MVMP Labs
All posts

Build a custom agent for someone else's customers

By Connor Finlayson7 min read

ChatGPT, Claude, and Cursor are for you. A custom agent is your version of that, for your customers or your team. Eve is how you stand one up.

Build a custom agent for someone else's customers

TL;DR

Building custom agents for businesses is one of the bigger opportunities right now if you're a heavy dabbler in AI.

Not for the people who hang out on AI Twitter. For normal businesses that still talk to their customers the old way.

I've been experimenting with Eve this week, a framework from Vercel. It makes setting up an agent for a business a lot easier than it sounds.

ChatGPT, Claude, and Cursor are for you. A custom agent is for everyone else.

ChatGPT, Claude, Cursor. That's what you use as a business to brainstorm, write, and code. The interaction is between you and the tool.

A custom agent is your own version of that. For your customers. Or your team.

Same idea. Different person on the other side.

"I can just use ChatGPT for that"

This is the bit most people miss. Swap in Claude or Cursor. Same objection.

When someone says they can just use the tool they already have open, they're thinking of a tab. A custom agent is a thing you can hand to someone else.

ChatGPT, Claude, or CursorA custom agent
For the ownerFor the customer or the team
TalksDoes the work, then asks before anything irreversible
Starts over every chatKeeps the thread
Knows the internetKnows that business
Lives in a browser tabLives in Slack, email, or the signup page

You can't give a stranger your ChatGPT, Claude, or Cursor login and say "sign up here."

Those tools are something you rent. A custom agent is something you can charge for, because the client doesn't have to become good at prompting. Their customers just talk.

The opportunity

Most founders I talk to are comfortable using AI in-house. Almost none of them have given that same experience to the person on the other side of the signup flow.

That's the gap. That's the work. That's what I'd charge for.

Using this on your own product is useful. Selling it to other businesses is the actual opportunity. If you already build directories or marketplaces, this is the same game. You're modernizing how that business talks to people.

What it looks like

Signup is the example that makes it click.

Right now the flow is usually a form, a seven-step wizard, then a generic "check your email" screen. A custom agent flips that. The customer talks. The agent already knows the business. It asks the next useful question, and only brings in a human when it should.

Same job as the form. Familiar the way ChatGPT, Claude, or Cursor is familiar. That's why I think it converts better.

Eve is the example

Vercel didn't just write a think piece about agents. They built Eve.

You engage with a custom agent they created, and that agent helps you build more agents. That's the whole point, in product form.

Using Eve is basically building a miniature version of your own Codex or Cursor. Same idea as the tool you already use to write code. Except this one belongs to a business, and other people can talk to it.

People are already taking that further. Software factories. An agent that takes a task from GitHub or Linear, runs it through classify / analyse / implement / review, and comes back with a draft pull request.

Examples you can actually look at

Eve has a bunch of templates. These are the ones I'd click first.

  • Eve itself: the framework is the demo. You talk to it. It helps you stand up an agent.
  • Marketing Team: one lead in Slack, five specialists. Blog in Notion, social in Typefully, email in Resend. Closest thing to a productized service you can fork this week.
  • Foreman (software factory): GitHub or Linear task in. Classifier, analyst, implementer, reviewer. Draft PR out. That's the tiny Codex for a product team.
  • Kody (GitHub maintainer): weekly issue digest to your email, PR summaries, answers @mentions, picks up Linear issues.
  • Content Agent: @mention it in Slack, it drafts blogs, LinkedIn, release notes, and newsletters in a house voice.
  • Sanity Copilot: Slack copilot for a CMS. Queries, edits, drafts. Asks before it touches production.
  • Social / Typefully: drafts for X, LinkedIn, Threads, Bluesky, Mastodon. Queue and publish on approval.
  • Design agent: answers Slack from your approved design guidelines. Won't make things up from the internet.
  • Chat template: a real web chat you can put in front of a customer. Closest starting point for the signup idea.

Full list: eve.dev/templates.

How I'd actually build it

If you already use ChatGPT, Claude, or Cursor, you already use an agent. You just never see the wiring. Eve is that wiring, as a folder you can deploy.

  • An agent is a folder. instructions.md is the brain. tools/ is what it can do. Add a file, Eve picks it up.
  • Start it like any other project: npx eve@latest init my-agent, then pnpm dev. Cursor can write the files.
  • It deploys like a website. vercel deploy.
  • It doesn't flake out mid-job, and it can ask before it sends an email.
  • It's still beta, and it lives in the Vercel world. Fine for experiments. Not a forever, lock-in-free platform yet.

That's why Eve matters here. The idea above is a bit abstract. Eve is how you make it shippable this week.

What I'd do next

Don't keep reading takes. Including this one.

Pick one real business. Yours or a client's. Stand up a signup agent, or fork the Marketing Team template and point it at their actual tools.

Watch one real conversation. That's faster than any more theory.

FAQ

What is a custom agent? Your own version of ChatGPT, Claude, or Cursor, built for a specific business, that customers or a team can talk to.

How is that different from ChatGPT, Claude, or Cursor? Those tools are for you. A custom agent is for everyone else. It lives in their signup, Slack, or email. It can do work, keep the thread, and follow that business's rules.

What is Eve? A Vercel framework that turns an agent into a folder you can deploy, like a small website. It's how you stand one of these up without becoming an AI engineer.

Who is this for? People who already dabble in AI and want to make money helping normal businesses talk to their customers in a chat, not a form.

Ready to learn more? Check out our Playbooks for step-by-step guidance.

View Playbooks

Continue Reading

Related Posts