Resend, an email-sending service focused on developers, keeps appearing in recommendations made by automated agents even when developers have a “home bias” toward a larger incumbent. Zeno Rocha, Resend’s co-founder and CEO, explained the practical reasons behind this pattern in a recent podcast interview.
Recommendation data
An AEO study cited during the conversation found that the Claude Code model picked Resend first in 64 of the 102 extractable email recommendations. SendGrid was first just seven times but appeared as an alternative in 55 suggestions. The hosts noted their later, broader AEO study produced more competitive results, with Postmark coming out on top for several models.
How Resend gets into developer workflows
Rocha says one of Resend’s earliest effective distribution mechanisms was an email template library. While the team was building the sending API, developers repeatedly complained about how hard it was to make emails look right. Resend responded with React Email, a template tool documented so templates could even be sent through SendGrid. As a result, by the time a developer considered switching providers, Resend-related code was sometimes already in the repository.
Agents and machine-readable presentation
Resend also made design choices that help automated agents evaluate the product. Rocha highlighted the pricing page’s slider: set a volume and the price is visible. Append ".md" to the URL and the page becomes a machine-readable table that an agent can fetch in a single request — a small change that simplifies model-based comparisons.
MCP server and growing toolset
At the time of recording, Resend’s MCP server had expanded from roughly 20 tools at launch to more than 80. Rocha’s intent is to allow people working in a chat client to perform tasks they would in the dashboard: sending an email (enough for a demo), inviting teammates, checking usage and managing billing for long-tail workflows.
Rocha described opening collapsed tool calls in his own sessions to see what the agent actually did. In one example the result was too large to fit in context; simply reducing the number of tools would not have made that response smaller. He suggested a cheap experiment for MCP designers: expand the result of a frequently used call and see how much of it the model actually requires.
What happens after “send”?
Choosing an SDK does not solve the business and operational work of delivering mail. For example, click tracking is often implemented by rewriting links through a shared tracking domain to count clicks. If one customer uses that domain for phishing and it gets blocklisted, links in other customers’ legitimate emails can be affected. For that reason Resend introduced custom tracking domains. Rocha also explained why they want to keep a free tier despite the risk that spammers can sign up — and why the company communicates a $3 million safety commitment to handle abuse.
Users and customers
At the time of the recording Rocha said Resend had 4.3 million users and 120,000 paying customers. He shared an anecdote about discovering Warner Bros. Discovery in the customer database on a Saturday by filtering for short .com domains.
Closing notes: product packaging and naming
Toward the end of the episode Rocha discussed the personal side of building a company: running a business while raising a family and the surprisingly important task of finding a name and category label that others actually use. The podcast also pointed listeners to practical resources mentioned in the interview, including React Email’s history, Resend’s anti-abuse handbook, and a chapter list for the episode.
Related materials and chapters
The episode’s chapters cover topics such as Rocha’s path from the Dracula editor to email, when agents choose an API, DNS hurdles, the React Email wedge, measuring adoption, owning the mail stack, MCP design, CLI and SDK craft, mail for agents, writing for machines, winning recommendations, enterprise and pricing, building with a family, and naming a category. The hosts additionally referenced Resend’s MCP server integration and anti-abuse documentation as further reading.
