Building a Personal News Editor: How I Tamed 161 RSS Feeds with an AI Agent

The Problem: Information Overload

Every morning, I used to wake up and do the same thing: grab my phone and scroll through dozens of unread messages, trending hashtags, and push notifications. After ten minutes, I felt like I knew a lot but couldn't name one thing worth caring about. That's the paradox of the modern news diet—we're fed constantly, yet starved for signal.

One day, I decided to fight back. I set up an AI agent to curate my news, but quickly learned that feeding it raw internet was like asking a librarian to summarize a garbage dump. The results were shallow and repetitive.

Step One: Building a Trusted Feed Pool

My first attempt was simple: a scheduled task that scraped five tech stories each morning. It worked, but only for a few days. The same press release would appear three times with different headlines, and yesterday's news would show up as 'new' just because another site rewrote it.

The fix was to stop relying on random web searches and start with a curated list of RSS feeds. I've collected 161 over the years, but quantity isn't quality. I organized them into a tiered pool based on trust:

• Primary sources: official company blogs and financial reports (OpenAI, DeepMind, etc.)
• Serious media: Bloomberg, Reuters, The Information, WSJ
• Secondary aggregators: The Verge, Techmeme, MacRumors
• KOLs and bloggers: for hands-on opinions and leaks

This isn't just about avoiding fake news—it's about preserving context. The closer you get to the original source, the less distortion you'll see.

Step Two: Organize with a Modern RSS Reader

Managing 161 feeds in one flat list is chaos. I use Folo, an RSS reader that lets me organize subscriptions into a tree structure. It feels like building my own magazine—six main categories covering tech, gaming, culture, AI, and autos.

But the real magic is Folo's CLI interface. It turns my reader into an external brain for my AI agent. The agent can read unread items directly from Folo, which means it's already working with a filtered, credible set of sources—not random web noise.

Step Three: Teaching the Agent to Judge

An agent doesn't know what you're tired of. After days of AI model news, I typed: "Too much AI stuff. I want more consumer electronics and hardware." The agent saved this to a MEMORY.md file and adjusted future picks. It worked.

That's the key: you have to train your agent like a new intern. It won't read your mind, but it will remember your preferences if you state them clearly.

Step Four: From Chat to a Personalized Newspaper

Chat output is fine for quick scans, but I wanted something prettier. I asked the agent to generate an HTML page with a clean card layout, separating the five top stories with titles, core facts, and why they matter. Each card has a link to the original source. It feels like reading a newspaper designed just for me.

I took it further with a long-running story: the foldable iPhone. Rumors, denials, re-rumors—it's a mess. I asked the agent to build a self-contained HTML tracker with a timeline, a parameter tree, and source cards. It categorized each rumor as confirmed, multi-sourced, single-source, or unverifiable. The result was a clear, structured report that killed my FOMO.

Why This Matters

In 2025, Merriam-Webster chose 'slop' as its word of the year—referring to AI-generated junk content. We're drowning in it. Building your own trustworthy feed is the best defense.

My agent doesn't make decisions for me; it just filters. I still read the original sources when something matters. But now I spend less time swimming in the flood and more time reading what counts.

You can do this too. Start with a few feeds you trust, organize them, and connect them to an AI. It won't happen overnight, but soon you'll have your own quiet dam against the noise—and that's worth it.