My experience building systems of record for businesses is that they often require multi-level, hierarchical, relational data. Architecting the database demands diagnostic effort, engaging users at each role level to understand their workflows before a suitable schema can be designed. Getting a complete picture can take weeks or even months of conversation. No builder would sign a scope of work based on a handful of prompts from a single user. Yet that is precisely the constraint under which AI currently operates.
As a database architect, I find AI to be a genuinely powerful assistant. I own the overall architecture and use AI for discrete, well-scoped tasks that it can execute far faster than I can (e.g., add a table named XXX with the following fields…). I use it for custom code in workflow steps which gives me tremendous flexibility and for interface blocks, especially now that they support CRUD actions. The productivity gains are real and significant.
That said, AI has also shifted customer expectations, in large part due to uneven marketing messaging (industry wide, not Softr specific). The most challenging scenario I encounter is something like: “I already vibe-coded 95% of my app — I just need help with the finishing touches.” What I often find underneath is a database architecture that doesn’t match the actual use case, built without a full understanding of it. Untangling this is painful, partly because it feels like regression to the customer, especially when they’ve been told that AI makes building so easy that anyone can do it.
Current AI co-builder functionality is genuinely well-suited to internal tooling for small teams. The natural next frontier, though, would be expanding that capability to larger, more complex use cases by collecting input from multiple stakeholders and reconciling contradictory requirements into a coherent design (essentially, a structured diagnostic). That would represent a meaningful step change in value.
I would love to hear others’ experiences and thoughts on opportunities and limitations.