I built an AI email triage agent and tested it on my own 300-email backlog first. One session later: a clean inbox, three real decisions, and a system that runs in minutes a day.
You open your inbox and immediately don't know where to start. Urgent client requests sit next to newsletters and automated receipts with no way to tell them apart. Every email demands a micro-decision — read it? skip it? reply? delete? That's decision fatigue at scale, and it happens before you've done a single minute of real work.
303 unread inbox messages, 640 newsletters, 81 flagged — impossible to know what actually matters.
Action items, FYIs, receipts, and cold outreach all lived in the same pile with no labels, no structure.
Time spent searching for real emails, re-reading things already decided, and context-switching — invisible but real.
Five components working together — connecting your Gmail, classifying every email against a personal rulebook, proposing batch actions, executing only what you approve, and logging everything.
Your rules are the brain of the agent — defined once, applied to every email automatically. Below is a sample of the rules we built in this session. Each one encodes a real decision you used to make manually, every single day.
Click any of the boxes to show details
Set up once, then it runs in minutes a day. Each step is built around one principle: you stay in control of every decision that matters.
A secure connection you approve once — no passwords shared, no access beyond reading emails and organizing them. Then we encode how you decide: who always gets a reply, what gets archived unread, how newsletters differ from client requests. 21 rules, written once, applied to every email after that.
The agent reads the backlog and sorts everything into a 7-section digest: Action Required, Awaiting Reply, Calendar, Clients, FYI, Newsletters, Archive. Each email comes with the recommended action and the reason — traceable to the exact rule that triggered it. No black box.
This is the key decision that turns a tool into an agent. For each category of email, you decide: does the agent handle it automatically, surface it for your review, or flag it for someone else? You can change your mind at any time.
Agent reads, classifies, and proposes. Nothing moves until you say go. Full control, every run.
Agent auto-handles low-risk categories (newsletters, receipts, cold outreach). Surfaces a summary for you to review, then archives after confirmation.
Runs automatically — daily or weekly. Clears the noise, surfaces only what needs a human, sends a digest. Inbox stays clean without you touching it.
Approved actions run in batch. Every single one — what was done, to which email, why — is written to a permanent record. We ran the loop a few times, refined the rules where edge cases appeared, and surfaced the three items that genuinely needed a human. Inbox 303 → Inbox 0.
This isn't a one-time cleanup. These are the reusable components that make the next triage take minutes, not hours.
A live, persistent bridge between the AI and your inbox — approved by you once, limited to reading and organizing. Reconnects instantly each session.
Your personal classification logic — encoded once, reusable forever. Updated as your preferences evolve.
A standardized output structure: Action Required, Awaiting Reply, Calendar, Clients, FYI, Newsletters, Archive. Consistent every run.
Every triage action recorded with timestamp, email details, and the rule applied. Full audit trail, always recoverable.
Claude remembers your preferences across sessions — your rules, your exceptions, your context. No re-explaining every time.
A documented process you can run daily in minutes: connect → pull → classify → approve → execute. The agent handles the rest.
Three emails remain — each one a genuine human decision. The inbox is now a to-do list, not a landfill. Every item visible has a reason to be there.
The rules are set. The connections are live. Running triage tomorrow takes the same 5–10 minutes as it did today — but without building anything from scratch. The system compounds.
Time estimate based on the actual action record: ~150 emails triaged across the first sessions, ~100 archived with reasons logged, review time of 5–10 minutes versus roughly an hour a day of manual sorting — about 4 hours back every week.
The most common question we get — answered plainly. This system is built on trusted platforms our clients already use every day, and it never acts without your sign-off.
Nothing moves without your sign-off. The agent proposes; you decide — and you set how much it handles over time.
"Archive" only means "out of the inbox." Every email stays in your account, searchable and recoverable.
A permanent audit trail: what was done, to which email, and the rule behind it. Full transparency, always reversible.
Built on tools your business likely already trusts — Google Workspace and Anthropic's Claude. You approve access once; no passwords are ever shared.
"I was the test case — 303 emails, one session. My inbox is now a to-do list, not a landfill. I build every system on my own business before I bring it to yours."— Mathieu Arsenault, Founder, Think Forward Solutions
Book a 15-minute call — I'll show you exactly what this would do on your inbox. Your workflow, your clients, your rules.