How I Built My Own AI News Editor with 161 RSS Feeds

Morning Dread in the Feed

Every morning, I do the same thing: grab my phone and see what happened overnight. WeChat pings, Weibo trends reshuffle, news apps push alerts, and tech forums churn out 'breaking' updates. After ten or fifteen minutes of scrolling, a weird feeling creeps in. I've seen a lot, but I can't tell you what actually matters today.

A phone launches. An AI model updates. Another company announces a plan. Every item looks click-worthy, but after reading them all, little sticks. We used to worry about not having enough information. Now the problem is the opposite: there's so much that it swallows not just time but also judgment about what's worth our attention.

So I ran an experiment. I built a system using AI agents to filter my news and take back control of what I read each day.

Stage One: Ditch the Scattershot, Tap a Pro Wire Service

My first idea was simple: set a daily 8 a.m. task in an agent to grab five tech stories and send them to me. That gave birth to my first briefing assistant.

It worked—sort of. Instead of hopping between news sites, apps, and social feeds to piece together overnight tech news, I'd have a neat summary waiting while I made coffee.

But the novelty wore off in days. The same model launch, rewritten three ways, took up three slots. A story I'd seen yesterday got rehashed by another outlet, and it popped back as 'new.'

To be fair, the agent wasn't slacking. Web search returns a sack of loose info: press releases, articles, secondhand takes, reposts, clickbait—all looking fresh. The AI can summarize fast, but it can't judge what really matters to you.

The fix starts with sources. Over the years, I've curated 161 RSS feeds. But quantity isn't quality. To stay sharp in this info swamp, I needed a clear filtering strategy.

Action 1: Build a Trust Ladder

In an age of recycled content and AI-generated fluff, the core principle is traceability. Information loses context and gets distorted through repeated retelling. The closer you get to the original node, the more accurate the picture.

So I sort sources by trust level:

• Primary sources: Official company blogs (OpenAI, Google DeepMind, Anthropic) for unvarnished announcements.
• Authoritative media: Bloomberg, The Information, WSJ, Reuters, Caixin—they have solid reporting and fact-checking.
• Quality aggregators and niche outlets: The Verge, Techmeme, TechCrunch, MacRumors—they turn raw info into readable analysis.
• Bloggers and KOLs: Digital gadget influencers on Weibo, Bilibili tech YouTubers—good for hands-on impressions and spicy opinions.

If you're not picky, you can add ifanr and APPSO—they're solid too.

Action 2: Organize with a Tree Structure

When you've got over a hundred sources, a flat list is chaos. You need a tool that can hold them all—I use Folo.

It's an RSS reader, which sounds old-school in the age of algorithmic feeds. But for actively aggregating and staying in control, nothing beats it. You can organize subscriptions into folders like a custom magazine. My Folo has six core sections: tech, gaming, culture, AI, cars—from Bloomberg to Ars Technica to DIGITIMES. Opening Folo feels like flipping through a magazine I've curated myself.

Action 3: Open the API, Make Folo the Agent's Brain

What makes Folo special is its CLI tool. It turns those 161 feeds into an information library the agent can call directly.

Once set up, the agent reads unread items from my subscriptions. It's not scraping random headlines; it's pulling from my pre-filtered sources. Plus, each item has a direct link, so no more AI hallucinated URLs.

My upgraded morning briefing prompt looks like this: 'Use Folo. Read the skill docs and follow instructions. Read unread items from the past 24 hours. Apply editor judgment: merge duplicate stories, cross-verify details, prioritize primary sources, and mark multi-source confirmations.'

Stage Two: Train Your Assistant by Complaining

Even with a quality pool, a new problem appears: industry hot topics aren't the same as your personal interests.

For a while, open-source AI models dominated the news. Day one, I click. Day two, parameter breakdowns. Day three, I know it doesn't affect my life. What I really care about are concrete hardware changes—like a new laptop's specs or a phone's release date.

Humans filter automatically. Agents don't. They see a hot topic and keep pushing it. So I told my agent directly: 'Too much AI news today. I want more consumer electronics and hardware. Note that for future briefings.'

The agent created a MEMORY.md file and added a long-term rule. And it stuck. My next briefing was full of hardware news, not AI models. That's what I love about agents: they don't magically know you, but they remember what you don't like. A good assistant is often born from complaints.

Stage Three: Stitch the Fragments Into a 'Cyber Newspaper'

Chat bubbles are fine, but reading long summaries in a chat window is meh. Since AI can code, why not have it turn the scattered clues into a proper digital newspaper?

So I upgraded the prompt: 'Take the five final briefs and make an HTML page. Minimalist UI, card layout. Title, key facts (multi-source), why it matters. Put source buttons at the bottom.'

The result: clean white cards with verified facts and commentary, no cramped text. Click a button to jump to the original article.

The same method works for long-running stories. Take the foldable iPhone rumors. They go back and forth: leaks, denials, more leaks. Each one looks like a big deal, but together they're just variations on the same question.

I challenged my agent to build a 'dynamic encyclopedia' that tracks the foldable iPhone saga. The prompt was detailed: create a self-contained HTML file with a 100-word status summary, a tree diagram for specs, a timeline of key leaks, a keyword frequency chart (counting independent sources, not reposts), clue cards with credibility labels and source links, filters, and a source list. Rules: prioritize official and primary sources, merge duplicates, distinguish confirmed vs. multi-source vs. single rumor vs. unverifiable, never present speculation as fact, and make every claim traceable. Output only the HTML, no explanation.

The agent delivered. The page has a dark theme, works on mobile and desktop, and uses pure SVG/CSS for charts. At the top, a blunt status summary: production progress and core unknowns. Below, a tree diagram lays out screen ratio, liquid metal hinge, price—all the rumors. The timeline shows how a rumor evolved and reached consensus. Each clue card is tagged 'confirmed,' 'multi-source,' or 'single rumor.' For something like 'no crease,' you can see exactly how many independent sources back it up, and click through to Kuo or Bloomberg.

This cross-referencing saves you from piecing together truth from scattered pages. It hands you a coherent investigation report.

After seeing that tracking page, I'm actually less anxious about when the foldable iPhone will launch. Not that I've lost interest—rumors can't replace hands-on experience. But when you clearly see which rumors are supply-chain consensus and which are just clickbait, the fear of missing out evaporates.

That relief comes from cutting out junk. In 2025, Merriam-Webster chose 'slop' as word of the year—AI-generated, low-quality digital content. The more info there is, the harder it is to judge.

In this flood of digital slop, building your own reliable sources is the best defense against AI garbage. That's why I built this system: AI can collect, dedupe, and organize, but it can't outsource your judgment.

The info flood won't stop. Instead of trying to swim faster, build a dam. Subscribe to sources you trust, keep diverse voices, and when you see a conclusion, go back to the original. Whether it's curating feeds or teaching an agent your preferences, you're just adding a filter upstream. What reaches you is already settled. You can tell what's worth a deep read and what's just a glance.

Being able to extract real value from noise—that's meaningful enough.