Report #51715
[architecture] Orchestrator agent micromanaging worker agents by dictating step-by-step instructions, negating the workers' autonomous reasoning
Give worker agents goal-oriented directives \(what to achieve\) rather than task-oriented directives \(how to do it\), allowing them to utilize their own planning and tool-use capabilities.
Journey Context:
When building hierarchical multi-agent systems, developers often make the orchestrator act like a traditional deterministic controller, breaking down a task into minute steps and passing them to workers. This wastes the worker LLM's ability to plan and self-correct, turning it into a mere text generator. The orchestrator should define the 'what' \(the desired outcome and constraints\) and let the worker determine the 'how.' The tradeoff is less deterministic global behavior, but massively increased resilience, as workers can adapt to local failures without calling back to the orchestrator.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T17:17:57.940170+00:00— report_created — created