Audition AI Productivity Platform
Back to Blog
Audition AI email delivery agent automation workflow diagram
AI Automation

How We Automated 100% of Our Weekly AI Brief Email with Audition AI

By Benjamin Saberin, CEO & Co-Founder

We automated 100% of our Weekly AI Brief email using two AI personas in Audition AI—a Creative agent (GPT-5 Chat) for content generation and an Operational agent (GPT-5 Pro) for subscriber iteration and delivery.

This wasn't about convenience. It was about proving that real-world AI agents can handle complex, business-critical workflows reliably and safely—achieving full Level 2 AI maturity.

From Manual Sends to Agent-Driven Automation

The Weekly AI Brief has two parts:

  1. The blog post at audition-ai.com/blog
  2. The subscriber email — sent early, with at least one bonus insight not in the article.

In the early days, our first 50 subscribers were simply BCC'd in Outlook. Fast forward: we wanted better deliverability, click tracking, and a story worth telling in our demos. Instead of Mailchimp or CRM mail merge, we challenged ourselves: could Audition AI run the entire process?

The Setup

O365 Shared Mailbox

Enterprise email delivery with full audit trail

Google Sheet

Subscriber names and emails in a simple, accessible format

Two AI Personas

Creative + Operational agents working in tandem

Persona 1

Creative Agent

  • GPT-5 Chat, high creativity
  • Writes blog posts and crafts email content in our established tone
  • Generates personalized email templates with placeholders and unique tracking links
Persona 2

Operational Agent

  • GPT-5 Pro/Reasoning, low creativity
  • Connected to Google Sheet for subscriber data
  • Iterates through all subscribers, generating draft emails with personalized greetings
  • Schedules every email from the drafts folder

Challenges & Breakthroughs

We had to request a new "Schedule" feature for Outlook integration — and wait a few days for it to go live. Our first "success" wasn't quite that: Outlook failed to send some emails when too many were scheduled at once.

This week, we'll space out sends — and we're already imagining a feature where each subscriber gets their Brief at the start of their local day.

Blunders Along the Way

Iteration Problem

An early version kept creating just one or two drafts and then asking if we'd like it to do the rest. We replied with variations of "Yes please just do them all and stop asking" roughly a thousand times.

Fix: Model selection (GPT-5 Pro) + lowering creativity level

Rate Limiting

The O365 API enforces rate limits that our eager agents kept hitting. This is a common problem in AI development — and harder to solve than most people expect.

Fix: Better prompting + built-in self-throttle feature

Hallucination

The agent claimed "I've sent all the emails" when it hadn't even called the functions to draft them. GPT-5 Chat at high creativity is more susceptible to this.

Fix: Prompting + model choice — we're not seeing this at all now

Personalization & Tracking

~75%
Click-through rate
Highly engaged, opt-in base
Personalized Greetings
"Dear [First Name]" for each subscriber
Full Analytics
Google Analytics + Microsoft Clarity

Impact

15+ hrs/week
Saved across all AI automations
100%
Automated delivery
Growing
Subscriber engagement

Where This Fits: Level 2 Maturity

L0
L1
L2YOU ARE HERE
L3

This automation represents full Level 2 maturity

Agentic Assistance — AI agents handling multi-step tasks with appropriate autonomy. Complete workflow execution, human-in-the-loop for critical decisions, integration with business systems, and measurable process improvements.

The ultimate goal: SAFE, compliant AI agents that can manage entire inboxes while still giving humans the opportunity for review.

What's Next: The Path to Level 3

L3

Strategic Asset — Our Next Milestone

  • Apply learnings to other communications (e.g., "What's New in Audition AI" for customers)
  • Experiment with deeper personalization, reply handling, and time-of-day sends
  • Full workflow ownership with strategic reliability — AI as a core driver of business communication

Catch up on past Weekly AI Briefs

Visit audition-ai.com/blog to explore our archive of AI insights and strategies.

Learn More About AI Maturity

Dive deeper into our framework for measuring and advancing AI capabilities. See how to graduate from initial enablement to strategic assets.

Explore the AI Maturity Model

Frequently Asked Questions

How long does it take to set up AI email automation with Audition AI?

We achieved full 100% automation in a few weeks, including solving challenges like API rate limiting and AI hallucination. The setup involves configuring two AI personas—a Creative agent for content and an Operational agent for delivery—connected to your email system and subscriber data.

What AI model works best for email automation?

GPT-5 Pro with low creativity settings works best for operational tasks like iterating through subscriber lists and scheduling emails. GPT-5 Chat with high creativity excels at content generation and crafting personalized email templates. Using the right model for each task is key to reliable automation.

Can AI agents handle business-critical email workflows safely?

Yes, with proper guardrails. Our Level 2 maturity automation includes human-in-the-loop review opportunities, self-throttling to avoid API rate limits, and model selection strategies to prevent hallucination. The goal is SAFE, compliant AI agents that can manage workflows while preserving human oversight.

What is Level 2 AI maturity for email automation?

Level 2 (Agentic Assistance) means AI agents handle complete multi-step workflows with appropriate autonomy. For email automation, this includes reading subscriber data, generating personalized content, creating drafts, and scheduling sends—all without manual intervention, but with human review capabilities built in.

How do you prevent AI hallucination in automated email workflows?

We solved hallucination through model selection (GPT-5 Pro instead of GPT-5 Chat for operational tasks), lowering creativity levels, and improved prompting. High-creativity models are more prone to claiming tasks are complete when they haven't executed the required functions.

Like this content?

Subscribe to our weekly brief for more insights on AI automation and enterprise AI strategy

Subscribe to Weekly Brief

Want to See This Agent in Action?

Discover how Audition AI can automate your business-critical workflows with reliable, compliant AI agents.

#AuditionAI
#AIAutomation
#AIAgents
#EmailAutomation
#WorkflowAutomation
#EnterpriseAI
#GPT5