Part 6: Control Your Agent with Email
The simplest control plane: your inbox. No dashboards, no logins, no apps.
Just send an email and the agent executes. You already know how to use it — you've been doing it for decades. That's what makes email the most underrated interface for autonomous agents.
Why Email as a Control Plane?
You could build a dashboard. You could build a Slack bot. You could build a custom app. But email wins for one simple reason: zero adoption friction.
You already check email
No new tool to learn, no dashboard to bookmark, no app to install. Your inbox is already open.
Natural language commands
“Pause posting until Friday” works. No menus, no buttons, no syntax to memorize.
Mobile-first by default
Control your agent from your phone, your couch, the airport, anywhere you have email.
How It Works Technically
No code here — just the flow. The implementation details come when you build it. Here's what happens every 5 minutes:
Agent checks email every 5 minutes using Gmail API
Emails are classified by sender — only authorized senders trigger actions
Peter’s emails → Claude parses the natural language into a structured command
Command is executed → Agent replies with confirmation
Everyone else’s email → Marked as read, ignored. Zero responses to strangers.
Real Commands Jenny Understands
These aren't hypothetical. Every command below is something Jenny processes in production, parsed from plain-English emails.
Posting Control
“Pause all posting”
Agent stops all scheduled posts across every platform. Confirms via reply email.
“Resume posting”
Agent restarts the full posting schedule immediately.
“Pause LinkedIn only until Monday”
Platform-specific pause with automatic resume. Other platforms keep posting.
“Focus this week on AI tools content”
Adjusts content selection priority — the agent front-loads AI tools articles for all platforms.
Reporting
“Send me this week's analytics”
Generates and emails a full HTML report with engagement metrics, growth data, and top posts.
“How are partnerships going?”
Sends a partnership pipeline summary — who was contacted, who replied, what's pending.
“Sponsor report”
Current sponsor outreach status — pitches sent, responses received, deals in progress.
Content Feedback (Self-Learning)
“The LinkedIn post about AI tools was too long. Keep them under 1000 characters.”
Jenny extracts the lesson, stores it in her memory, and consults it before every future LinkedIn post.
“More hot takes on X, less generic promos”
Adjusts content angle weights — X posts shift toward opinionated commentary, away from promotional language.
Hub Management
“Generate 50 new glossary items”
Triggers the hub content generation pipeline — 50 new glossary entries written, validated, and stored.
“Run a quality sweep on prompts”
Reviews all hub prompt content, flags low-quality entries, and reports findings via email.
The Self-Learning Feedback System
This is the magic. When you tell Jenny “that post was too formal,” she doesn't just fix it once — she stores the lesson and applies it to every future post. The agent gets better over time without code changes. Every piece of feedback becomes a permanent improvement.
The Security Model
Agents that respond to anyone are dangerous. An open inbox is an attack surface. Here's how Jenny locks it down.
How Jenny's Security Works
Peter’s iCloud address
Full command access — the only email that triggers actions
Newsletter content emails
Stored silently, never responded to
Featured.com emails
Stored for Monday PR report, never responded to
HARO journalist queries
Scored by Claude, auto-responded to selected ones via the reply address
Everything else
Marked as read, zero action, zero reply
Bounce detection
Mailer-daemon emails auto-mark prospects as bounced
Strict sender verification is non-negotiable.
Your agent MUST have strict sender verification. An agent that responds to anyone can be tricked into sending spam, leaking data, or taking unauthorized actions. Whitelist your email address and reject everything else. No exceptions.
Building Your Own Control Plane
Email works for us. But the principle is what matters: pick the channel you already live in.
Alternatives to Email
Slack Bot
Good if your team lives in Slack. Real-time responses, channel integration.
Telegram Bot
Great for personal use. Instant feedback, lightweight, easy to build.
Possible but harder to automate. API access is more restrictive.
Simple Web Dashboard
More work to build, but visual. Good if you want a shared interface for a team.
The point: pick whatever you already use every day. The best control plane is the one you'll actually use.
Your Day-1 Command Template
List 5 commands your agent should understand on day 1. Start here:
Pause / Resume
Stop all operations. Restart all operations. The emergency brake.
Send Status Report
On-demand summary of what the agent has done and what's queued.
Change Content Focus
Shift the topic priority for upcoming content. Adapt to what matters this week.
Give Feedback on a Specific Post
Tell the agent what worked or what didn't. It learns and adjusts.
Trigger a Manual Action
Run a specific task on demand — generate content, pull a report, send outreach.
What You Just Learned
- Email is the simplest control plane — you already use it, no new tools to learn
- Natural language commands parsed by the LLM means no rigid syntax to memorize
- Self-learning feedback — the agent improves over time without code changes
- Strict sender verification is non-negotiable for security — whitelist and reject everything else
- Pick the channel you already live in — email, Slack, or Telegram
Next: Learn how to automate video and visual content creation — branded videos generated without an editor.