An AI Approach to Email

Every successful founder gets the same advice: "Hire an assistant." When you ask what that assistant should actually do, the answer is almost always the same — manage your inbox and calendar.

Before you spend $10K+ a year on a person to do that work, here's the AI setup doing most of it for me — including the move that changed how I think about email entirely: hooking Superhuman's MCP into Claude Code.

The cost of email, honestly

I was losing 10+ hours a week to email — context switches, half-finished drafts, scheduling tetris with people three timezones away. The fix wasn't to grind through it harder. It was to stop being the bottleneck.

The default play is hiring a virtual assistant or executive assistant. Offshore VAs run cheap; competent US-based EAs are $25–40/hour. Even at the low end, 20 hours a week of email and calendar gets you well into five figures a year — plus the hours you spend onboarding, training, and managing them.

The alternative is a small AI stack that runs 24/7 and doesn't need vacation. Below is the exact one I use, plus the workflow change that made it click.

The stack: two tools

I used to run more tools alongside these — a separate inbox-triage product, an automation layer for repetitive replies. Once Superhuman's native features caught up and Howie covered scheduling end to end, the rest became overkill. Less surface area, fewer subscriptions, same outcome.

Superhuman — drafting and triage

Superhuman is Gmail with an AI layer. Two features carry the weight for me:

• AI drafts in my voice. Press ⌘+J, type a few bullets, and it expands into a reply that sounds like me. Tighten it, hit send.
• Auto-triage and auto-labels. I describe what I want in plain English ("Label any email about invoices, payments, or receipts as Finance; surface customer emails immediately, batch newsletters") and Superhuman keeps the inbox organized in the background.

The auto-triage is what let me retire a separate triage tool. One product, fewer rules to maintain, the same end state — a quiet inbox that surfaces only what needs a real response.

Howie — scheduling, now via text too

Howie handles scheduling end-to-end. CC it on a thread (or email it directly) and it negotiates the time, confirms with the other person, and adds the event to your calendar.

The newer addition: SMS. I can text Howie with the request and it'll handle the back-and-forth from there. Scheduling no longer requires sitting at my inbox or even being on email at all — I can offload it from anywhere.

Why this beat Calendly for me: nobody loves clicking a Calendly link, and Howie reads thread context, proposes specific times in plain language, and feels like a human assistant instead of a booking widget.

Managing email from inside Claude Code

This is the move that changed how I think about email entirely.

Superhuman shipped an MCP server — Model Context Protocol, the open standard that lets AI tools like Claude connect to external systems. Once you install it in Claude Code, Claude can search your inbox, draft and edit replies in your voice, send messages, and manage calendar events directly.

In practice: I stay in my coding flow and have Claude handle email for me. Same on my phone — I can manage my inbox from inside Claude without ever opening Superhuman. The AI does the work; I never break flow.

Worth noting: MCP requires Superhuman's Business tier or higher. Setup steps live in Superhuman's help docs.

The math, side by side

Hiring an EA

• $10–40K/year depending on whether they're offshore or onshore
• Working hours only
• Has to be onboarded, managed, and covered when away
• Handles a lot more than email — travel, gatekeeping, errands, projects

This stack

• Well under $100/month, total
• Always on, including from your phone via Claude
• No management overhead
• Only does email and calendar

Where this doesn't quite work

Honest read: this stack replaces the email-and-calendar slice of an EA, not a full one. If your real bottleneck is travel booking, expense reports, gatekeeping inbound calls, or owning project logistics — hire the human. The AI stack won't cover that work.

If your bottleneck is genuinely email and calendar (mine was), the AI stack wins on every axis: faster, cheaper, available everywhere you are, and you don't have to manage it.

What changes

After settling into this setup:

• Reply drafting goes from minutes to seconds
• Your inbox stays sorted in the background — newsletters batched, invoices labeled, customer email surfaced
• Scheduling threads disappear, handled via Howie from your inbox or your phone
• Email itself becomes something you can manage from inside whatever AI tool you're already in

You don't need a babysitter for your inbox. You need a copilot — one that's everywhere you are, including inside the tools you actually work in.