The Scale Problem
At Meta Reality Labs, we shipped AR features to 2 billion+ users. At that scale, small mistakes become massive problems. A gesture that confuses 0.1% of users is still millions of people.
We learned to validate everything before production because the cost of mistakes was measured in millions—not just dollars, but user experiences broken at scale.
What AR Taught Us
AR interactions are physical. They require body movement, spatial awareness, gesture memory. You can't spec your way to understanding how a pinch gesture should feel, or how fast a virtual object should respond to head movement.
We built prototypes constantly—not as deliverables, but as thinking tools. A 2-day prototype could answer questions that 2 weeks of meetings couldn't resolve.
This wasn't a philosophical preference. It was survival. The complexity of AR meant that specs were always wrong in ways we couldn't predict.
The Pattern That Emerged
Over 7 years, a pattern hardened:
- → Build something small that demonstrates the core question
- → Put it in users' hands (or on their faces, with AR)
- → Watch behavior, not opinions
- → Pivot or proceed based on evidence
- → Repeat until conviction is earned
This pattern worked for gesture recognition systems. It worked for commerce features. It worked for never-before-built hardware like Orion glasses. The domain didn't matter—the principle did.
From Meta to Rationale
Startups face the same problem as Meta, but with higher stakes. Meta had resources to recover from mistakes. Startups don't. Limited runway means you can't afford to build the wrong thing.
That makes validated learning even more critical. The methodology that worked at billion-user scale works even better at 0-to-1 scale.
Validate early. Pivot cheap. Ship with conviction.