Zapier bets on orchestration as APIs fade
Mike Knoop, co-founder of Zapier, on Zapier's LLM-powered future
This points to a shift in where the durable value sits, away from special AI friendly wrappers around software, and toward the orchestration layer that can reliably move context, permissions, and actions across many tools. Zapier used plugins and natural language actions as a bridge for early models, but the broader direction is that models increasingly call tools directly, or operate software through existing interfaces, while Zapier remains useful as the system that connects thousands of apps and turns messy multi step work into a controlled workflow.
-
The practical reason these API adaptations are short lived is that they mainly help immature models navigate rigid schemas. Zapier found early models could already fill parameters with limited extra work, and its later AI strategy shifts from plugin formatting toward orchestration, governance, and long tail connectivity across 8,000 apps.
-
The closest comparison is universal APIs and embedded integration layers. Those products speed up common read and write use cases, but they usually cover only a slice of real workflow complexity. As customers need custom fields, logic, approvals, and edge cases, the value moves back to a fuller workflow engine.
-
The market has also moved toward models using software more directly. OpenAI now centers tool use around function calling in its newer API stack, while Anthropic offers computer use that lets Claude click, type, and read screens like a person. That makes one off natural language endpoints less defensible than broad execution infrastructure.
Going forward, the winners are likely to be the platforms that make AI actions dependable inside real business systems. As models get better at calling functions and using interfaces on their own, the bottleneck becomes security, routing, approvals, monitoring, and support for the long tail of business software. That is exactly where automation platforms can compound.