One agent, too many jobs
It's tempting to hand one agent a huge, capable model and let it do everything: research a topic, write the draft, fact-check itself, and polish the tone. In practice this tends to underperform, not because the model lacks knowledge, but because all of those responsibilities compete for space in the same context and the same set of instructions. A prompt that says 'research thoroughly, then write persuasively, then verify every claim' is asking one agent to hold three different mindsets at once, and the result is often a blend where none of the three gets done as well as it would on its own.
Splitting the task across multiple agents, each with a narrow role and a focused set of instructions, tends to produce more reliable results. A research agent only has to be good at finding and summarizing information. A writing agent only has to turn that summary into clear prose. A fact-checking agent only has to compare claims against sources. This is the same reason a team of specialists usually outperforms one generalist on a complex project: each member's attention is fully dedicated to one kind of judgment instead of being split three ways.