Four Years at Magic Leap: The Long View
Reflecting on nearly four years of building mixed reality systems - what I learned, what changed, and what's next.
Four years ago I knew nothing about AR/VR. Now I lead perception for devices shipping to customers worldwide. Time to reflect.
The Arc
Year 1 (2016): Learning everything. Depth sensors, SLAM, embedded constraints. Building foundation.
Year 2 (2017): Building teams and systems. Synthetic data team founded. Eye tracking progressing. Spec freeze pressure.
Year 3 (2018): The crucible of shipping. V1 launch. Field feedback. Debugging under pressure.
Year 4 (2019): Maturation. V2 planning. Scaling synthetic data. Global team coordination.
Each year taught different lessons.
Technical Evolution
What I understood in 2016 vs 2019:
Sensors: Thought more pixels = better. Now understand: right pixels at right time, well-calibrated, matter more.
Algorithms: Thought clever algorithms solve everything. Now understand: data quality and system integration dominate.
ML: Thought neural networks were magic black boxes. Now understand: they're functions we can analyze, optimize, and understand.
Architecture: Thought good specs led to good systems. Now understand: specs are starting points; iteration is everything.
Leadership Evolution
Team size: 3 → 45 (including contractors)
Scope: Individual contributor → Director spanning multiple sub-teams
Communication: Mostly technical → 50% technical, 50% people/process
Decision making: Implemented others' decisions → Made decisions others implement
The role transformation surprised me. I came to build, ended up enabling others to build.
What I'd Do Differently
More real-data collection early: Synthetic data is powerful, but we under-invested in real data. The combination is stronger than either alone.
Earlier production calibration focus: We scrambled on calibration close to ship. Should have been a Year 2 priority.
Better spec-to-test traceability: Requirements existed but weren't systematically tested. Some V1 failures were specified features that weren't verified.
More user research: We built what we thought users wanted. More direct user involvement would have shaped priorities differently.
Industry Perspective
MR in 2016: "This is the future! Billions will wear these!" MR in 2019: "This is hard. Progress is slower than hoped. But still the future."
The hype cycle played out. We're in the trough of disillusionment, climbing toward productive deployment. The technology works. The use cases are still emerging.
What's Next
V2 is on track. The team is strong. The technology is maturing.
For me, I'm starting to wonder: after nearly four years here, is there more I can learn? Are there other challenges where my experience could have more impact?
No decisions yet. But the question is there.
Whatever happens, Magic Leap gave me a masterclass in building complex systems from scratch. That travels.
Grateful for every challenge, every failure, every late-night debug session. This is what building looks like.