The Email Problem Nobody Solves
Every productivity guru tells you to batch your email. Check it twice a day. Use filters. The advice is fine for people with 20 emails a day. When you run seven businesses, your inbox is not a communication tool. It is an intake system for opportunities, obligations, and emergencies arriving 24 hours a day.
I needed something different. Not an email filter. Not a canned autoresponder. An AI concierge that reads every incoming email, understands who sent it, knows the context of our relationship, and responds intelligently — as itself, not pretending to be me. A digital colleague that handles the front door while I focus on the work behind it.
So I built one. It runs every 15 minutes via a scheduled task. It uses the Gmail API with OAuth2 for full read/send access. Claude handles classification and response generation. And it has been live since March 21, 2026, autonomously handling business communications across active client relationships.
The Classification Engine
Every incoming email gets classified into one of five categories before any action is taken:
BUSINESS — Known contacts from active relationships. These people have opted into the AI workflow by emailing my address. The agent responds as itself — Claude, my AI business partner — not pretending to be me. It can answer marketing questions, discuss project scope, share relevant insights, and move conversations forward.
COLD_OUTREACH — Unknown people with personalized pitches. This triggers the reverse funnel. More on that below.
NEWSLETTER — Mass marketing, subscriptions, promotions. Ignored entirely.
NOTIFICATION — System alerts from banks, hosting providers, domain registrars. Ignored unless flagged by the VIP monitor.
UNKNOWN — Anything that does not fit cleanly. Flagged for manual review. The agent never guesses on ambiguous messages.
The Reverse Funnel
Traditional cold outreach response: ignore it or send a template. Both waste the opportunity. The reverse funnel does something counterintuitive — it engages cold outreach warmly, but with a strategic purpose.
When someone cold-emails me, the agent responds conversationally. It asks what they are working on. It learns about their business. It delivers genuine value — marketing insights, AI implementation ideas, strategic suggestions. Over the course of 2-3 exchanges, the relationship reverses. The person who was trying to sell me something is now receiving free consulting. And the natural close becomes: “I actually help businesses with exactly this. Want to hop on a call?”
The person who cold-emailed to sell me SEO services is now a potential client for my agency. The funnel reversed. And the AI handled the entire nurture sequence.
Surge Mode: 3-Minute Response When It Matters
The standard scan runs every 15 minutes. But when the agent detects a new reply from an active conversation, it activates surge mode — a temporary 3-minute monitoring cycle focused exclusively on that contact.
When a key contact replies, the system creates a dedicated rapid-response task that checks for follow-up messages every 3 minutes. After one hour of inactivity, surge mode automatically disables itself. During that hour, the contact experiences near-real-time conversation with the AI.
This solves the biggest problem with scheduled email agents: the 15-minute gap feels robotic when someone is in an active back-and-forth. Surge mode makes the conversation feel natural and responsive while still being fully autonomous.
The Work Order Builder
When contacts express interest in a project — a website, a content campaign, an SEO audit — the agent does not just say “let me have Will call you.” It becomes a consultant.
Through back-and-forth email conversation, the agent asks clarifying questions about goals, audience, features, timeline, and existing branding. It assembles a rough scope document through natural dialogue. When the prospect is ready for pricing, the agent escalates to me with the full context packaged in Notion — not a vague “someone is interested” note, but a structured work order ready for pricing and proposal.
The AI handles the consultative selling. I handle closing and pricing. The division is clean and plays to each party’s strength.
Per-Contact Knowledge Base
Every person the concierge communicates with gets a profile in a dedicated Notion database. Each profile contains background information, active requests, completed deliverables, a research queue, and an interaction log.
Before composing any response, the agent reads the contact’s profile. This means the AI remembers previous conversations, knows what has been promised, and never asks a question that was already answered. The contact experiences continuity — not the stateless amnesia of typical AI interactions.
The research queue is particularly powerful. Between scan cycles, items flagged for research get investigated so the next conversation elevates. If a contact mentioned interest in drone technology, the agent researches drone applications in their industry and weaves those insights into the next reply.
Frequently Asked Questions
Does the agent pretend to be you?
No. It identifies itself as Claude, my AI business partner. Contacts know they are communicating with AI. This transparency is deliberate — it positions the AI capability as a feature of working with the agency, not a deception.
What happens when the agent does not know the answer?
It escalates. Pricing questions, contract details, legal matters, proprietary data, and anything the agent is uncertain about get routed to me with full context. The agent explicitly tells the contact it will check with me and follow up.
How do you prevent the agent from sharing confidential client information?
The knowledge base includes scenario-based responses that use generic descriptions instead of client names. The agent discusses capabilities using anonymized examples. A protected entity list prevents any real client name from appearing in email responses.
The Shift This Represents
The email concierge is not a chatbot bolted onto Gmail. It is the first layer of an AI-native client relationship system. The agent qualifies leads, nurtures contacts, builds work orders, maintains relationship context, and escalates intelligently. It does in 15-minute cycles what a business development rep does in an 8-hour day — except it runs at midnight on a Saturday too.
{
“@context”: “https://schema.org”,
“@type”: “Article”,
“headline”: “I Built an AI Email Concierge That Replies to My Inbox While I Sleep”,
“description”: “An autonomous email agent monitors Gmail every 15 minutes, classifies messages, auto-replies to business contacts as an AI concierge, runs a reverse.”,
“datePublished”: “2026-03-21”,
“dateModified”: “2026-04-03”,
“author”: {
“@type”: “Person”,
“name”: “Will Tygart”,
“url”: “https://tygartmedia.com/about”
},
“publisher”: {
“@type”: “Organization”,
“name”: “Tygart Media”,
“url”: “https://tygartmedia.com”,
“logo”: {
“@type”: “ImageObject”,
“url”: “https://tygartmedia.com/wp-content/uploads/tygart-media-logo.png”
}
},
“mainEntityOfPage”: {
“@type”: “WebPage”,
“@id”: “https://tygartmedia.com/i-built-an-ai-email-concierge-that-replies-to-my-inbox-while-i-sleep/”
}
}