On January 14^th, I attended a live event titled ChrisGPT Presents AI Agents for Everyone, hosted by the American Society for Engineering Management (ASEM). The speaker was Chris McDermott, a digital transformation leader at Denso and a CrossFit-owning Navy reservist who has successfully "cloned" himself into a digital twin to handle administrative overload.

The presentation was a wake-up call. As professionals, we often drown in non-value-added work. McDermott’s philosophy is simple: if we can offload the boring, repetitive tasks to AI agents, we can reinvest that time into our teams and high-value leadership activities. Inspired by his success—saving hundreds of hours annually—I am starting a project to automate the cleanup of my personal email inbox.

Here is a summary of the key concepts from the talk and how I plan to apply them to my "Inbox Scrubber" project.

McDermott defines an agent as software that can "perceive inputs, reason about them, and take action toward a goal". Unlike a standard automation that just follows a rule (if X, then Y), an AI agent adds a layer of "reasoning" in the middle.

The workflow looks like this:

1. Trigger (Input): The event that starts the process
2. Reasoning (The Brain): The AI evaluates the input
3. Action (The Hands): The automation performs the task

McDermott shared a case study where he built an email triage agent that was 90% effective at sorting requests and drafting replies. I plan to adapt this concept. Instead of drafting replies, my agent will focus on "scrubbing"—identifying junk or low-value emails that don’t need archiving.

• Trigger: I activate the AI Agent to analyze my email inbox.
• Reasoning: The agent analyzes the sender, subject line, and body content. It asks: Is this a receipt? Is this a newsletter I never open? Is this spam?
• Action: If it’s junk, delete it. If it’s questionable, tag it for review. If it’s important, archive it.

One of the most valuable insights from the Q&A was about safety. When giving an agent permission to act on your email, you need guardrails. McDermott advised that unless the AI has a 90% confidence level, it shouldn’t act alone.

For my project, I will implement a "preview mode" first. As McDermott suggested, you don’t want an agent running "squirly" in a live environment. I will initially set the agent to label emails as "To Delete" rather than actually deleting them, allowing me to audit its performance before giving it full autonomy.

During the live demo, McDermott built a research agent in under 10 minutes using Zapier. While there are many tools available, he highlighted Zapier as the fastest entry point for beginners because it now allows you to build agents using natural language prompts. You simply tell the interface what you want the agent to do, and it builds the workflow structure for you.

The goal isn’t to become a programmer; it’s to become a better manager of time. As McDermott noted, we don’t care if we do the task or if the agent does the task—we only care about the result. I’m ready to get my result: an inbox zero state, courtesy of my new AI assistant.