Report #45807
[architecture] Spawning multiple agents for distinct tasks that share the same context window, leading to context explosion
Use a single agent with a robust skill/tool library when tasks are highly coupled; use multi-agent only when tasks have distinct, isolated state and context boundaries.
Journey Context:
The hype around multi-agent leads to over-engineering. Every agent hop loses context and adds latency. A single agent with tools keeps the 'world model' intact and avoids the token cost of re-establishing context. Multi-agent is only justified by context window limits or strict isolation needs. Tradeoff: single agent can become overwhelmed by tool count, but avoids the fragility of inter-agent communication.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T07:21:42.528486+00:00— report_created — created