https://news.ycombinator.com/user?id=dubovskiyIMHacker News RSShttps://hnrss.org/hnrss v2.1.1Sat, 09 May 2026 04:17:17 +0000Two questions on the design: "lead-agent reads sub-agent output" makes the lead an LLM judge. Are there post-conditions checked, or is LLM-judge approval the final gate? Related: does the system reason over typed state or over raw output?

]]>Thu, 07 May 2026 15:28:54 +0000https://news.ycombinator.com/item?id=48050604dubovskiyIMhttps://news.ycombinator.com/item?id=48050604https://news.ycombinator.com/item?id=48050604This laws works only if there is human in the loop. When the consumer in an AI agent and it is autonomous - rules are breaks. Agent read output and decide what to do himself. I do not explain how this rules are breaks - it is obvious, i only want to say, that this rules should be structural. Not behavioral. Agent layer (or something else) should declare what is allowed and what is not.

]]>Tue, 05 May 2026 20:21:08 +0000https://news.ycombinator.com/item?id=48027962dubovskiyIMhttps://news.ycombinator.com/item?id=48027962https://news.ycombinator.com/item?id=48027962