Zapier's OAuth-Like AI Onboarding
Mike Knoop, co-founder of Zapier, on Zapier's LLM-powered future
Zapier is trying to turn automation setup from building a workflow into granting scoped access. The strategic implication is that Zapier sees its real moat less in the editor screen itself, and more in the hidden translation layer that already knows how to map plain language into thousands of app actions, field IDs, and safe defaults. That makes connecting an app feel closer to install and approve, then let the system fill in the form behind the scenes.
-
In the old Zap editor, users had to pick from long argument lists, custom fields, and dropdowns that often exposed machine style IDs instead of human labels. The NLA API collapses that into a plain text instruction, then Zapier interprets the request, signs the API call, and returns a trimmed result that is easier for humans and models to use.
-
The OAuth comparison is also about trust and control. Zapier wanted setup to look like connect account, approve access, and choose which actions are allowed, not hand over an open ended agent. In Mike Knoop's example, a user might allow message drafting but lock down the Slack channel choice, so the model can help without making the riskiest decision.
-
This is different from unified API companies like Merge or embedded iPaaS vendors like Prismatic. Those products reduce integration work for developers, but they still assume someone has to model the fields and build the experience. Zapier is pushing one step higher, using AI to absorb the last mile setup work so non technical users can authorize tools and start from intent instead of configuration.
This points toward a future where Zapier sells orchestration that is increasingly invisible. The near term move is making setup feel effortless, but the longer term move is using the same translation, access control, and workflow logic during execution, so automation starts to feel less like filling out software and more like assigning work to a reliable digital operator.