Two Years at Meta: From Zero to Platform
VPS evolved from concept to platform in two years - reflecting on the journey, the team, and what's next.
Two years at Meta, and the shape of them is easy to see from here: VPS shipped on Quest Pro, the team grew to 80, and the end of December is as good a time as any to look back and forward.
The Arc
- July 2020: Joined as IC, one of 5 on a nascent VPS team
- December 2020: Leading VPS engineering, team of 15
- June 2021: MVP launched to developers
- December 2021: Team of 50, expanded coverage
- October 2022: Quest Pro integration shipped
- December 2022: Team of 80, VPS as platform
Zero to platform in 29 months.
Key Accomplishments
On the technical side we got visual localization working at scale, built neural map compression, kept the architecture privacy-preserving, and stood up the crowdsourced mapping pipeline. On the product side VPS shipped in Quest Pro, covers 1000+ mapped locations, has 100+ developer partners, and daily active users keep growing. And the team went from 5 to 80, with three strong engineering managers and a culture of shipping and quality.
What I Learned
The Meta lesson first: the resources are real, but so is the complexity, and getting alignment across orgs is the hardest part of the job. Scale taught me that problems at scale are qualitatively different - what works at 10 fails at 10,000. Leadership changed underneath me too; my job shifted from solving problems to making sure problems get solved, and letting go is essential to that. And building a platform is a different discipline from building a product, because developer experience matters as much as end-user experience does.
What is still hard?
Indoor localization is still not production-ready. It keeps turning out harder than expected. I gave it a 2022 beta in September of last year and a 2022 launch in December, and both dates are behind us. The economics have not cracked either - mapping remains too expensive and needs a 10x improvement. Of the four priorities I set last December - indoor, the 10x on mapping cost, self-service tools, the next hardware - only the hardware integration shipped, and the other three reappear below with the year changed. The team number, 80, was the forecast that landed exactly. Developers want more coverage before committing. And the competition is real now: Google and Apple are building similar capabilities.
Looking to 2023
The technical bets are indoor localization launch, next-gen learned features, and real-time map updates. Real-time updates is the retraining problem I flagged in January as the thing that decides whether neural maps deploy; a year later it is still deciding. Product focus goes to Quest 3 integration, exploration of the Ray-Ban AI glasses, and developer self-service tools. The team plan is consolidation rather than growth, with more senior technical depth and stronger cross-functional capabilities.
Personal State
Energy is still high; the problems remain interesting. My own learning rate has slowed, though, and I notice myself looking for new challenges. Balance is better than year one - the pace finally feels sustainable. VPS is on solid footing, which frees me to ask an uncomfortable question: is there more I can contribute here, or is it time to apply these skills somewhere else? I am not leaving yet. But I am thinking about what comes next.