I’d probably start with one agent and keep the scope controlled. Multiple agents sound powerful, but if the user has to figure out which agent to talk to, the experience can get confusing pretty quickly.

I think the answer depends on how different the jobs are. If support, sales, and onboarding need completely different context and permissions, specialized agents make more sense. Otherwise, one agent could keep things much simpler.

For me, one agent feels better from the user’s perspective, while multiple agents might be better behind the scenes. The customer shouldn’t necessarily need to know how many agents are working underneath.

Honestly, I think we’ll see a hybrid approach become pretty common. Users will interact with what feels like one AI assistant, but behind the scenes it’ll delegate tasks to different specialized agents depending on what the user is trying to accomplish.
‎

What I’ll do is to look at it from the failure side. With one agent handling everything, a bad decision can potentially affect several parts of the product. Specialized agents give you clearer boundaries around what each one is allowed to do, which could make testing, permissions, and debugging a lot easier as the SaaS grows. The interesting part is deciding where that boundary is worth the extra complexity.

‎I can see specialized agents being especially useful when they have different permissions. A sales agent probably shouldn’t have the same access as an agent handling refunds or account changes. Separating those responsibilities could make the system safer too.
‎

I’d probably start with one main agent and keep the specialized parts behind the scenes. For the user, having one place to ask for help feels simpler, while different agents can handle support, sales, onboarding, or specific workflows when needed. That also seems easier to maintain than making users figure out which agent they should talk to.

I’d lean toward starting simple. Build one agent around the most valuable use case, learn what users actually ask for, then introduce specialized agents when the data shows there’s a real need. I wouldn’t design a whole multi-agent system just because it sounds impressive.
‎

‎There’s also a maintenance angle here. One huge agent handling everything sounds convenient until you’re debugging why it made a bad decision. Smaller agents with clear responsibilities could be easier to test and improve.
‎

‎I’d probably build one interface with multiple specialized agents underneath. The user gets one simple place to ask for help, while the system can route each request to the agent that actually understands that particular workflow.

‎

‎I think users care less about whether there’s one agent or ten agents and more about whether the experience feels seamless. If multiple agents can work together without making the user repeat themselves, I don’t really see a problem with specialization.

‎

What matters to me is whether those agents actually need to think and act differently. If support, sales, and onboarding are all working with the same kind of information, one agent might be enough. But once they need different rules, goals, or access, forcing everything into one assistant could get confusing fast. Its better keep things simple for the user, but split the work behind the scenes where it genuinely makes sense.

The biggest thing I’d worry about with one agent is context getting messy. A support conversation, a sales conversation, and an onboarding flow all need different information. At some point, separating those responsibilities could make the system much more reliable.
‎

I’d start with one capable agent, but I wouldn’t try to make it do everything on day one. Once the workflows become complex enough, I’d split specific tasks into specialized agents where it actually improves reliability.
‎

I’m facing one of the most critical architecture decisions right now. For me, the monolithic vs. multi-agent debate comes down to user friction vs. technical feasibility. If I’m building an AI SaaS today, I’m building multi-agent under the hood but presenting it as a single, unified agent to the user.