Build a Custom Email Client with AI Agents

What this is, in plain English

This is the story of someone building their own email app, exactly the way they wanted it, by directing several AI coding agents (AI tools that write and edit software for you) instead of coding it all by hand. The app runs on their own Mac, keeps Gmail as the real source of emails, and adds custom touches like split inboxes, keyboard shortcuts, a 20-second “undo send”, and one-click unsubscribe.

Be honest about the level: this is an advanced, multi-session build, not a copy-paste recipe. The author moved between paid AI tools (one to draft, another to polish and fix), iterating many times. You cannot reproduce it in one sitting, and you need to be comfortable running a development project. What is genuinely useful here, and very copyable, is the method (how to drive AI agents to build a custom tool step by step) and the actual prompts they used.

What you can use it for

The real lesson is bigger than email: this is how to build a custom tool tailored to you when no off-the-shelf app fits.

• A bespoke email client. Split inboxes, your own rules, and shortcuts that match how you actually work.
• Any personal productivity tool. The same drive-the-agents method builds a custom task manager, reading-list app, or dashboard.
• A learning project. Watching agents build, debug, and speed up a real app teaches modern app development fast.
• A prototype to validate an idea. Get a working version of a tool in front of yourself (or users) before committing real engineering time.

Tools you need

• AI coding agents (paid): the author used Codex to draft the first version, then Factory to polish the interface, fix bugs, and tune performance, reaching for Claude Opus or GPT-5.5 on harder problems. Any capable coding agent can play these roles.
• Gmail (freemium): stays the source of truth for your email; the app reads and writes through it.
• Google Workspace CLI (free): a command-line tool to sign in, fetch emails, and manage labels in Gmail.
• Superhuman (paid): not required, just the polished email app the author used as a benchmark for the features to aim for.

How it actually works

You build in passes: draft, then polish, then fix, feeding the agent clear instructions and the prompts that worked. The realistic shape:

1. Draft the first version. Describe the whole app in one detailed prompt so the agent has the full picture:

   I want you to build me an email client so I can have things exactly how I want them. I'll only be using it on my MacBook. It can run locally to start. Gmail should stay the source of truth. The end product has split inboxes, shortcuts, command palette, reply/compose, 20 second undo send, one-click unsubscribe, search, rules, filter sync, email rendering and built in a way that an agent can use it natively.

   Expect a rough, unpolished first version.

2. Polish the interface. Switch to a refinement-focused agent and have it improve the look, feel, and usability.

3. Hunt down lag. Make every action feel instant:

   Investigate why there is a delay with anything in this app. I want everything to feel instant.

   The fix is usually to show saved (cached) data immediately, refresh in the background, and update the screen optimistically (assume an action worked and show it instantly while it finishes behind the scenes).

4. Sync rules with Gmail. Make sure labels and rules you set in the app actually use Gmail’s own filtering, not just a local file, so they work everywhere.

5. Build full reply and compose. Including reply-all, editable to/cc/bcc, attachments, and send-and-archive with the 20-second undo:

   Build full reply functionality. Default reply-all, editable to/cc/bcc, attachments. Cmd+Enter sends and archives, but waits 20 seconds before actually sending so Cmd+Z can undo.

6. Tame email rendering. Get plain text, HTML newsletters, receipts, calendar invites, and inline images to display cleanly. This is fiddly, so expect several rounds.

Words you’ll see, explained

• AI coding agent: an AI tool that writes, edits, and debugs software from your plain-English instructions.
• Email client: the app you read and send email in (Gmail’s website is one; here you build your own).
• Source of truth: the real, authoritative copy of your data; Gmail stays that, the app is just a nicer window onto it.
• Cache: a local copy of data shown instantly so the app feels fast while it refreshes in the background.
• Optimistic update: showing an action as done immediately, then confirming it behind the scenes.
• CLI: a command-line tool you type instructions into, used here to talk to Gmail.

Original source

Based on “Ben’s Builds #3 - an email app” by Ben’s Bites, documenting the author’s process of building a personalized email client by orchestrating several AI coding agents.

Notes & variations

• Do you even need this? If a ready-made client (Gmail, Superhuman) already fits you, building your own is a big effort. This pays off when you want behavior no existing app offers, or you want to learn by building.
• Free-tier alternative: the paid agents give the most polished results, but you can draft with a local open-source coding model (via Ollama or LM Studio) for free, expecting rougher output and more iteration. The Google Workspace CLI is free.
• Common pitfall: leaning on a single agent and one prompt. The author’s progress came from iterating and switching tools (draft with one, refine with another).
• Tip for better results: spell out every feature and constraint up front, as in the first prompt. A precise brief is what let the agents build the right thing.