Video wird geladen...
Video konnte nicht geladen werden
Anthropic engineer: "90% of our engineers were already running self-improving loops Now everyone is building agentic graphs" "Prompting is basically over" In just 10 minutes, she builds her complete Claude Code system live from an empty terminal Agents → Loops → Graphs → Self-Improving Systems Prompting was the old... show more
531,957 Aufrufe • vor 3 Tagen •via X (Twitter)
8 Kommentare

@grok what are graphs in this context

@LunarResearcher one thing that caught me off guard was over-tuning loops. Kept iterating without letting them stabilize. Costly lesson.

watched it. empty terminal to a full system in 10 min is the real flex graphs aren't magic tho - they just force the part most loops skip: verification, durable state, and a recover path when a node dies

Agree. Loops + graphs are the real unit of work now, durable state, tool routing, verification, recovery. That’s what turns a smart model into a system you can actually run.

While forever loop to spend tokens on Claude. Great idea

I think this is just inadequate literacy on almost all parts. The strength is in feedback looping not just generic feedback. Fixating and identifying deltas and other constructs gives enough detail to do basic wavetable synthesis ( subpar to actual fm synth ) but either way this kind of iteration is clearly the most underhanded snake oil I’ve ever seen. You can’t polish a turd into a golden nugget. Furthermore, the answer is almost always patience versus action since 1) objectivity is supreme tokenization and 2) models only get better

Prompting is over, so the same workflow just got renamed and resold as a $1,

Is that claude 4 video from last year still relavent to the harness you are talking aabout?
