The AI "Signal Scanner" Agent Build

This week’s issue is a bit lighter than usual — not because there’s less to say, but because I wanted to stay focused on finishing the Signal Scanner agent I’ve been building behind the scenes. The project is finally coming together, and we’re almost at the finish line.

Below, I’ll walk through what I did to get it to this point, what worked, and what still needs tightening. Thanks for sticking with me as I’ve gone a little deeper on the build.

Summary: Meta's CMO, fiercely advocating for the impact of brand-building, suggests retiring the term "performance marketing" as AI blurs the lines between brand and direct response. This signifies a shift in industry mindset towards a more holistic view of marketing effectiveness, where all efforts contribute to both brand equity and measurable outcomes.

Why it matters: This challenges marketers to rethink traditional silos between brand and performance, pushing for integrated strategies where brand building is seen as a driver of long-term performance. It suggests that AI will enable more sophisticated attribution, making every marketing touchpoint accountable and impactful.

Summary: The chosen words of the year, like "AI slop" and "vibe coding," reflect a growing consumer fatigue with an overwhelming and often inauthentic digital landscape. These terms highlight a generation's complex relationship with technology—they're tethered to the internet but increasingly critical of its quality and impact.

Why it matters: Marketers need to be acutely aware of this pervasive digital fatigue and skepticism when crafting online content and campaigns. Authenticity, transparency, and high-quality, human-centric messaging will be crucial to resonate with consumers wary of "AI slop" and digital overload.

Summary: An AI security expert from Google shares crucial insights into data privacy and security risks associated with interacting with chatbots and AI tools. The expert warns against sharing sensitive personal or proprietary information due to potential data breaches, misuse, or unintended model training.

Why it matters: This underscores the critical importance of privacy and security in the age of AI, compelling marketers to exercise extreme caution when using AI tools for content creation, data analysis, or customer interaction. Brands must educate their teams, establish strict data governance policies, and clearly communicate their AI usage to build and maintain customer trust.

Quick reset for anyone new here: I’m building a “Signal Scanner” agent. The idea is to scan a set of RSS feeds and surface a short list of news stories based on rules I define.

I’m using this newsletter as the main use case, but the pattern is pretty flexible. An agent like this could just as easily:

Last week, I walked through how I mapped the workflow and started building in n8n, the tool I’m using to wire this together. I also started documenting the process in a Scribe SOP so anyone can follow along or build their own version.

Before getting into this week’s progress, it’s worth pausing on RSS feeds, since they’re a core part of this agent—and the source of most of my early friction.

An RSS feed is basically a way for websites to publish their latest content in a format that other tools can read automatically. Instead of me checking a bunch of sites manually, those sites push new articles into a feed, and my workflow pulls them in.

In theory, this should be the easy part.
In reality, it wasn’t.

While building the Signal Scanner, I kept running into issues caused by the feeds themselves: broken links, blocked requests, and random errors that would stop the workflow mid-run. Rather than letting those bugs derail everything, I decided to create and manage my own RSS feed list using RSS.app. That gave me more control over the sources and made the workflow far more reliable (read: way less buggy).

Once the feeds were working and I filtered out stories that clearly weren’t about AI or marketing, I needed a way to turn raw articles into something readable. This is where the Message a model node does the heavy lifting.

Instead of just passing links around, this node takes each article and rewrites it into newsletter-ready copy.

Here’s how I used it.

For every article, I wanted the same structure:

Nothing fancy — just clean, consistent blocks I could drop straight into the newsletter.

Each article already had a title, link, and short excerpt from the RSS feed. I passed that into the model and let it do the rest. No extra scraping, no long prompts.

The key was separating how the model should behave from what it was processing. The instructions stayed the same every time; only the article content changed. That made the output much more predictable.

Once configured, n8n sent each filtered article through the Message a model node automatically, one by one. Same instructions, different inputs, clean output every time.

To finish the workflow, I aggregated the results and used the Gmail node to send myself an email with the final output.

Overall, the system worked. The filter narrowed about 225 stories down to 9. A few of those still weren’t as relevant as I’d like, which tells me there’s room to tighten the relevance checks before the final send.

That’s what I’ll be refining next — and I’ll walk through that process in the next issue, which will be a wrap-up of this entire build!

As always, stay curious and have fun!

Best,
Skyler Neal