bookmarks
5 bookmarks by @kieranklaassen

I have tested GPT-6 Astra this week. It is a show horse, not a workhorse. Super fun to use and showy, does crazy good stuff, and design-wise it's the most interesting model I've run. But I don't know if I'd use it for a real day to day engineering/product work. It's not as trustw…↗

Most agent architectures add a reviewer subagent to catch mistakes. You pay the cost twice: once to do the work, again to check it. Claude Skills flip this. They load domain expertise exactly when the agent needs it—during execution, not after. Real-time guidance, zero review ov…↗