Claude Code picks Resend first in 64 of 102 email answers

AI coding tools now pick the vendors a developer uses. Latent Space tested this on email providers. Its host swyx published the piece on 10 November 2026, under the title Designing for Agents. The guest is Zeno Rocha, chief executive of Resend. The test: ask Claude Code to pick an email API, 102 times. Latent Space reports that Claude Code picked Resend first in 64 of those 102 answers. SendGrid was named first in 7. SendGrid was named as an alternative in 55.

Resend was founded in 2023. It sells one thing: an API that sends email. Zeno Rocha says Resend has 4.3 million users. He says 120,000 of them pay. Those are his own figures for his company.

What Resend changed in its documentation

The podcast walks through specific edits Resend made so an AI coding tool can read it. The pricing page used an interactive slider for humans to pick a volume. Zeno Rocha says Resend added a .md version of the same URL. The .md page is a plain table of prices. An agent can read it in one request. It does not have to work the slider.

The second change is the company's Model Context Protocol server. Resend released it with about 20 tools. It now has more than 80. The podcast names the kinds of tool exposed: send an email, invite a team member, check usage, read billing. The dashboard and the MCP server now cover the same ground.

A third edit is older. Resend wrote and released its React Email template library before it finished the core sending API. A developer could build a template with React Email and send it through SendGrid. Resend code sat in developer repositories before the API choice was made.

Why the number matters

A developer who asks Claude Code for an email API gets a short list. Latent Space's test of 102 answers is the first public figure of its kind for this category. The score is lopsided. Resend is picked first more than nine times as often as SendGrid. For a buyer that spends on paid search, the money is not the only path to the shortlist anymore.

Zeno Rocha also describes a cost that comes with writing for agents. The company reads the tool-call results its MCP server returns. Long results fill the model's context. A longer result can push earlier context out. Resend trims the output.

What it means for a team selling to developers

The item is an interview with the vendor, so the engagement numbers and the MCP tool count are Resend's own. The 64 of 102 figure is Latent Space's test. A different test on a different day with a different model would score differently. The useful fact is the practice, not the exact count. Three edits put Resend on the list: a markdown price page, a full MCP server, and a template library a developer can run against a competitor. A vendor that waits for a buyer to open a sales page gets none of those visits.

A product leader selling an API to developers can act this week. Add an .md version of every page an agent would need. Release an MCP server that mirrors the dashboard. Trim the length of each tool-call result against a real prompt run. Those are the steps Zeno Rocha names. The test will change. The practice will decide who is on the short list.

Source

• Designing for Agents — Zeno Rocha, Resend, Latent Space.