Key Takeaways
- Josh Woodward looks for high unlearning rates: the speed at which a builder abandons a polished belief when user feedback contradicts it.
- Long-term delivery in AI requires what Woodward calls Roger Federer's "explosive endurance": sprinting hard during major model milestones while pacing teams across multi-year horizons.
- Early product-market fit reveals itself in live human eye reactions during user testing rather than aggregate numbers on analytics dashboards.
- As autonomous agents handle more code generation, rapid trust-building and human collaboration become rarer, higher-value team differentiators.
The Mispriced Skill: Unlearning Rate
Most hiring rubrics in tech test for raw acquisition speed. Hiring managers ask how quickly an engineer can learn a new framework, prompt a model, or digest documentation. Woodward, who leads Google Labs, AI Studio, and the Gemini app, watches a different dial entirely.
In generative tech, the underlying capabilities shift every quarter. Workflows built in January become obsolete by April. When builders spend three weeks perfecting a workflow, they develop sunk-cost attachment to their code. High performers accept when new model releases render their architecture pointless. They drop their prior convictions instantly when live testing disproves them. Woodward watches for eye reactions across the table during testing sessions; if the user's eyes stay flat, the thesis is wrong, and the team must abandon it without mourning the wasted hours.
Roger Federer and Explosive Endurance
Startup culture worships non-stop urgency. Founders brag about seven-day workweeks and midnight releases. Woodward argues that AI development runs on a multi-year cycle, making continuous redlining fatal to execution.
To balance model deadlines with team longevity, Woodward borrows a concept from tennis champion Roger Federer: explosive endurance.
“There was this phrase in there about Federer had explosive endurance,” Woodward says. “It's just this beautiful turn of phrase and with the alliteration everything but it's like there there's something around like how are people taking care of themselves and the people around them where they can go really hard and that's the like explosive part but they can do it over a while.”
Model drops and major benchmark releases create sudden, intense sprints where teams must move fast. But between those bursts, squads must recharge. “We want to be doing this as a team for many many years,” Woodward points out. “And so it's like you can't just like empty the tank in 9 months and just like kill yourself, right?” If your team burns out after one release, you lose the match.
Scaling Trust in the Agent Age
With autonomous agents generating boilerplate code and summarizing research, some founders predict engineering teams will dissolve into solo operators. Woodward sees the opposite dynamic taking hold inside Google Labs.
“I still feel like collaboration and like how you work with people matters even more now,” Woodward notes. “And so maybe one thing I'd say there is like people that can build trust, that can scale trust, that that underpins any kind of great collaboration.”
When execution velocity increases, interpersonal friction becomes the primary bottleneck. Software agents can draft PRs and automate tests, but agents cannot resolve creative disputes or align cross-functional leaders around ambiguous roadmaps. The ability to establish trust quickly across disciplines separates teams that ship durable products from teams trapped in organizational gridlock.
What to Do With This
Tomorrow morning, audit your current sprint for sunk-cost traps. Put your latest feature prototype in front of two target users, ignore their polite post-demo comments, and watch their direct eye reactions while they click. If their eyes look detached, kill the branch by midday and reassign your engineers to what is almost possible on newer models.