I’d look for a niche where people are still relying on spreadsheets, emails and repetitive manual work then build an AI product that connects those pieces into one workflow. The best opportunity might not be the most exciting problem on paper but the one where removing that daily friction creates obvious value for the customer.

Specialization may become more important as the product gets more complex. A single agent might work well early on, but as support, sales, onboarding and operational tasks grow, separating those responsibilities could make each workflow more reliable and easier to improve. The challenge is doing that without making the user experience feel fragmented.

The natural language approach could make agent management much better for teams that are constantly tweaking their setup. Being able to ask for an update, review what changed and approve it without digging through several menus sounds practical. It would be enticing to see if it also makes troubleshooting and improving the agent faster over time, not just the initial configuration.

Another benefit is how AI agents can help teams standardize the way work gets done. Rather than relying on different people to remember every step or carry out the same situation differently, the agent can follow defined processes consistently. That can be useful in support and operations, where small inconsistencies can quickly turn into bigger problems. The challenge is making sure the process and knowledge behind the agent are actually kept up to date.

SaaS is being judged by how much task it can actually take off the user’s plate not just how many tasks it can automate. Once an AI agent can understand context, make choices and take action, providing details starts to feel less valuable. But that also raises the bar for reliability and judgment. If it’s going to act like a teammate, it needs to solve problems without creating more work to supervise it.

From what I’ve seen, getting the first 100 users is usually more about being willing to put yourself out there than finding some perfect growth hack. Talking directly to people, asking for referrals, sharing the product in the right communities and learning from early users can go a long way. I also think the unexpected channels are often the ones that end up working best you start with a plan, then your users show you where the real opportunity is.

The big test here is what actually makes the support team’s day easier. AI native tools sound great if they can resolve issues, keep context, route tickets and hand things off to humans without creating extra work. But Zendesk and Intercom are already pretty mature, so I’d be eager to hear from teams that have actually switched.

The real value of an omnichannel AI agent is having one source of truth rather than simply having AI everywhere. When knowledge, workflows, and customer context are managed centrally, it should reduce the headache of updating multiple systems and help keep responses consistent. But I also think each channel needs some flexibility, what works well in website chat may not work the same way over WhatsApp, email, or voice. One question is whether the operational savings actually hold up once the edge cases and human handoffs start piling up.

This is one of the biggest challenges with adding AI to SaaS products. Customization can make the product much more useful for specific customers but every extra prompt, workflow, integration, or piece of business logic creates something the team has to maintain later. I’d probably focus on building flexible systems rather than heavily customized ones and only add complexity when it clearly enhances the user experience or solves an important problem. Over time, simplifying the AI stack might actually create more value than constantly adding new layers.

Launching early makes more sense especially for a new SaaS. A waitlist can build anticipation but nothing compares to having real users try the product and provide feedback on what works and what needs improvement. You can build the community and generate excitement along the way.