I’ve been curious about this for a while now, and I finally had a chance to test it with a real workflow.
The pitch is appealing: describe what you want your workflow to do in plain English, and the AI generates a ready-to-run automation. In theory, that would save massive amounts of time compared to manually building workflows step by step.
But I’m skeptical about how much of that actually survives contact with real requirements. When I’ve used code generation tools before, the output is maybe 60-70% useful. You get the structure right, but then you spend hours tweaking details, fixing edge cases, and adding error handling.
I’m wondering if workflow generation works the same way, or if it actually produces something that’s close enough to production-ready that you’re not rebuilding most of it.
The reason I’m asking is because we’re trying to compare Make and Zapier for a migration project, and if we could generate equivalent workflows on both platforms from the same plain-language description, we’d actually be able to do a fair cost comparison. Right now, we’re building each one from scratch, which means the cost is heavily influenced by how efficiently we design it on each platform.
Has anyone actually used AI workflow generation for something beyond a proof of concept? How much customization did you end up doing after the AI generated the initial version?