I watched GPT-6 play StarCraft.
GPT-6 controls the game through computer use. No MCP, no game-specific tools. Just a simple build for now, with a few prompts to prime GPT-6 and get it ready for the actual game.
The game is StarCraft: Brood War. Two players: GPT-6 as Zerg, the computer opponent as Terran.
Opus 5.5 is really good at creating explainer videos. wild!
I shipped this launch video for a startup working on agentic engineering and agent-native infrastructure in about 20 minutes, for roughly $40.
Opus 5.5 writes the film, and an agent renders it.
No video generation model was used here. No additional libraries. Just JavaScript, Playwright, and ffmpeg. Claude coded and drew every frame of this animation, then took a video of it.
A software factory is a system where triggers, such as a Linear issue being created, a schedule, a CI failure, or another session finishing, start work rather than a human. More work happens while you’re away, leaving human attention for decisions that still need it.
It doesn’t have to cover the entire software process. A codebase, or part of one, where no human reviews the factory's output is a dark factory.
That's the AI Coding Dictionary's definition: https://www.aicodingdictionary.com/?term=software-factory
Geoff Huntley takes it further: your product is the factory. The automation belongs inside the product, enabling it to build and improve itself.
| <!DOCTYPE html> | |
| <html lang="en"><head><meta charset="utf-8"> | |
| <meta name="viewport" content="width=device-width,initial-scale=1"> | |
| <title>GPT-6 Astra: A new generation of intelligence | OpenAI</title> | |
| <meta name="description" content="Introducing GPT-6 Astra, our most intelligent and aligned model yet."> | |
| <style> | |
| :root{--bg:#0b0c0e;--fg:#ececf1;--muted:#9aa0a6;--line:#22252a;--accent:#7cc4ff} | |
| *{box-sizing:border-box} | |
| body{margin:0;background:var(--bg);color:var(--fg);font:400 18px/1.7 -apple-system,"Segoe UI",Inter,Helvetica,Arial,sans-serif} | |
| header.hero{position:relative;min-height:60vh;display:grid;place-items:center;text-align:center;overflow:hidden;border-bottom:1px solid var(--line)} |
One word keeps coming up in AI discussions: taste.
Everyone talks about it, but almost no one defines what they actually mean.
If taste is supposed to be the moat in the agentic AI era, we should probably define it first.
Taste is:
“Taste” is the ability to consistently make high-quality qualitative judgments where no objective metric exists. It’s the creation of something that feels right intuitively, with no real justifiable way to measure that. But when you do it, people feel it.
A person with “good taste” is someone who can do this repeatedly, consistently. The funny thing about taste is that it’s hard to create, but its result is very easy to copy. Once someone makes a tasteful decision, others can imitate it almost immediately.
This is usually an argument against the existence of taste: “look how easy I can copy your work!” And yet, you couldn’t create the work without first having someone to copy it from. One has taste, the other doesn’t.
There have always been people with consistently good taste. But taste is coming up more regularly than ever before. It is becoming a critical differentiator.
Learning where AI is perhaps going, agentic AI workflows, AI disruption, and SWE transformation.
Over the past year, I've been thinking about AI agents, loop engineering, and verification systems.
A few areas I'm particularly excited about:
its a myth
a fable
Claude Fable 5 by Anthropic, releasing tomorrow: https://news.ycombinator.com/item?id=48450521