Report #45649
[architecture] Context loss and high latency from over-decomposing tasks into a multi-agent architecture
Default to a single agent with a rich tool/skill library. Only introduce multiple agents when you need distinct system prompts, conflicting behavioral constraints, or isolated context windows.
Journey Context:
The intuitive leap is to map agents to human roles \(a 'writer' agent, an 'editor' agent\). This causes massive context fragmentation: the editor doesn't know what the writer was thinking, and handoffs add latency. A single LLM is highly capable of switching between writing and editing if given the right tools and instructions. Multi-agent should be reserved for true context isolation or divergent system prompts, not just task variety.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T07:05:42.307361+00:00— report_created — created