Zapier Turns Models Into Workflows
Wade Foster, co-founder & CEO of Zapier, on AI agent orchestration
This is the standard platform playbook, the foundation model company wins by making the base model broadly useful, while the real product margin shifts to specialists who wrap that model in workflow, data, and trust. OpenAI built GPTs and the GPT Store as a no code layer for custom chat experiences, but its broader developer push has centered on general tools like the API, Responses API, Agents SDK, and Agent Builder. That leaves room for companies like Zapier to turn raw model capability into concrete, repeatable business workflows for non technical users.
-
Zapier is not competing by training a better model. It is competing by taking many tools, Salesforce, Gong, web search, email, Slack, and turning them into a step by step system with approvals, access controls, and handoffs. In practice, that is what enterprises buy when they want AI to actually do work reliably.
-
The market keeps producing deeper wrappers around general models. Bardeen goes after browser based scraping and text driven automation. Harvey, Hebbia, and other vertical AI companies go after legal and research workflows. The pattern is the same, general models create the raw intelligence, specialists package it around a job to be done.
-
This also matches OpenAI’s own incentives. Building frontier models and mass market surfaces is capital intensive and broad by nature. OpenAI was already pursuing platform plus consumer app in 2023, and by 2025 to 2026 it expanded developer tooling for agent building rather than trying to own every niche workflow itself.
The next layer of value should keep moving away from the model alone and toward orchestration products that know a workflow cold. The winners will be the companies that can take a powerful general model and make it feel like a dependable coworker inside sales, support, finance, legal, and other concrete operating loops.