The setup: two reports, opposite titles

In mid-2025, two reports landed close together and seemed to contradict each other. Anthropic published "How we built our multi-agent research system," justifying a multi-agent approach (covered in Chapter 2). Cognition published "Don't Build Multi-Agents," arguing for a single agent (covered in Chapter 3). The titles look like a direct clash, and the community treated it as a debate: should agents be single or multi?

McGuinness's answer is that this is a false dichotomy. Both positions are valid, because they are answers to differently shaped problems, and underneath both lies the same real concern: managing context.