Codex is currently eating the entire industry's lunch and the rest of the labs are writing strongly-worded safety papers about it.
OpenAI has turned Codex into something that actually does real work: background agents, browser automation, PR reviews, memory, long-running tasks, and devboxes that don't need you holding their hand the whole time. Three million weekly users. Meanwhile everyone else is still trying to sell you a $200/month 'personal AI computer' that is just a really expensive search bar with commitment issues.
Anthropic is out here dropping Claude Design like it's going to own the entire creative pipeline, but let's be honest — it's fancy moodboard software until it can ship production code without crying. They're also locking down gigawatts of compute and adding pharma CEOs to the board like they're building a regulated enterprise fortress. Respect the pivot, but the models are still too polite to be truly dangerous.
The infrastructure war is the only one that matters right now. TSMC just printed another absurd quarter while everyone fights over chips like it's Black Friday at Best Buy. Cerebras is going public, OpenAI is doubling down on them, and the real bottleneck isn't intelligence anymore — it's power, cooling, and who can actually get their hands on enough silicon before the lights go out.
Robotics is finally starting to look less like expensive puppet shows. Google DeepMind's latest Gemini Robotics stack is doing actual embodied reasoning instead of the usual 'watch this robot fail to pick up a block for the 47th time.' Still not ready for your kitchen, but the gap between demo and deployment is shrinking faster than the cope on AI Twitter.
Hacker News has its priorities straight for once: NIST dropped integrated photonics chips that can do any-wavelength lasers. That's the kind of boring-but-world-changing work that actually matters, while half the timeline is arguing about whether Opus 4.7 is allowed to have opinions. NASA still making hard choices on Voyager 1 is a nice reminder that real engineering still exists.
On the biotech front, Lilly and Nvidia just dropped up to a billion dollars on an AI drug discovery lab with robotics and digital twins. Peptides are having a moment in the journals, but the real shift is AI moving from 'maybe this molecule works' to industrialized wet-lab + dry-lab loops. The miracle drug era is being replaced by actual factories.
AI Twitter is doing its usual thing: pretending to argue about models while actually panicking about who controls the workflow layer, who has the compute, and who can ship agents that don't require a 22-year-old prompt engineer on standby.
Bottom line: the demo era is dead. The winners won't be the ones with the smartest tokenizer. They'll be the ones who can actually close the loop between thinking and doing at scale, without needing a human to babysit every step. Everything else is just expensive performance art.