Read this & let it sink in. 13 days into this cold spell & Zohran Mamdani was letting the homeless sleep under the Manhattan Bridge. For the last 30+ years the NYPD did Code Blue sweeps of the homeless off streets in freezing weather. Mamdani ended that policy & 19 people died.

If you liked this, you'll love @packyM and @evanbeard's new piece on robotics

🎏 Our @reddotblue takeaway: Gen-P is real. The TAM is human physical labor. But physics doesn’t care about hype The winners will be those who cross the uncanny valley by pairing ambition with economics, and turn robots from demos into systems that actually work Full essay

💰 This bifurcates capital: - Mega-funds back broad-loop bets where money is a weapon - Smaller funds win with narrow-loop systems that optimize BoM, TCO and reliability Durable value is likely to accrue closer to applications, data loops and workflows than pure physics

🚗 #AutonomousVehicles are the cautionary tale. Demos in 2009 but still constrained ~15 years later Manipulation is likely harder than driving: more degrees of freedom, messier environments, weaker structure Time + economics matter as much as intelligence

🐋 The hard truth: robotics lives in the long tail After ~80-90% coverage, progress slows dramatically because reliability depends on mastering rare, messy edge cases, not improving the average case This is why impressive demos so often fail in production

🛣️ 2 paths are forming: 1. Broad-loop generalizability: Push complexity upstream into giant physics models. Long feedback cycles. Massive capital. Zero-shot ambition 2. Narrow-loop specialization: Solve problems in context. Tight hardware+software. Short feedback loops. Clear

🦾 But a Gen-P stack is emerging: - Physics model learns how the world responds to actions - Application model maps that intuition onto specific HW - Distribution happens through deployed robots IRL Unlike LLMs, HW, embodiment and end-effectors matter enormously

🥽 Why it’s harder: There is no internet-scale dataset of physical interaction Robotics data must be generated: via teleop, simulation, real-world demos or direct actuation, and each has sharp tradeoffs in speed, realism and cost Data, not algorithms, is the core bottleneck

🧮 The mental model investors reach for is LLMs: - Foundation models scale w/ data + compute - Application layers turn raw power into usable systems - Distribution + feedback compounds advantage Gen-P borrows this, but the world of atoms is less forgiving than the world of

🤖 Robots today are impressive but brittle. They work when the world is tightly controlled and the task is narrow General-purpose robotics (Gen-P) asks a harder question: Can robots adapt across tasks & environments with minimal retraining, the way LLMs did for software?

#CES2026 is peak hype Cool demos don’t tell you how close we are to robots that generalize and work reliably in the real world Here's how to think about the true trajectory of #AI + robotics - short thread below👇🧵 Full essay here: 🔗 https://t.co/T1BOwZoAGw

My 2 year old sneezed in my face and then said “bless you dada”

BREAKING: FCC bans foreign-made drones over national security

We just published the definitive account of how Netflix ended up buying Warner Bros. There is a lot here that has never been reported. Will share some throughout the day, but please read/share: https://t.co/yev8mz8pTS

Here's yet another study showing that kids who get smartphones earlier have worse mental health as teens. So what's the right age for a first smartphone? I'd say its whatever age you want them to cut back on sleeping, reading, exercise, and socializing. I advise not doing that

Things that make me bearish when assessing robotics companies approaching the 2026 year: - jibo re-skins - teleop demos that dont label themselves clearly as teleop demos (including speed) -hd renders - software-only robotics companies with the pitch “hardware is solved” -

A reminder of what we're up against in the NYC mayoral election. This isn’t about Mamdani's policies/rhetoric. It’s about the ugliness he emboldens, and the violence his supporters feel empowered to commit/ignore

The work is mysterious and important