When AI Agents Do the Wiring
Zapier built its empire on a simple promise: connect your apps without writing a single line of code. For years, that promise was enough. Small businesses, marketing teams, and solo operators ran entire workflows through Zapier’s “if this, then that” logic, paying monthly fees to automate tasks that would otherwise eat hours. The platform now connects thousands of apps and serves millions of users. That installed base is exactly why OpenAI’s latest push into agentic automation should make Zapier’s leadership uncomfortable.
OpenAI’s Operator product – its AI agent capable of browsing the web, filling forms, and executing multi-step tasks autonomously – does something Zapier was never designed to do. It doesn’t require pre-built connectors. It doesn’t need a defined trigger and action pair set up in advance. It figures out how to complete a goal by navigating the actual interface of whatever tool is in front of it, the same way a human would.
That is a different category of automation entirely.
The Connector Model Is Starting to Show Its Age
Zapier’s architecture is fundamentally a library of pre-approved integrations. Each “Zap” depends on both connected apps playing nicely within Zapier’s defined API relationship. When an app updates its backend or restricts API access, Zaps break. When a business wants to automate something that doesn’t have a dedicated Zapier integration, they either wait for one to be built or hire a developer to work around the gap. This brittleness has always been the platform’s quiet vulnerability, but it was tolerable when no better alternative existed for non-technical users.
Operator-style agents sidestep that entire dependency chain. Because they interact with software through its user interface rather than its API, they don’t need a formal handshake between platforms. An AI agent can log into a CRM, pull data from a spreadsheet, compose and send an email, and log the outcome – all without any of those tools needing to know the agent is even there. For businesses running workflows across poorly integrated legacy software, this isn’t a minor upgrade. It removes a category of friction that Zapier simply cannot address with its current model.
The practical gap shows up most clearly in enterprise contexts. Large organizations often run workflows across tools that predate modern APIs entirely. Zapier has made genuine efforts to expand upmarket, but its product logic is still rooted in the clean, consumer-friendly world of SaaS integrations. OpenAI is building agents that don’t care whether your software is from 2024 or 2004.
Zapier Isn’t Sitting Still, But the Race Has Changed Shape
Zapier has been adding AI features at a real pace. Its AI-powered Zap builder, natural language workflow creation, and integrations with various large language models show a company that understands where the threat is coming from. The platform is trying to layer intelligence on top of its existing connector infrastructure rather than rebuild from the ground up. That approach buys time, and it keeps the existing user base loyal because their current workflows don’t break.
The problem is that layering AI onto a connector model doesn’t change what the connector model fundamentally is. Zapier can make it easier to build a Zap using natural language, but the Zap still depends on a pre-existing integration being available. OpenAI’s agents, by contrast, are designed to handle ambiguity and novel situations by reasoning through them in real time. A Zapier workflow is a pipeline. An OpenAI agent is closer to a junior employee who figures things out on the fly. For users whose needs fit neatly inside Zapier’s library, that distinction may not matter much yet. For users who have been bumping up against Zapier’s ceiling, it matters a great deal. The pattern of established productivity tools losing ground when a newcomer redefines the task itself is becoming familiar territory in tech.
Zapier also faces this challenge from multiple directions at once. Microsoft’s Copilot integrations are working their way into enterprise workflows. Google’s Gemini is embedded directly into Workspace. Make (formerly Integromat) has been eating into Zapier’s more technical user segment for years. OpenAI’s Operator is one more front in a war Zapier is fighting on several sides simultaneously.
The Real Question Is About Switching Costs
Zapier’s strongest defense is not its product – it’s inertia. Businesses have hundreds or thousands of active Zaps running mission-critical workflows. Rebuilding those in any other system requires time, testing, and risk. OpenAI’s agents, however capable they become, still need users to trust them with real business processes, and trust takes time to earn when the stakes are a broken sales pipeline or a missed invoice. But switching costs erode faster than most platforms expect, especially when the new alternative promises to handle not just existing workflows but ones the old tool could never touch in the first place. The moment an Operator agent successfully replaces three separate Zaps and a manual step that Zapier couldn’t automate, the calculation for that user changes permanently.
