Global Team Coordination: US-Israel Collaboration
Lessons from managing perception development across continents - what actually works between Florida and Israel.
Half of our perception team sits in Florida and the other half in Israel, 7 time zones apart, and nearly every process we run exists because of that gap. None of it happened by accident. Making this work requires intentional structure, and this post is a tour of the structure we ended up with.
How much overlap do we actually get?
Israel runs on GMT+2 or +3 depending on the season, Florida on GMT-4 or -5, and in practice the gap is 9-10 hours. When Israel ends their day at 6pm, Florida is just starting at 9am. When Florida wraps at 6pm, Israel is asleep at 3am. The shared window comes out to roughly 2-3 hours on a good day, and everything else about how we operate follows from that number.
Communication Patterns
Those overlap hours are reserved for things that genuinely need everyone awake at the same time: the daily 15-minute standup, technical deep dives, decision meetings, live debugging sessions. We guard the window fiercely and keep routine meetings out of it.
Everything else defaults to async. Written design documents, recorded demos, issue tracker discussions, pull request reviews. If it can be written, write it.
Documentation Culture
Global teams can't rely on hallway conversations, so documentation becomes infrastructure. Every significant change gets a written design doc: problem statement, proposed solution, alternatives considered, open questions. When we decide something, we record why - the context at the time, the options on the table, the rationale for the choice, and what would change our mind. And every meeting produces a written summary with the decisions made, action items with owners and dates, and open questions pushed back into async discussion. It sounds bureaucratic until you watch it save a day of cross-timezone back-and-forth.
Building Trust Remotely
Trust is harder to build without casual interactions, so we manufacture the conditions for it. Video is the default, cameras on, because seeing faces does something text can't. We run a weekly "coffee chat" video call with no agenda at all. We fly people in both directions quarterly, which costs real money and is worth every dollar. Appreciation that would happen casually in an office has to be deliberate here, so we make it deliberate. And when a message reads curt, we assume timezone fatigue rather than hostility, because nine times out of ten that's what it is.
Ownership Clarity
Ambiguous ownership kills distributed teams, and drawing the boundary between the two sites was still an open item when I wrote about hiring in 2017, so ours is now written down:
| Israel Owns | US Owns | Shared |
|---|---|---|
| Synthetic data | SLAM | Integration |
| Eye tracking | Mapping | API design |
| Sensor eval | Hand tracking | Architecture |
| Calibration | ML infra | Release |
Where ownership is shared, decision authority is explicit. Nobody should have to guess who gets the final call.
What Doesn't Work
Expecting real-time response, for one. If you message someone and they're asleep, you wait. Synchronous-first defaults fail too - scheduling meetings across 9 time zones for every discussion burns everyone's evenings and mornings for no gain. Hidden context is the quiet killer: decisions made in local meetings without documentation, which the other side discovers weeks later. So are separate backlogs, where one team's priorities are invisible to the other.
Our Rhythm
- Daily async standup: written update by end of local day
- Weekly technical sync: all senior engineers, video, overlap hours
- Bi-weekly all-hands: full team, recorded for async viewing
- Monthly planning: align priorities, allocate resources
- Quarterly visits: build relationships in person
None of this replaces being in the same room, and I've stopped pretending it does. But with discipline, a team split across an ocean can execute at high velocity anyway.