Skip to Content

Why AI agents will soon help choose your software

An AI assistant that suggests a vendor picks what it can read easily. Not what you require.
November 18, 2026

More and more often, a developer asks an AI assistant to pick a service and wire it up straight away. The assistant suggests one vendor, the code works, and nobody tries the second option anymore. That is how an AI helps decide which software your business uses, based on what it can read easily, not on your requirements.

What happened

On the Latent Space podcast of 10 November, Zeno Rocha, co-founder and CEO of the email service Resend, explains how his company targets AI assistants as a new "customer".

What the assistant picks

  • In a recent study Latent Space cites, Claude Code named Resend as first choice in 64 of 102 cases for an email service. SendGrid came out on top seven times, but showed up 55 times as an alternative.
  • A later, broader study by Latent Space itself was less one-sided: there, Postmark came out on top for most models.
  • The question the podcast asks: how often does anyone still look at option two when the first patch simply works?

How a vendor responds to that

  • Resend's pricing page has a slider for humans. Add .md to the address and an agent gets a table it can read in one go.
  • Resend's MCP server, the connection that lets an AI assistant operate the account, grew from about 20 to more than 80 tools. The goal: to do everything in a chat that you can do in the dashboard.
  • At the time of recording, Rocha says, Resend had 4.3 million users and 120,000 paying customers. Those are his own figures.

Our view: choosing software is a process decision. An AI may put the options forward, but the requirements and the choice belong to a human.

What this means for your organisation

Even if you do not write code yourself, this affects you. Your IT partner or your own developer probably already uses an AI assistant, and it helps decide which services end up in your systems. It works the other way round too: if you sell a service yourself, an agent will soon read your website on behalf of a customer.

What you do now

  • For any new integration, ask which alternatives were considered and why this one was chosen. Set your own requirements (cost, data in Europe, support) in advance.
  • Ask your AI assistant which vendor it recommends for a task you know well, and see whether you agree with it.
  • Put your prices and product information on your website as plain text or a table, so an agent can read them in one go.
  • Use your own tracking domain for email. According to Rocha, a shared domain can end up on a blocklist because of another customer's behaviour, and then your clean emails suffer too.

How it works under the hood

More tools is not automatically better. In his own sessions, Rocha expands the collapsed tool calls to see what the agent really did. In one case the result no longer fitted in the context window, and fewer tools would not have made that result smaller. His tip for anyone building an MCP server: expand a call you make every day and see how much of the result the model really needs. With Code Mode, Cloudflare goes to the other extreme: the whole API through two tools, search() and execute().

This news touches chapters 9, 23 and 24 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.

What you can do with this

Want to know what this means for your business? Plan an introduction: half an hour, no sales pitch, but an honest answer to the question where to begin. Rather read on first? The course AI(l) or nothing is free and starts with the basics.

Written in response to a publication by Latent Space: https://latent.space/p/resend/