Reid Robinson

Reid Robinson is a product manager whose AI work at Zapier connected language models to searches and actions in business applications. As Zapier’s lead AI product manager at OpenAI DevDay in 2023, he explained AI Actions for GPTs and the company’s new Assistants API support. His public guidance paired enthusiasm for automation with limits: constrain the tools a model can use, check accuracy, and add human review before critical uses.[1]

In his November 2023 Latent Space interview, Robinson described two jobs performed by Zapier’s team-built AI Actions: translating a model’s natural-language instruction into an application API call, then refining the returned data into information the model could use. A request to check someone’s calendar for Monday became a Google Calendar call; a large, confusing Gmail response needed processing rather than being passed back as its entire payload.[1]

Robinson favored GPTs’ ability to specify both which apps to use and how to use them. He contrasted this with Zapier’s earlier ChatGPT plugin, which exposed every action a user had enabled. Asked about the following week’s agenda, it sometimes searched Gmail because calendar invitations arrived there. Constraining the available tools and instructions offered a way to narrow that ambiguity.[1]

He distinguished human-triggered, one-off actions from workflows that could run in the background. His illustrative reporting workflow exported blog-traffic CSV data every Friday, sent it to an assistant instructed to identify top-performing articles and analyze relevant details, and posted the resulting report or charts to Slack. Teammates could then continue the conversation with the assistant in the thread.[1]

Robinson also recalled an internal expense-assistant experiment combining company policies with employee budget data. That context could let the assistant explain that a requested webcam exceeded the recommended budget while the employee still had personal budget available. The example showed how retrieved policy documents and live account information could inform the same response.[1]

At SAAS NORTH 2023, Robinson argued for starting with existing business problems rather than trying to predict every advance in AI. As he told BetaKit, “I try to tie that back to the problems we have.” One example was drafting lead emails from templates matching a user’s usual responses, with the user reviewing and finalizing the drafts in Gmail.[8]

His DevDay advice was to “double check accuracy” and include human review before critical use. He warned that freeform input could introduce prompt-injection risks and recommended internal use cases as a starting point for learning the technology’s constraints.[1]

Robinson authored Zapier’s announcements of MCP support through OpenAI’s Responses API on May 21, 2025, and Anthropic’s Messages API on May 22. His Anthropic explanation described creating a Zapier MCP server, selecting the actions Claude should use, and passing the server URL into a Messages API call. This gave the model access to tools such as sending Slack messages or updating CRM records.[6][5]

In a retrospective updated in January 2026, Zapier CEO Wade Foster credited Robinson with focusing full-time on AI, both building capabilities and helping teammates learn to build. Foster separately credited Mike Knoop and Bryan Helmig with earlier LLM experiments. He also said the early OpenAI plugin and AI Actions “didn't take off,” while describing those experiments as useful groundwork for moving quickly when MCP gained momentum in early 2025.[7]

Before this AI product work, Robinson managed strategic partnerships at Hootsuite and Zapier and co-founded cocoNFT. BetaKit’s account dates his participation in a small Zapier team experimenting with AI integrations to 2022, supplying the earlier context for his 2023 product explanations and subsequent MCP announcements.[3][8][1][6][5]