If I were launching an AI SaaS from scratch today, I wouldn’t build another wrapper that just generates text or images. I would build what I think of as agentic middleware — a very specific, outcome-driven vertical agent.

What I find most interesting is how AI agents are changing what SaaS is actually selling. Instead of paying for a set of tools and then having to figure out how to use them, you start paying for an outcome and let the software handle more of the process itself. That can make products way more useful, but it also raises the bar for accuracy… Read more

What would make Backstage really useful for me is if it also helps my team remember why we made certain changes in the first place. Once a few of us start updating instructions, workflows, or knowledge, it gets hard to track what changed and what problem it was supposed to solve. If Backstage can keep that history clear and show who changed… Read more

Another way I think about the first 100 users is that the product itself can drive a lot of the growth. If a user has to invite a teammate, share a report, or send a link to get work done, then one user naturally brings in another without me having to do extra marketing. I might not have planned for that loop at the start, but once I notice… Read more

From what I’m seeing, SaaS teams are moving past that early hype phase where everyone tried to customize everything with AI. In the beginning, most teams including mine built heavy, custom wrappers around the models just to get the exact behavior we wanted. Now we’re realizing that creates a lot of technical debt. To balance a tailored… Read more

In my experience, running one AI agent across every support channel makes maintenance simpler overall, but it makes prompt engineering and testing a lot harder at the detail level. Platforms like Voiceflow and Chatbase push the “train once, launch everywhere” idea, which does help me avoid keeping separate knowledge bases in sync, but once I… Read more

For me, I’ve learned it’s better to ship a simple MVP fast than to sit on a waitlist for months. You only get real feedback when people can actually use the product, click through it, and pay for something that works. I prefer to get v1 out quickly, put it in front of users, and let their actual behavior shape what I build next.

I think the real shift is less about replacing live chat and more about using AI for the right interactions. I’ve noticed that AI agents can be great for quick questions, product recommendations, and lead qualification. But when a visitor has a complex or high-value concern, having a human available still makes a difference. The strongest…

Read more

I think the real question isn’t whether PocketBase can handle production, but whether it fits the complexity of the SaaS you’re building.

I’ve found tools like PocketBase really appealing for getting an MVP out quickly. The challenge comes later when requirements around scaling, integrations, background jobs, or more complex data workflows…

Read more

One thing I’ve noticed is that some AI features create more friction than value. I recently used an AI SaaS tool that kept suggesting things while I worked. The suggestions were impressive, but rarely relevant, so I spent more time dismissing them than using them. The best AI features should feel almost invisible—helping when needed without… Read more