Tired: AI Chatbots Wired: Real-World AI

It’s time to rebrand the title “Product Manager.” With 𝕏 hiring Nikita to run product, it’s clear that having a strong product team matters. But, “product management” has drifted too far from its original purpose and now mostly resembles project or program management. You

Grok 4 benchmarks look super impressive. Really solid results so far in my limited tests. What is clear is that we are nowhere close to having enough compute (for both inference and training) based on what is coming. More data + more compute, still = better performance.

So cool to see the Tesla Diner finally open! Can’t wait to check it out, The team at Tesla has been giving a lot of thought to every element of this project for quite some time. Exiting to see yet another vision become reality ⚡️

Have been trying all the frontier code models for a side project I’ve been working on. Here is the quick verdict so far: • Claude 4 Sonnet: backend ace and relentless bug-hunter. Best for complex tasks. • Gemini 2.5 Pro: best on UI and good at building to design spec •

Big week for AI models: • OpenAI released open-weight models (likely been testing as Horizon Alpha/Beta on OpenRouter) and they look strong • Anthropic dropped Claude Opus 4.1 for better real-world coding and agentic use cases • xAI launched blazing-fast image + video

Options to get to dinner tonight: Walk: Free Bus: $3 per person Robotaxi: $3.82 UberX: $9.93 Lyft Standard: $10.85 Waymo: $21.54 ($11.54 w/ $10 off referral) Low cost per mile will be disruptive. People will own cars way less if autonomy makes it this cheap to get around.

Serious question: why doesn’t most of Europe have A/C? Cultural? Economic? Energy costs? Regulation?

Even the most conservative Robotaxi math starts to melt my brain 🤯 ~19T global passenger miles/year (EIA, OICA, ITF) 25% capture = 4.8T miles. At 1.55 occupancy → ~3.1T vehicle miles. At $0.35/mile (Tesla/ARK) → ~$1.1T revenue. Exclude ~90 low-GDP / poor-infra countries