THEY GAVE CLAUDE CODE A SUBCONSCIOUS

They just gave Claude code a subconscious thinking process through observer subagents #claudecode #claude #aiagents #ai #vibecoding They just gave Claude code a subconscious. This is because experimental feature called observer subagents. Observer subagents are one half of a two subagent hole, where the other one is an implementer, a researcher, or actually doing the work. Actually doing the work. Agent gets deployed at the exact the implementer or worker sub agent and actually monitors its transcript as it gets longer as the implementer actions using different tools. That observer agent actually gets reads calls and its thinking for any errors that the implementer is taking. Notices the worker making a mistake, it can actually issue report directly to its live session to correct it. This is almost like adversarial code review, where you have one agent create and another agent context window review that PR. Except the observer agent is reviewing every tool block as it comes in and happens. Of course, deploying an observer subagent with every worker subagent will consume more tokens. But the theory kind of is in the long run, maybe it will reduce overall token consumption small error corrections when they happen, error corrections that can propagate through long running autonomous projects that kind of correct all at the end. Just one of dozens of super cool features that Harness has. If you want video breakdowns on all of them, those videos are hosted in my Skool community. In my bio. In my bio. Would love to see you there.