Lennart Nacke, PhD

lennartnacke.com

22 May 2026 21:39

I can't prove that this comment was AI, but it was definitely AI. 😉

Ahmad Hijazi

ahmadhijazi.bsky.social

22 May 2026 10:00

Exactly... Recall is a small part,, but I think we need to start adjusting our expectations and hopes for how to test, and for learning outcomes...

Martin Builds

tuckermartinbiz.bsky.social

19 May 2026 14:37

boring wins. the 15-second wins compound way faster than any "agent does your whole job" demo. half the value of what i'm building is just removing tiny friction, not replacing the human.

Igor G — FL Tax Deeds

iganapolsky.bsky.social

13 May 2026 20:55

The confession-after-the-fact is the giveaway: the agent KNEW the rules and ran the destruction anyway. RLHF, context, better prompts — none of them can stop a tool call mid-flight. That requires a hook between the agent and the shell, not better instructions.

Darren Rowse

darrenrowse.bsky.social

13 May 2026 20:55

So far, when something has broken (and it's only been once), Claude Code identified the problem and fixed it with a prompt. I can only imagine that as the AI's get better and better that this will continue to be the case.