Building a Personal News Filter with an AI Editor and 161 RSS Feeds

The Firehose Problem

Every morning, I wake up to the same ritual: checking my phone to see what happened overnight. Dozens of unread messages in WeChat, a refreshed trending list on Weibo, a stream of push notifications from news apps, and then the tech communities—each one claiming some "breaking" update. After fifteen minutes of scrolling, I'm left with a strange feeling. I've seen a lot, but I can't tell you what actually matters today.

A phone releases. An AI model updates. A company announces a new plan. Each item looks worth clicking, but when I read them all, very little sticks. The problem isn't a lack of information anymore. It's the opposite: there's too much, and it's drowning out my judgment about what deserves attention.

So I decided to run an experiment. I built an AI agent to curate my news, a system that filters the noise and hands me back control over what I read each day.

Step 1: Ditch the Random Scraping, Plug into a Professional Network

My first attempt was simple: set a scheduled task in the agent to grab five tech news items every morning at 8 AM. That gave birth to my first briefing assistant. It worked okay. Instead of hopping between news sites, apps, and social feeds to piece together what happened overnight, I had a summary waiting for me by the time my coffee was ready.

But the novelty wore off in a few days. The same model launch, written three different ways, would take up three slots in the briefing. Yesterday's news, re-reported by another outlet, would pop up again wearing a "latest" tag.

I can't blame the agent entirely. A web-wide search returns a grab bag of "loose information"—official announcements, media coverage, secondary commentary, reposts, and clickbait—all looking equally like "new news." Hand that pile to an AI, and it can summarize quickly, but it can't judge what's genuinely worth your time.

To cut down on echoes, I started with the sources. Over the years, I'd accumulated 161 RSS subscriptions. But quantity doesn't mean quality. To stay sane in this information black hole, I needed a clear filtering and management strategy.

Action 1: Build a Tiered Source Pool by Trustworthiness

In an age of recycled content and AI-generated fluff, the core principle is traceability. Information loses crucial context—and sometimes its original meaning—when it's passed around and aggregated. The closer you get to the original node, the clearer the picture.

I sorted my sources by credibility:

• Primary sources (official blogs and earnings reports): OpenAI News, Google DeepMind News, Anthropic News—these give me untainted original info.
• Established media (serious journalism and investigative reports): Bloomberg, The Information, Business Insider, WSJ, Reuters, Caixin—they have strong editorial networks and rigorous cross-checking.
• Quality secondary sources (vertical analysis and aggregators): The Verge, Techmeme, TechCrunch, MacRumors—they turn raw info into readable deep dives.
• Bloggers and KOLs (leaks and expert opinions): tech vloggers on Bilibili, big Weibo accounts. They supplement with hands-on experiences and unique takes.

If you don't mind extra reading, you can also subscribe to ifanr and APPSO—they're solid sources.

Action 2: Organize with a Tree Directory

When you have over a hundred sources, managing them in one flat list is a mess. You need a unified tool. I use Folo, an RSS reader that's been around for ages. In an era of algorithm-driven feeds, RSS feels old-school, but for actively aggregating information and keeping control, nothing beats it.

Folo lets me treat my subscriptions like a custom magazine. I've split mine into six core sections—tech, gaming, culture, AI, autos—covering everything from Bloomberg and Ars Technica to DIGITIMES and MacRumors. Opening Folo feels like flipping through a magazine I've built for myself.

Action 3: Expose an API with Folo CLI

What really sets Folo apart is its CLI tool. It turns those 161 feeds into a library my agent can call directly. Once configured, the agent reads unread items from my subscriptions. It's not scraping random headlines anymore; it's reading a vetted list. Plus, each item comes with a direct link, so no more hallucinated URLs.

Here's the upgrade I made to my morning briefing:

Use Folo
Read https://api.folo.is/skill.md and follow the instructions.
Please read my Folo subscriptions from the last 24 hours. Perform an "editor-in-chief" analysis:
- Merge duplicates: identify multiple sources covering the same event, avoid repetition.
- Cross-verify: compare details across sources, compile the most complete summary.
- Prioritize traceability: if an official announcement or primary report exists, use that wording and note "multi-source verified."

Step 2: A Good Assistant Is Trained by Criticism

Even with a quality information pool, a new problem emerged: industry trends aren't the same as my interests.

For a while, open-source model releases dominated the news. Day one, I clicked. Day two, another parameter breakdown. By day three, I knew it wouldn't affect my day. What I really cared about were concrete hardware changes—like a new laptop's specs leaking or a phone's launch date.

Humans naturally scroll past what doesn't interest them. An agent doesn't. It only knows the topic is hot and sources are writing about it, not that I've had enough. So I told it directly:

"There's too much AI news today. I want more consumer electronics and hardware stories. Remember this for future briefings."

The agent created a MEMORY.md file in the background, turning that preference into a long-term rule. And it worked. The next briefing was full of hardware news, and the AI model stuff was gone. You could even see the reasoning in its thinking process.

That's what I love about agents: they don't magically understand you, but they remember what you don't like. A good assistant is often trained by being scolded.

Step 3: Sew the Fragments into a Cyber Newspaper

The agent could summarize and organize, but the chat interface felt cramped. Text bunched together, not great for reading. Since AI can write code, why not have it turn the scattered clues into a personalized HTML page—my own "cyber newspaper"?

Here's the upgraded prompt:

Turn the final 5 deep-dive briefings into an HTML page.
Minimalist UI, card layout.
Include: headline, core facts (multi-source summary), why it matters.
Buttons at the bottom for source links.

The result: clean white cards, no dense text. Multi-source verified facts and commentary laid out neatly. Click a button to jump to the original article.

I pushed the same approach to track a long-running topic: the foldable iPhone. Reports, denials, more leaks—each one looks like a big story, but together they're just the same question rehashed. I challenged the agent to create a "living encyclopedia" that tracks the whole saga.

I asked for a self-contained HTML page with a 100-word status summary, a tree diagram of specs, a timeline of key leaks, a keyword frequency chart by independent sources, and filterable clue cards. Each card had to have a credibility label—confirmed, multi-source corroborated, single rumor, or unverifiable—and link back to the original source. Dark theme, responsive, no external dependencies.

The agent delivered. It broke two years of scattered rumors into structured data. The top had a concise status summary. A parameter tree showed everything from screen ratio to the liquid metal hinge. Timelines and frequency charts revealed how rumors evolved and converged. Each card was tagged with its credibility, and you could click through to the original report from analysts like Ming-Chi Kuo or Bloomberg.

That kind of cross-referencing saves you from digging through dozens of pages. It hands you a logically organized investigation report.

After seeing that foldable iPhone tracker, I'm less anxious about when it'll actually launch. Not that I've lost interest—rumors can't replace hands-on experience. But when you can see which clues come from supply chain consensus and which are just clickbait, the fear of missing out fades.

That anxiety disappears because you've filtered out the junk. In 2025, Merriam-Webster chose "slop" as its word of the year—a term for low-quality, AI-generated content. Information keeps growing, packaging gets slicker, and judgment gets harder.

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

The information will keep pouring down. Instead of trying to swim faster, build a small dam upstream. Subscribe to sources you trust, keep diverse voices, and always go back to the original when you see a claim. Whether it's hand-picking feeds or training an agent to follow your preferences, what we're really doing is installing a gate at the top of the stream.

What reaches you is already settled. You can easily tell what deserves a close read and what deserves a skip. Extracting something genuinely useful from the noise—that's enough.