It's a flywheel. Operating robots builds the judgment. Judgment scores the data. Consensus across hundreds of people turns that into quality at scale.

That's exactly the instinct you need to score data. Watch a demonstration and you can tell fast whether it's clean or a mess, because you've been on the other side of the controls.

Teleoperating a robot teaches you something you can't get from reading. You feel what smooth, competent, useful operation actually looks like, in your own hands.

Almost two weeks ago we opened Verify Quality, our data QA product, to the public. The people getting good at it fastest are our own community, and most of them got here by teleoperating our robots.

Today's fleet: red, white, and blue. Happy 4th from PrismaX 🇺🇸 Run a session this weekend: app.prismax.ai/live-control

The demo that goes viral and the demo that changes how someone feels about a robot in their home are usually not the same demo. The first is the one that's often rewarded. The second is the one that actually matters.

Standing in a room, what people lean in for is utility. Can it fold my shirts? Can it do my dishes? Can it organize the shelf I've been ignoring for a year?

Online it's the opposite. The clips that travel are the flashy ones. Backflips, dancing, sprinting. When people scroll, they want a quick hit of entertainment.

The robot demo that draws a crowd in person is almost never the one that goes viral online. At every robotics floor we've been on, from CES to ICRA, the longest line isn't the backflip. It's the robot folding shirts.

We shipped something at @PrismaXai this week and most of the work that went into it is invisible in the final product. What people see is a clean way to score robot data and earn for it. What they don't see is the months of deciding what "good" even means, the rules we rewrote more than once, the small calls that felt enormous at 2am. Launches are strange that way. The visible part is a fraction of the actual work. The rest is judgment, repetition, and a lot of saying no to things that would have been easier. Grateful to the team, and to everyone who showed up for The First 100. The fun part starts now.

The founders who do well at this intersection take the coordination tools from one world and the humility from the other. That combination is rarer than it should be.

What got in the way: the instinct that everything can be learned from information. Crypto you can study by reading. Robotics you only learn through proximity to the machines. You have to watch them fail in person.

Also transferred: patience for long timelines. Crypto people are used to building infrastructure years before it pays off. Hardware is the same kind of bet.

What transferred: incentive design. Crypto is the best laboratory we have for getting thousands of strangers to coordinate around a shared goal. Robotics needs exactly that to collect data at any real scale.

I spent years in crypto before I came to robotics. Some of it transferred better than I expected. Some of it actively got in the way.

Claude is good because of the people behind it. We're doing the same thing for robots, and for the first time you don't need a lab to be part of it.

The 100 people who are best at it become our first cohort of validators. We're calling them The First 100. It's actually the human layer that makes good models possible.

We opened up data quality scoring on @PrismaXai. You watch real robot demonstrations, score them against the bar that decides what's training-grade, and earn for the calls you get right.

The most underrated job in robotics is the simplest to describe. Look at a robot doing a task and know whether it's actually good. For a while that judgment lived inside a handful of labs. As of this week, anyone in our community can do it.

The teams that show up early to something like this are usually the ones who actually get why it matters. Glad these did. Thank you @theBBFund @ChainGPT_Pad @monad @FabricFND @peaq @PerleLabs @SentientAGI @StanfordSBA @virtuals_io @xmaquina @ZenO4AI