AI Tinkerers STL

Agent Mail System

A Replacement for Software Engineering Talent

How I use headless agents and MCP Agent Mail to ensure quality and continuous operation of my sprawling computer vision project.

ClaudeMCP Agent MailYOLOv8cron

Presented at the Sketch Development office on February 4, 2026.

View event post on LinkedIn

Live at AI Tinkerers STL

Full-size presentation photo from Sketch Development office, February 4, 2026.

Mark presenting Agent Mail at AI Tinkerers STL at the Sketch Development office on February 4, 2026

Click to fullscreen

The Problem

DeerAI Tracking Response

A computer vision system that detects, tracks, and deters deer using:

*Multiple Wyze/Reolink cameras covering property
*YOLOv8 + MegaDetector for animal detection
*Kalman filtering for multi-object tracking
*Continuous retraining pipeline on new data

My Weakness

I have strong middle management muscles but very weak software engineering muscles.

The project sprawls across data pipelines, model training, inference servers, and mobile notifications. Keeping it all running is a full-time job I don't have time for.

The Solution: Hire Middle Managers

🌙

Night Watchman

Runs nightly via cron. Checks system health, validates data pipelines, reports anomalies.

☀️

Day Watchman

Monitors inference quality during active hours. Escalates detection failures.

🎓

Model Training Master

Coordinates retraining cycles. Validates new model performance before deployment.

By monitoring their communications in the project mail inbox, I can manage agents instead of writing code.

MCP Agent Mail

View on GitHub

What It Is

An MCP (Model Context Protocol) server that provides a messaging system for AI agents to communicate with each other and with humans.

+Project-based organization
+Agent identity registration
+Inbox/outbox messaging
+File reservations for coordination
+Thread-based conversations

Why It's Manager-Friendly

Instead of debugging code or reading logs, I read email threads between my agents.

“Night Watchman noticed the inference queue backed up at 2am. Model Training Master is investigating whether the last deployment caused a regression.”