I used to spend 90 minutes a day on email. Now it’s about 15.

Here’s exactly what I built and how.

The Problem With Email

Email is mostly reactive. Someone sends you something, you respond. The problem is that most of what arrives doesn’t need your specific judgment — it just needs a response, or to be filed, or to wait.

Once I started seeing my inbox as a triage problem, the solution became obvious: most messages follow predictable patterns. Patterns can be automated.

What I Automated

Categorization — Every incoming email gets labeled automatically: Client, Newsletter, Administrative, Personal, Urgent. This alone saves mental energy. I’m not making micro-decisions about what to read first.

Draft responses — For common request types (meeting requests, status updates, questions I get repeatedly), AI drafts a response. I review and send. This takes 15 seconds instead of 5 minutes.

Newsletter digest — Instead of newsletters hitting my inbox, they get collected and summarized into a weekly digest. One email instead of 30.

Automatic archiving — Anything that matches patterns I’ve defined (shipping notifications, receipts, calendar confirmations) gets archived immediately. Never touches my attention.

The Stack

  • Gmail filters for the initial heavy lifting
  • Make to route emails to different workflows
  • Claude API for the actual reading and drafting
  • A simple Notion database to track response templates

Total cost: about $8/month in API calls.

What I Still Do Myself

The 20% I handle manually is: anything emotionally important, anything with genuine ambiguity, and anything where my specific context matters.

The system is good at pattern-matching. I’m better at nuance. So the system handles patterns, I handle nuance.

How to Start

You don’t need my exact stack. Start with Gmail filters and labels. That alone will change how you experience your inbox.

Once you have categories, identify which category takes the most time. Build one automation for that. Measure the time saved. Iterate.

The 80% reduction didn’t happen in a day. It took about three months of incremental improvements. But each improvement compounded.