •Being at the frontier is an active practice: constant testing, reading, and observation
- 9:42 – 11:52
Future-betting as a discipline: predictions, entertainment, and behavior research beyond the Bay Area
Labs builds explicit points of view about the future—even while expecting most predictions to be wrong. Josh shares how they explore domains like entertainment, map technical prerequisites, and study behavioral trends by getting outside Silicon Valley to find real shifts early.
- •Labs maintains many explicit predictions to cultivate a future-oriented point of view
- •Entertainment and immersive experiences are a major frontier area being monitored
- •They decompose “future experiences” into technical requirements (latency, languages, modalities)
- •Behavioral trend research benefits from leaving the Bay Area bubble to observe broader reality
- 11:52 – 14:37
Signals worth betting on: surprise, obsession, and early examples (NotebookLM, Flow)
Josh describes how personal excitement isn’t enough—what matters is surprising capability paired with user pull. He recounts early “you have to hear/see this” moments from NotebookLM’s podcast feature and creative tools like Flow/Whisk as examples of frontier sparks that demanded action.
- •A strong early signal is genuine surprise and sustained obsession after seeing a demo
- •Tech magic must still map to behavior and durable user value
- •NotebookLM podcast generation created an immediate, memorable ‘must act’ reaction
- •Creative control and accessibility (Flow/Whisk) can be a breakthrough vector
- •Not every exciting demo becomes a good product—volume of attempts matters
- 14:37 – 17:57
Finding product-market fit in the lab stage: ‘eyes are the metric’ + love the problem, not the solution
Josh argues early PMF is more art than science: conviction and rapid prototype testing dominate before metrics are meaningful. He emphasizes in-person prototype reactions (“eyes lighting up”) and warns against falling in love with a solution before fully understanding the underlying problem.
- •Early PMF is often pre-metric; conviction and qualitative signals lead
- •Prototype testing should happen outside the building; visceral reaction is the key read
- •“Eyes lighting up” beats early retention dashboards for zero-to-one evaluation
- •Expect multiple pivots—fall in love with the problem, not the initial solution
- •When things hit, it feels like ‘swarming and holding on for dear life’
- 17:57 – 19:35
When to kill an idea: letting teams say the hard thing (and a real Gemini example)
Josh explains that teams often know a project is failing before leaders do; the leader’s job is to make it safe to acknowledge reality. He shares a recent example where a Gemini feature was stopped after weak early signals, praising the PM for protecting quality and focus.
- •The team typically recognizes failure before leadership does
- •Key signals: passion fades, repeated pivots don’t improve outcomes, user pull isn’t there
- •Leaders should design psychological safety for calling ‘not working’ early
- •Example: a Gemini feature was canceled after early user testing didn’t show ‘the eyes’
- •Healthy labs culture treats killing ideas as progress, not defeat
- 19:35 – 22:18
Why Google Labs and the Gemini app sit together: shared builders, AI obsession, and ‘users first’
Lenny probes the unusual structure: Josh oversees both frontier experiments and a massive consumer app. Josh explains the common thread is a zero-to-one heartbeat, builder profiles, and a consistent prioritization principle—users first, Google second, product third.
- •Gemini operates at massive scale but still needs a zero-to-one rhythm for rapid iteration
- •Labs and Gemini attract similar builder personalities and AI ‘nuts and maniacs’
- •Shared mantra: users first, Google second, product third (Labs/Gemini third)
- •Google Labs is all-in on AI, including hardware-adjacent bets like Google Beam
- 22:18 – 24:31
Next consumer AI breakthroughs: entertainment, messaging, and scarce human resources (time, money, memories)
Josh argues consumer is newly dynamic and competitive because AI changes what’s possible. He points to entertainment evolution, messaging/chat reinvention, and products that help with universally scarce resources—time, money, and meaningful real-world experiences.
- •Consumer isn’t ‘dead’—AI reopens the landscape for new category creation
- •Entertainment formats may shift beyond conventional feeds toward new interactive modes
- •Messaging/chat is being reimagined across the industry
- •Big opportunity areas tie to scarce resources: time, money, and shared memories/experiences
- •Form factors remain unsettled: chatbot, agent, or something not yet invented
- 24:31 – 27:19
Google’s personal assistant vision: fewer modes, more ‘just do it,’ and leveraging your Google data
Lenny challenges why third-party assistants can connect deeply into Google data while Google feels slower internally. Josh notes pieces exist today and frames the direction as removing toggles/modes in favor of a single prompt experience, built around “personal intelligence” that is personal, proactive, and powerful.
- •Trend: UX simplification—remove modes/toggles and move to one universal prompt surface
- •Gemini already offers some integrations, with more iteration underway
- •Many users prefer Google to use their existing data rather than connect it elsewhere
- •Vision framing: ‘personal intelligence’—a Gemini that’s personal, proactive, powerful
- •Josh signals near-term launches/experiments in this direction
- 27:19 – 31:17
Overhyped vs underhyped in AI: principles > benchmarks
Josh says principles and values are under-discussed: products encode beliefs about what future we’re building. He calls model benchmarks overhyped because most users don’t care—what matters is turning capability into products that deliver real value.
- •Underhyped: values/principles that guide what should be built (products as ‘encoded principles’)
- •Overhyped: benchmarks and rating obsession vs user value
- •Industry discourse often over-rotates to speeds/feeds instead of usefulness
- •Product opportunity is vast because ‘almost possible’ becomes possible weekly
- 31:17 – 35:46
Skills rising in value: unlearning rate, explosive endurance, and trust at scale
Josh outlines what he sees as mispriced traits for the AI era: the ability to rapidly discard outdated beliefs, sustain high intensity over long periods, and build trust quickly across people and AI agents. The conversation then digs into rhythms that prevent burnout while keeping momentum.
- •Unlearning rate: learn fast, then abandon assumptions when reality changes
- •Explosive endurance: sprint intensely, but sustainably over years
- •Trust-building becomes more important as teams form/reshuffle and work with agents
- •As execution costs drop, smaller teams ship more—chemistry and collaboration are differentiators
- •Leaders should name intensity modes and design seasonal rhythms (sprints + recovery)
- 35:46 – 46:26
Why roles aren’t collapsing into one ‘universal builder’ + planning in rolling windows
Josh argues ‘everyone is a builder’ is overhyped: specialties still matter like a major/minor, even as tools blur boundaries. He also describes how planning compresses to rolling six-month windows with 50–100 day milestones to match fast-moving model and product realities.
- •PMs may adapt fastest due to horizontal nature, but top builders come from any function
- •Specialization still matters; avoid losing expertise by flattening roles too far
- •Teams are getting smaller (often 2–4 people) as AI lowers execution costs
- •Planning has shifted to rolling ~6-month horizons and 50–100 day meaningful milestones
- •Long-term vision still exists, but detailed multi-year roadmaps are less realistic now
- 46:26 – 1:02:11
Labs lifecycle stages, setting up an internal lab, and culture rituals (awards, gratitude, ‘good trouble’)
Josh explains how Labs distinguishes stages (0→1, 1→10, 10→100) with different success metrics and risks. He offers concrete advice for building internal labs—independence, clarity on graduation vs new categories, and using the lab to challenge company processes—then closes with culture rituals like playful awards and surprise appreciation meetings.
- •Lifecycle framing: 0→1 (qualitative pull), 1→10, 10→100 (scaling/activation economics)
- •Tip: give labs real independence—can’t be nested under a business unit that smothers it
- •Define the lab’s winning game: graduate into core products vs invent new product lines
- •Use labs to create ‘good trouble’ by piloting new ways of working (e.g., job ladders)
- •Hiring signals: compulsive builders, curiosity, low ego, energy, comfort in ambiguity
- •Culture rituals: TPU Harvester, golden Band-Aid for paper cuts, LLM Whisperer, surprise gratitude meetings
