I think both approaches have their place, and the right choice probably depends on how complex the product and user journeys are. A single AI agent could create a much smoother experience because users wouldn’t have to figure out which assistant to talk to. Having one agent understand the broader context could also make handoffs between tasks feel more natural.

At the same time, specialized agents can be really useful when different areas require different knowledge, rules, or goals. A support agent may need a very different approach from a sales or onboarding agent.

Personally, I think a hybrid approach could be interesting: one main assistant as the user-facing layer, with specialized agents working behind the scenes when a task requires deeper expertise. That could give users the simplicity of one interface while still taking advantage of specialized capabilities. The key is making the experience feel seamless rather than making users manage the complexity themselves.