The hinge: both sides are really talking about context

Here is the move that resolves the debate. McGuinness points out that Cognition's whole position centers on how to best manage context. And that is exactly Anthropic's position too. Both companies make clear that a good agent system lives or dies on managing context: the prompts, the tool calls, the data, and the communication between agents. Doing that well is context engineering.

So the two famous articles are not really arguing about architecture. They are both saying that context engineering is the thing that determines reliability. Cognition's solution is to avoid dividing context at all, for the simplest and most reliable operation. Anthropic's solution is to divide context across subagents but then invest heavily in making the division clean: highly specific instructions to subagents and precise tool definitions, exactly so that miscommunication does not creep in. Anthropic faced the very context problems Cognition warns about, and engineered its way through them. Same problem, two valid strategies.