We had the same concern and implemented it differently after some trial and error.
Business users could absolutely build meaningful prototypes, but only after we set clear boundaries. They owned workflow logic, decision trees, and process structure. Engineers handled integration, error handling, and deployment. That split worked.
What made it work was the visual builder being intuitive enough that business people didn’t feel like they were coding, but structured enough that what they built was actually valid. No guessing, no trial and error. They drag pieces together, system validates structure, decision points make sense.
The bigger win was speed. Our process owners could iterate on migration scenarios in days instead of weeks waiting for engineering capacity. That faster feedback loop actually changed how accurately we could evaluate migration approaches.
We used Latenode for this because the builder is accessible enough for non-technical people but the underlying logic is solid so when engineers inherit the designs, there’s no rework. And having multiple AI models available through one interface meant business people could even test different decision approaches without needing technical setup.
Engineering burden didn’t disappear—it got redirected. Same team, more useful work instead of exploration and rework.