cd ~/
MetaReflectionScaleVPSLeadership

Six Months at Meta: Reflections on Scale

Half a year in - what I've learned about operating at Meta scale and building VPS from inception.

Evyatar Bluzer
3 min read

Six months in at Meta. A few notes on what operating at this scale actually looks like.

What Surprised Me

Process, mostly. I expected big-company process to slow things down, and often it's the opposite: well-designed process prevents chaos at scale. The depth of expertise surprised me too - whatever problem you have, someone at Meta has solved something similar, and finding that person is the real challenge. Years of infrastructure investment means features that would take months elsewhere take weeks here. And the ambition level is something else. The roadmaps assume we'll solve problems that don't have known solutions. Bold, occasionally unrealistic.

What I'm Learning

Navigation, for one: finding the right people, teams, and decision points requires active networking, because org charts don't tell the whole story. Influence without authority - I can't decree decisions here, I have to build coalitions, make the case, and earn buy-in. Planning horizons are long; decisions made now affect products launching in 3+ years, so I'm learning to think further ahead than I'm used to. And cross-functional coordination is real work. Design, research, policy, legal, and communications all have legitimate input, and technical correctness by itself doesn't carry a decision.

VPS Progress

The team started at 5 and is now at 15. We've hired well and the foundation feels strong. The VPS system design has been reviewed and approved, and implementation has started. Basic localization works in demo environments - nowhere near production. The day-one version of the mission was any device localizing in any mapped space at centimeter level. Six months in, the honest version is demo environments. We're aligned with the Quest team, the smart glasses team, and the mapping infrastructure team. The data pipeline is the bottleneck, and privacy review is rigorous, rightfully so. I wrote in May that data is strategy and the fastest data team wins. Knowing it did not stop the data pipeline from becoming the bottleneck here too.

Calls and how they agedThree horizontal rows on a month axis from May to January, each running from a dot at the call date to a dot in late December; the bottom row is highlighted.JunJulAugSepOctNovDecJan22 Dec 2020May: data is strategy; the fastest data team winsJuly: any device, any mapped space, centimeter levelNovember: prototype synthetic data pipeline in Q4 2020held; the data pipeline is the bottleneck anywaypartial: basic localization in demo environmentsslipped to 2021
Three calls from May, July and November 2020 and where each stood on 22 December: the data warning held and happened anyway, the mission is partial, and the Q4 synthetic pipeline slipped to 2021.

Cultural Observations

Move fast still feels true despite the company size; there's a bias toward action. Internal communication is remarkably transparent, with most docs accessible to everyone. Direct feedback is the cultural norm - people tell you what they think. And people genuinely believe in connecting the world. The mission talk is sincere.

What do I miss from Magic Leap?

Startup urgency. At Magic Leap every decision felt existential; Meta has more room for error. End-to-end ownership - there I owned hardware to software, here I own one component in a larger system. And scrappiness. Magic Leap made do with less. Meta has resources, but sometimes that means solving problems with headcount rather than cleverness.

What I Bring

Systems thinking, meaning I care how components interact rather than optimizing each piece in isolation. Hardware empathy: knowing what's actually possible on device, as opposed to in the cloud. Shipping experience, having launched real products rather than prototypes. And synthetic data expertise, which is still novel at Meta scale.

Looking to 2021

The goals: VPS MVP running on Quest, the synthetic data platform operational, the team scaled to 30+, and first external partnerships for mapping. One correction: last month I put a prototype synthetic data pipeline in Q4. It is on the 2021 list instead, which is the polite way of saying Q4 slipped. It's ambitious. But that's why I came here.

Comments