How I Built an AI Editor from 161 RSS Feeds to Beat the Noise

Every morning, I do the same thing: grab my phone and see what happened overnight. Dozens of unread messages in WeChat, a reshuffled list of trending topics on Weibo, a long chain of "important updates" from news apps, and then a scroll through tech communities full of "just announced" and "major release" posts. Ten minutes later, I feel weirdly empty—I've absorbed a lot, but I couldn't tell you what actually mattered.

That's the modern information paradox. We used to worry about scarcity. Now we're drowning in abundance, and the surplus doesn't just eat our time—it erodes our judgment about what's worth caring about. So I ran an experiment: I built a personal news-filtering system using an AI agent, and I took back control of my daily reading.

Step One: Ditch the Random Scraper, Plug Into a Real Intelligence Network

My first instinct was simple: set a scheduled task for 8 a.m., have the AI collect five tech news items, and send them to me. That was my first briefing assistant. It worked, sort of. Instead of hopping between websites and apps to piece together the night's events, I could pour a coffee and find a summary waiting on my phone.

But the novelty faded within days. The same product launch, written three different ways, would hog three slots in my briefing. A story I'd already read yesterday would resurface as "latest" just because another outlet had rehashed it. The AI wasn't lazy—it was just working with a garbage pile. Web searches pull in official announcements, media reports, second-hand analysis, reposts, and clickbait, all dressed up as "news." The AI could summarize fast, but it couldn't judge what was actually worth my attention.

To cut through the echo chamber, I started with the sources. Over the years, I'd curated 161 RSS feeds. But quantity doesn't equal quality. To stay sane in this information black hole, I needed a clear filtering system.

Action One: Build a Tiered Source Pool by Credibility

In an age where information gets endlessly repackaged and AI-generated content floods the web, the core principle is traceability. Every retelling strips away context and can distort the original meaning. So I focus on the original producers, ranked by trustworthiness:

• Primary sources: Official blogs from OpenAI, Google DeepMind, Anthropic—these give me the raw, unmodified facts.
• Authoritative media: Bloomberg, The Information, Business Insider, WSJ, Reuters, Caixin—they have solid reporting and cross-verification, offering a macro view.
• Quality secondary sources: The Verge, Techmeme, TechCrunch, MacRumors—they translate dense info into readable deep dives.
• Bloggers and KOLs: Tech influencers on Weibo and Bilibili—useful for hands-on impressions and unique viewpoints.

(Full disclosure: I also subscribe to ifanr and APPSO—they're excellent sources.)

Action Two: Organize the Chaos with a Tree Structure

When you have over a hundred feeds, a single list is a mess. You need a tool to hold them all. I use Folo, an RSS reader. RSS might sound old-school in the age of algorithmic feeds, but nothing beats it for active aggregation and control. With Folo, I can organize my subscriptions into categories like a personal magazine. I've got six main sections: tech, gaming, culture, AI, and autos, each packed with the best sources in that niche. Opening Folo feels like flipping through a magazine I've curated myself.

Action Three: Give the Agent a Brain with Folo CLI

What really sets Folo apart for me is its CLI tool. It turns my 161 feeds into a library my AI agent can directly tap into. After setup, the agent reads unread items from my subscriptions—not random web scraps, but a pre-filtered list. Plus, every item comes with a direct link, so the AI can't hallucinate sources.

I upgraded my morning briefing prompt:

• Merge duplicate stories about the same event.
• Cross-verify details across sources and produce one deep, comprehensive brief.
• Prioritize official announcements or primary reports, and mark them as "multi-source verified."

Step Two: A Good Assistant Is Trained by Criticism

Once the information pool was solid, a new problem surfaced: industry hot topics aren't the same as my personal interests. For a while, open-source AI models dominated the news. Day one, I clicked. Day two, I saw another parameter breakdown. By day three, I knew it wouldn't affect my day. What I actually cared about were specific hardware changes—a laptop's specs leaking, a phone's release date.

Humans naturally scroll past what doesn't interest them. Agents don't. They just see that a topic is trending and sources are writing about it. So I told my agent directly: "Too much AI news today. I want more consumer electronics and hardware. Remember that for future briefings." The agent created a MEMORY.md file in the background and stored that preference as a long-term rule. It worked—the next briefing was all about hardware, and I could see in its reasoning process that it was actively filtering for my preferences.

That's my favorite thing about AI agents: they don't read your mind, but they do remember what you dislike and what you care about. A good assistant is often trained by being scolded.

Step Three: Stitch the Fragments Into a Personal Cyber-Paper

Even with good summaries, the chat interface felt clunky—text piled together, hard to scan. Since AI can write code, why not have it turn the scattered threads into a custom HTML newspaper? I upgraded the prompt again:

• Take the five best briefs and format them as a single HTML page.
• Minimalist UI, card-based layout.
• Each card: title, core facts (multi-source), why it matters.
• Links as buttons at the bottom for easy source-checking.

The result was clean, white cards with verified facts and commentary, and I could jump to the original article with one click. It felt like reading a newspaper designed just for me.

I pushed the method further with a long-running topic: the foldable iPhone. Rumors, denials, more rumors—each one looked like a headline, but together they told a repetitive story. I challenged the agent to synthesize two years of leaks into a living, trackable document. The requirements were strict:

• A self-contained HTML page, dark theme, mobile-friendly.
• Top: a 100-word current status summary.
• A tree diagram of specs: size, hinge, screen, price, release date, and unconfirmed details.
• A timeline showing when each rumor first appeared, latest changes, and current status.
• A keyword frequency chart based on independent sources, not reposts.
• Each clue card tagged with credibility: confirmed, multi-source, single rumor, or unverifiable.
• Links to original sources for every claim.
• Filter by category, credibility, and status.

The agent delivered. The page broke down two years of noise into structured data. The status summary cut to the chase—production status and the core mystery. The parameter tree laid out screen ratios, liquid metal hinges, and pricing at a glance. The timeline showed how a rumor evolved and gained consensus. And every card was tagged with its evidence level. For a hot topic like "no crease," I could see exactly how many independent sources backed it, with buttons linking to Ming-Chi Kuo or Bloomberg.

That kind of multi-dimensional cross-referencing saved me hours of tab-hopping. It handed me a coherent investigation report on a silver platter.

After viewing the foldable iPhone tracker, I felt less anxious about the launch date. Not because I'd lost interest—rumors can't replace hands-on experience. But when you can see which leads are supply-chain consensus and which are clickbait, the fear of missing out dissolves.

That relief comes from filtering out low-quality information. In 2025, Merriam-Webster chose "slop" as its word of the year—a term for AI-generated, low-quality digital content. As information multiplies and looks more polished, judgment gets harder.

Building a reliable source pool is the best defense against AI slop. That's the whole point of this system: AI can collect, dedupe, and organize, but it can't outsource our judgment.

The flood of information isn't going to stop. Instead of trying to swim faster, build a dam upstream—subscribe to sources you trust, keep diverse voices, and go back to the original when you see a claim. Whether it's manually curating feeds or training an agent to follow your preferences, what we're doing is setting up a gate at the top of the stream.

What reaches you is already filtered. What's worth a deep read, what's a skim—it becomes obvious. And being able to extract genuine signal from the noise, that alone makes it worth it.