Building an Agentic Workflow

I’m sure you can relate: every morning, I open my phone to see if there are any new tips on using AI, and then spend an hour filled with rage at the definitely horrible news I’m seeing. There had to be an easier way, and that’s what robots are for.

That’s not to say it’s something I couldn’t have figured out on my own. RSS feeds exist, after all. But I wanted to do something cool.

I took a deep breath and made myself comfortable. I rubber-ducked with ChatGPT to come up with a plan for a hot minute. It gave me a “kickoff prompt,” I pasted that into Cursor… and suddenly I owned a moderately intelligent news robot.

Step 1: “Please Build Me a Robot”

“Hey,” I said to ChatGPT. “I’d like to only see the news I want, and avoid the news I don’t.”

“Cool. Here’s a full spec, a data model, and a plan. Sound good?” Chat said back. Obviously, I’m summarizing a bit.

Did it sound good? I honestly didn’t know, because it used all those scary terms and confusing acronyms I’d read about in articles on agentic workflows. But I figured that between the two of us, my robot friend probably knew more than I did on the subject.

“Sounds great!” I said.

I pasted the prompt into Cursor, and it started creating files faster than I can eat a package of Oreos. For educational purposes, though, I also asked it to explain what it was doing in layman’s terms, so I’d have a general idea when the time came to write about it.

And there was still a lot of manual work required. I did stuff! And you’ll have to do stuff too.

Step 2: The Bot Gets a Brain (Claude)

This was my contribution. One of the manual tasks I mentioned. I had to create an API key and then paste it into a chat box. It was hard work, I promise.

We hooked the project up to Claude (Anthropic’s AI model) so it could:

• Read articles
• Summarize them
• Extract actionable tips
• Score relevance

This part was crucial. It had to find the right articles, discard the bad ones, and focus specifically on content that provided actionable tips—how to use different models, emerging high-value tools, and things I could actually do. I wasn’t interested in generic AI news or endless think pieces about how economists believe AI will impact jobs.

Why did I choose Claude? Because they have a great pay-as-you-go model that doesn’t require a monthly subscription. That’s really all there is to it. It’s incredibly inexpensive, and I think it’s excellent. It’s also the model I use most often for coding.

Step 3: It Learned How to Find News

Instead of developing a severe case of high blood pressure from seeing all the bullshit that crosses the newsfeed every day, we taught our robot friend to find the news itself.

Every morning at 8 a.m., it pulls from RSS feeds (AI blogs, news sites, etc.) and grabs anything new from the last ~24 hours. Some days, that’s well over 200 articles. It then scores them for relevance, extracts any actionable tips, and decides what’s worth keeping.

Step 4: It Writes the Newsletter, of Course

This was part of the original ask when I talked to ChatGPT. The obvious solution was to email me the articles so they’re waiting when I sit down at my desk with a cup of coffee. It’s like a warm daily welcome from an old friend: The News.

And all of this is automated. Once the best articles are collected, the system formats them as an email and sends it off.

The email portion was another manual task I had to handle (and you will too, if you’re playing along). I had to set up an account with Kit, grab yet another API key, configure DNS records properly, and wire everything together. Cursor can’t do all of that for you, at least not yet.

So yes, the entire “daily email workflow” is now a robot’s job. I don’t have to do anything other than sit down at my computer in the morning.

Step 5: It Shows Up on the Website

The same articles used in the email also appear on the website. This wasn’t part of the original plan, but after wanting to re-read articles and digging through old emails a few times, it felt like a no-brainer.

Now the email and the website are in sync. I get daily content, and everything is stored on the homepage for easy reference later.

In summary, I went from “What is an API?” to “This thing crawls the web, filters AI content, summarizes it, emails subscribers, and updates my website automatically.”

And I did it in roughly a handful of hours spread across a couple of days. It’s been running for more than four months without interruption. You can head over to AI Daily Rundown to see it for yourself… and subscribe to the newsletter.