Workflow automation and AI agent infrastructure
Zapier
Zapier connects business apps, data, and AI tools so teams can automate onboarding, lead routing, and IT helpdesks. Its Zaps combine triggers and actions across more than 9,000 apps, with AI steps for summarizing, classifying, and drafting. Tables stores data for automation, while agents can work across connected apps. Business users can build without code; developers can use Zapier MCP to expose app actions to AI assistants or Zapier SDK to call them programmatically, with Zapier handling authentication, token refresh, and rate limits.
Founded by Wade Foster, Mike Knoop, and Bryan Helmig, Zapier launched in 2012 and remains led by CEO Wade Foster. Its engineering approach centers on reusable integrations: the public zapier-platform repository contains the developer CLI, core runtime, integration schema, and example apps. Its published TypeScript connectors make individual app integrations available as both agent skills and MCP tools, giving agent builders readable code they can install and adapt.
In its newsroom available in 2026, the company reported more than 3.4 million businesses using Zapier and more than 25 million Zaps created. Zapier raised $1.3 million in 2012 and reached a $5 billion valuation in 2021 without further company funding.
3 talks
Newest firstTurning Fails into Features: Zapier’s Hard-Won Eval Lessons
Rafal Wilinski · Vitor Balocco
AI Engineer World's Fair 2025 · 16:15
How Zapier Builds AI Products and Features With the Help of Braintrust
AI Engineer World's Fair 2024 · 14:59
4 speakers at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here
- How Zapier Builds AI Products and Features With the Help of Braintrust
Start here to learn how logic-based and LLM-based graders, CI runs, and provider load testing fit into AI product development.
Ankur Goyal · Olmo MaldonadoAI Engineer World's Fair 2024
- Turning Fails into Features: Zapier’s Hard-Won Eval Lessons
Learn how Braintrust MCP and a reasoning model can help investigate regressions and compare differing model behavior.
Rafal Wilinski · Vitor BaloccoAI Engineer World's Fair 2025
- Your Support Team Should Ship Code
Learn why gathering documentation, logs, and bug context became a key bottleneck, and how LLM-assisted diagnosis and testing addressed it.
Lisa OrrAI Engineer Code 2025
Messages from the stage
Testing accuracy alongside latency
The joint discussion with Olmo Maldonado and Ankur Goyal covers synthetic evaluation data, automated and manual tests, and regression detection. It also examines the accuracy-versus-latency tradeoff when choosing models for a streaming copilot.
Evaluating the whole agent execution
Rafal Wilinski and Vitor Balocco connect execution traces and user-feedback signals to business metrics. Their evaluation approach accounts for intermediate tool calls, generated artifacts, and grading bias rather than judging only final responses.
Putting diagnosis inside existing workflows
Lisa Orr describes how a standalone Autocode playground interrupted support workflows. Adoption improved when diagnosis tools moved into Zapier and Jira, with completed merge requests routed back to support for validation.
Affiliations reflect each recorded session, not necessarily current employment. The AI Zap Builder and Zapier Copilot session is a joint Zapier–Braintrust discussion.


