One of the most non-trivial challenges with VLA-driven robotics is few-shot improvement at deployment Real world variance explodes once the robot enters an unseen environment To handle edge cases, robots need adaption. These 3 ways guarantee it’s success: 1. Error taxonomy &

The secret to investing is to invest in unstoppable people building inevitable things.

I didn’t realize how badly I needed this

It’s remarkable seeing VLAs have a 21% increase in performance when fine-tuned with human egocentric data. The world has no idea what’s coming as new emergent capabilities are discovered with scale More compute & data = larger models = emergent capabilities = generalization

I’m so bullish on the future of robotics that it hurts

These policy-merging breakthroughs will continue to compound in 2026 and accelerate useful general purpose robotics. It’s so exciting

ALOHA or DROID? I’m becoming increasingly bullish on open-source VLA models for real world manipulator deployments Seeing strong policies fine-tuned on both ALOHA and DROID training data Which embodiment dataset/framework actually works best?

A $30B market is forming and no one is talking about it. This week I came across the RoboArena leaderboard for DROID-based policies built by Karl Pertsch, and it’s one of the clearest signals I’ve seen of how fast open-source, real-world robot learning is starting to compound.

all you have is now

This is easily the clearest and most accurate view of the robotics landscape today. Narrow commercial deployments seem like the highest-likelihood path to general manipulation

Chase greatness

The only path to general manipulation will be through deployment-first data

Physical AI intelligence is at an inflection point. Over the next 3-5 years, we will begin seeing robots create immense commercial value staring with workflows with low environmental and task variability. The future is sooner than we think

Incredibly impressive! Huge congrats to Jon Miller Schwartz kaan doğrusöz and Evan Wineland for pushing the limits with pi06 Bullish.