#3: How to Build a Newsletter Butler

Welcome to the Inside Track. Together we explore how AI is rewriting business rules in real-time, focusing on opportunities most don't yet see. We dive into productivity tactics, business opportunities, and industry shifts to help you understand and leverage AI as it rapidly evolves.

Here comes a tactical post. The best way to understand and find opportunities as the AI space evolves is to combine two things:

1. Forming high-level thoughts and mental models to understand second-order shifts.

2. Getting our hands dirty, building agents, testing tools.

Inside Track is all about exploring both layers. Today I'll show you a tactic.

I subscribe to many newsletters. You probably do, too! It's an important source of information for me. But it also leads to a messy inbox. Different newsletters come in at different times throughout the day, and so on.

There must be a better way, no?

Yes.

I spent ≈an hour building my own AI newsletter butler.

Each morning I receive one email in my inbox from my butler with a categorized list of newsletter editions, including summaries and links to read more.

Super-fast to scroll through:

Behind the scenes, the butler does all the heavy lifting; fetching emails, processing, organizing, etc.

Below I'll describe the different steps involved, because it's a good example of finding a small thing to improve, investing a little time to automate it, and reap the rewards. Longer term, the opportunity is to make a thousand such improvement automations that compound over time.

Building the Newsletter Butler

I set up a new email address for my butler. That's where I receive the newsletters.

Then I had to go through all my newsletter subscriptions and unsubscribe from my main account, and resubscribe with the butler account.

Next, we fire up a new automation in my (current) go-to tool – N8N.

The automation runs every morning to deliver the summary to my inbox at 8am.

The automation is built up of four main components:

1. Fetching emails

2. Processing them

3. Create summary

4. Clean up

I'll go into detail of each below:

Fetching emails

The butler fetches all incoming emails in the past 24 hours. It then filters out any that have already been processed. It then uses a text classifier to split out newsletter emails and ignore other emails (for instance a newsletter subscription confirmation, or those "signed on" notification emails from Gmail etc.).

(In the screenshot you see it also classifies emails from Readwise separately. That's because I'm already looking at ways to embed daily Readwise highlights into the summary email).

Processing emails

Now it gets more fun. The butler first takes each newsletter email and uses an OpenAI prompt to extract the read online link for each newsletter. We need this to build a summary email with clickable titles.

Next step is to categorize the newsletters and generate a short summary for each.

I'm currently using the following categories:

• AI

• Crypto

• Tech

• Business

• Virtual Worlds

• Entertainment

• Other

The categories can be changed by editing the prompt that's used to categorize.

Here's the prompt:

Classify the following newsletter into one of these categories: Business, Crypto, Entertainment, AI, Tech, Virtual Worlds, Other.
  
Then, summarize the content in two short sentences, focusing on key points.  

Newsletter Title: {{$json["subject"]}}  
Newsletter Content: {{$json["body"]}}  

Provide your answer strictly formatted as:  
Category: [CATEGORY]  
Summary: [SUMMARY]

Create and send summary

The butler has everything it needs to create the summary. It takes the inputs and generates a categorized list, each bullet point includes a title, name of the sender, summary and links to the post itself.

This makes up the body for the email. It adds today's date in the subject (to make a searchable archive) and sends it to me.

Clean up

Being a proper butler, it then tidies up, ready for tomorrow; Marks processed emails as read, and adds a label (processed), so they're easy to filter out for later automation runs.

All in all, the workflow looks like this:

All you would need for something similar is an OpenAI API account with some funds, and an N8N account.

The workflow queries OpenAI for three things:

1. Email classification

2. Extracting link

3. Categorize and summarize

After a couple of daily runs, the cost for all of them average out at $0.03 per run. Which is...acceptable.

That's it for now. Go forth and automate!

PS: Let me know if you like these tactics type posts!

Take care!

– Martin