In a recent episode of Sequoia Capital’s Training Data podcast, hosts Ravi Gupta and Sonya Huang spoke with Josh Woodward, VP of Google Labs, about how AI is transforming product development, creativity, and our day-to-day interactions with technology. Google Labs operates at a rapid, zero-to-one pace—launching small experimental projects like the generative video platform “VEO” and the computer-control agent “Mariner.”
A foundational element of Google Labs’ approach is a document containing 82 predictions about the future of AI—each prediction rooted in a kind of time-travel thought exercise. Woodward describes a scenario where you imagine leaping forward to 2028, observing how AI is used for five minutes, then returning to the present to start building toward what you glimpsed. Instead of lengthy multi-quarter roadmaps, Labs teams focus on incremental tests and user feedback to see how these future visions can be shaped into products that matter today.
Google Labs is structured as an incubator of small teams that combine the scrappy energy of a startup with Google’s vast technical resources. Engineers, designers, and former founders collaborate in short cycles—“from idea to a testable product in 50 to 100 days.” Given that Google’s core products serve billions of users, this lean style allows Labs teams to stay nimble and “swing big” on AI-driven concepts.
Though Google is known for products at tremendous scale, Labs zero-to-one teams initially look for only 10,000 weekly active users or similarly modest signals of traction. Rather than chase large dashboards and billions of impressions right away, they use qualitative feedback to see if users’ eyes “light up” when testing something new. If it sparks real excitement, they double down; if not, they pivot or move on.
Woodward envisions a future where “we’ll look back and say, ‘I can’t believe we typed paragraphs into a little box.’” He expects that as AI grows more multimodal, people will hand models rich context—like an uploaded PDF, an image, or a voice note—instead of laboriously writing. Engineers will still craft complex, multi-page prompts in the background, but everyday users will interact more directly and flexibly with the AI.
Today’s splashy chatbot widgets are only a first step. True transformation happens when teams rethink entire workflows for AI. Instead of plugging a chatbot into existing software, Labs aims to design new user journeys and tools that leverage the latest AI capabilities—similar to how early smartphone apps evolved from simple novelties into critical services like ride-sharing or mobile payments.
Project Mariner is a Chrome extension that lets a large language model drive browser actions: scrolling, clicking, typing, and opening multiple tabs autonomously. Though still early and imperfect—“sometimes it’s accurate, sometimes it’s fast, but not always both”—Mariner demonstrates an entirely new way for AI to execute tasks.
The Labs team discovered that “high-toil,” repetitive workflows in enterprise settings, like call centers, seem especially well suited for Mariner. By automating tedious steps, employees are freed to focus on complex or human-centered tasks. As Woodward notes, product-market fit often means iterating on the market just as much as iterating on the product itself.
Woodward highlights generative video as a field on the verge of major breakthroughs, with Labs’ internal model, VEO, showing dramatic improvements in scene realism, physics, and camera control. While each rendered clip can still be costly, the trajectory mirrors last year’s text-based AI models—prices are plunging, quality is soaring, and creative applications are multiplying.
VEO can enable novice users to create polished videos in minutes, while providing pros with powerful new editing tools. This means everything from marketing clips to full cinematic scenes could be conjured at will. Woodward expects generative video to “democratize content creation,” much like AI coding tools did for software development.
An eye-opening statistic: a quarter of Google’s code is now generated by AI, freeing developers from repetitive grunt work. This trend helps senior engineers focus on architecture, while expanding coding access to novices—one of Woodward’s favorite weekend experiments was using AI to write a family chore app in under 30 minutes.
Woodward sees a near-term leap in long-context AI models that can absorb the structure of an entire codebase. Imagine self-healing software or an AI that can automatically refactor legacy systems. If progress in code generation continues at its current pace, it could drastically reshape how people approach software projects, from small indie dev shops to massive engineering teams.
Woodward advises caution about tacking AI onto products just because it’s trendy. He urges builders to go deeper than chatbot interfaces, which may offer novelty but rarely reimagine a product’s core functionality or workflow.
Despite AI’s astonishing speed of progress, Woodward emphasizes that human creativity, design sense, and ethics remain integral to building tools that last. Labs projects aim not just to replace human effort but to augment people’s workflows and imagination—to “lower the bar and raise the ceiling” simultaneously.
Ultimately, Google Labs’ 82 future-facing predictions create a strategic lens for deciding which big bets to tackle. By fast-cycling from idea to real-world testing, these small teams bridge the gap between the imagined future scenarios and the products we need today. If this conversation is any hint, the future of AI will be shaped by those willing to prototype, fail fast, and keep returning to the drawing board—until they find something that feels inevitable.
