The agent loop got good enough to expose its own ceiling. The next discipline is designing agent systems as explicit graphs: boring nodes, typed edges, checkpointed state.
We recreated the WordPress pull requests behind CVE-2026-63030 and CVE-2026-60137 to test whether today's code reviewers and SAST tools would catch both vulnerabilities.
Nosdesk was perfect for about a year, because it only existed in my head. What happened when real people got hold of it, and an old Greek story about the same mistake.
I gave an AI agent access to a cable modem, a serial console, Ghidra, and a custom GDB stub, then asked it to reverse the firmware and hunt for bugs. It re-discovered known vulnerabilities, found new ones, and wrote working exploits. This post covers the setup, what it managed to do on its own, and where I still had to step in. The real problem is that the same approach now works on millions of devices nobody plans to fix.
Kevin sent me a link to this email from Linus Torvalds the other day, and I found that to be a great example of something that has been bothering me …
Can you prove a human decision-maker was actually there? Caneni is the founding methodology for AI Oversight Evidence.
Jack Conte, the founder & CEO of Patreon and half of the musical duo Pomplamoose, just gave a keynote about his take on AI, comparing decades-old responses to new technologies with the new technology of today.
Happy World Emoji Day 2026! Emojipedia's annual celebration of all things emoji is now in its twelfth year, and here's a quick rundown of what's been happening as part of this year's celebration! 🌎📅🥳
Get structured output from one API call. Run extraction, web research, and browser automation without managing LLM orchestration, browser infra, or pipelines.
Drop in a rough idea. StartupForge AI generates a complete startup blueprint: market sizing, pricing, MVP features, and a 90-day roadmap. No consultants. Just clarity.
Improve your piano sight-reading with free online note-reading drills, real song challenges, and instant feedback. Practice on a phone, computer, or MIDI keyboard.
Tasks, projects, CRM, calendar, focus, and a visual canvas. One connected system, on your machine. One-time purchase. No subscription.