Claude Code Was Getting Dumber. Stopped It from Auto-Spawning Sub-Agents. Performance Came Back Instantly.

Published: April 15, 2026

A few weeks back Opus 4.6 went from god-level to painful. Rework loops, same mistakes repeated, context getting lost. Then I noticed what was happening. It was auto-spawning sub-agents for everything. Every new task - straight to a sub-agent - which was doing sloppy work. And when the sub-agent reported back 'Done' it just passed on the message...no validation, no checking, just messenger.

I mean as soon as I gave it a task...it just dumped on a sub-agent(s) and sat back doing god knows what.

Claude confirmed that's as per its system instructions too - it is supposed to auto spawn agents.

Maybe the main agent wasn't instructing properly. Maybe the sub-agent was a Haiku. I didn't debug further. Just tried 2 simple things:

1. CLAUDE.md global rule: zero agents without my express authorization

2. Every conversation starts with the same instruction - no agent delegation unless I approve or expressly ask for it

Performance came back instantly.

I still use sub-agents - for repetitive downloads, bulk front-end tasks, simpler UI changes. But I set up the process with the Opus 4.6 first, validate it, then have it delegate to sub-agent. Main agent has the context. Sub-agent is labor.

And the last week or two...it seems to have become even more intelligent...and human like...