Zapier's Human-in-the-Loop Approach
Mike Knoop, co-founder of Zapier, on Zapier's LLM-powered future
Human approval is the bridge between AI as a clever guesser and AI as trusted automation. Zapier is letting models fill in the easy parts of a workflow, like drafting text or mapping plain language to an app action, while keeping riskier choices, like the exact Slack channel, email send vs draft, or which app actions are even allowed, under explicit user control. That design protects trust while still making setup much faster.
-
The key product shift is from building a zap field by field, to granting a model a narrow set of powers. Users connect accounts through an OAuth style flow, expose only specific actions, and can hard set certain parameters so the model cannot improvise on them.
-
This also explains why Zapier fits AI better than pure chat products. Zapier already sits between thousands of apps, manages auth, and knows how to turn messy APIs into cleaner human readable actions. The model handles interpretation, while Zapier handles permissions, execution, and output shaping.
-
Competitors show the tradeoff. Make has leaned into deeper endpoint coverage and lower cost for more hands on builders, while Workato has pushed upmarket into enterprise workflows. Zapier's AI angle is to make automation easier to start, then add approval steps and deterministic workflow scaffolding where reliability matters.
This points toward a hybrid future where automation runs mostly in the background, but humans stay inserted at the moments that carry cost, compliance, or reputational risk. As models improve, more steps will become safe to automate outright, and Zapier's advantage will come from deciding exactly where judgment stays with the user and where software can act alone.