Embodied Collective

8/ Narrow sites compound At @RLWRLD_ai, each deployment feeds the model, so the same task starts near 90% next time, not 70% Full write-up: embodiedcollective.org/insig… Next event: luma.com/embodiedcollective

7/ Every extra moving part is a liability A 21-DoF hand worked in the lab and broke on day one on site. @ChestnutRobotic limits degrees of freedom on purpose. Sensors break first

6/ Design the human handoff @cyberneticphysx robots run a safety check before every action and hand off to a remote operator when unsure @Evan_Tao_T: even at 99.99% autonomy, teleop won't disappear

4/ Start with customers who need it most and decide fastest Urgency beats brand. Mid-sized manufacturers watching margins every month move faster than famous logos 5/ Track usage hours, not revenue. Revenue lags

3/ Define success before the robot arrives @Robotcom's first pilot with Sodexo started with a clear definition of success. It became 40+ campuses and a 7-year contract (@Juda_Chris)

2/ Customers buy outcomes The question isn't when they pay. It's who's responsible when the robot stops. Formic bundles maintenance, spare parts, financing and insurance, and only gets paid if it works. 98% of subscriptions renew

1/ Reliability multiplies A lab demo at 85% isn't a product. Uptime = model x hardware x repair speed, and the total has to land near 99.99%. @goformic runs at 99.96% across ~150 factories (@samanfarid)

Thank you for coming to The Deployment Moat. 300+ registered, 2 panels on putting robots to work. Thanks @samanfarid @Juda_Chris @RLWRLD_ai @cyberneticphysx @Evan_Tao_T, @Scobleizer for opening, @Scav and @MissionRobotBay for hosting 8 lessons from the night:

Demos raise rounds. Robots still running in March build companies. Is deployment the real moat in Physical AI? Tomorrow in SF: @Scobleizer @samanfarid @goformic @Evan_Tao_T @ChestnutRobotic @cyberneticphysx @Juda_Chris @Robotcom @RLWRLD_ai Model or deployment? One word below