The Cognition team's blog post "Don't Build Multi-Agents" and Anthropic's blog post "How we built our multi-agent research system" present seemingly opposing views on building multi-agent systems. However, upon closer inspection, they share common insights on the importance of context engineering, which is crucial for making agentic systems work reliably. The two teams also recognize that multi-agent systems designed primarily for "reading" tasks tend to be more manageable than those focused on "writing" tasks, as read actions are inherently more parallelizable than write actions. Both blogs highlight the need for durable execution and error handling, agent debugging and observability, and effective evaluation of agents. Ultimately, both teams agree that there is no one-size-fits-all solution and that the choice of multi-agent system depends on the specific problem being solved.