How I Built an AI-Powered Art News Curator from 161 RSS Feeds

The Morning Scroll That Went Nowhere

Every morning, I do the same thing: grab my phone and see what happened overnight. There are dozens of unread messages on WeChat, the trending list on Weibo has rotated several times, and my news app pushes a long queue of 'important updates.' I open a few art forums and see 'breaking' and 'major announcement' everywhere.

After ten minutes of scrolling, I feel a strange emptiness. I've absorbed a lot, but I can't tell you what actually matters today. A new gallery opens, an artist releases a new series, a museum announces a partnership. Each item seems worth reading, but when I finish, little sticks.

We used to worry about not having enough information. Now the problem is the opposite: too much, and it's swallowing my time and my judgment about what's truly worth my attention. So I decided to run an experiment—build a personal curation system using AI agents, and take back control of what I see each day.

Step One: Ditch the Net, Build a Ladder of Trust

My first attempt was simple: a scheduled task every morning at 8 a.m. that would gather five art news items and send them to me. That became my first briefing assistant. The result was decent—I no longer had to hop between apps and websites to piece together what happened in the art world overnight. While my coffee brewed, a tidy summary was already waiting.

But the novelty wore off in a few days. The same press release about a new exhibition would appear three times, rewritten slightly by different outlets. Yesterday's news would resurface as 'latest' because another site had reposted it. The agent wasn't lazy—it was just feeding on an undifferentiated pile of 'loose information': official announcements, media reports, secondary interpretations, reposts, and clickbait. All of it looks like a 'new story.'

To cut through the echo, I had to start at the source. Over the years, I've accumulated 161 RSS feeds. But quantity doesn't equal quality. To stay clear-headed in this information black hole, I needed a clear filtering system.

Action One: Build a Tiered Source Pool by Credibility

In an age of recycled content and AI-generated noise, the core principle is 'traceability.' Information loses critical context with every retelling. So I sort my sources by how close they are to the original event:

• Primary sources: Official museum press rooms, artist foundations, auction house announcements—first-hand, unaltered info.
• Authoritative media: Artforum, The Art Newspaper, Hyperallergic—they have solid editorial standards and cross-checking.
• Quality secondary sources: Artsy, Artnet News, and niche blogs that turn raw data into readable stories.
• Bloggers and influencers: Independent critics and Instagram art sleuths—good for street-level observations and personal takes.

If you don't mind, I'd add a couple more—like the newsletters from museum curators. They're excellent.

Action Two: Organize with a Tree Directory

When you have over a hundred feeds, managing them is a chore. Stuffed into one list, you'll drown. I use a tool called Folo—it's an RSS reader, and it lets me organize subscriptions into folders like a custom magazine. My Folo has six main sections: Painting, Sculpture, Photography, Digital Art, Museums, and Galleries. Opening it feels like flipping through a magazine I've slowly built for myself.

Action Three: Give the Agent a 'Second Brain' via CLI

What sets Folo apart is that it's agent-friendly. It offers a CLI that turns my 161 feeds into a searchable database my AI can call. Now the agent reads only from my curated list, not random web scraps. Each item comes with a direct link, so no more hallucinated citations.

Step Two: A Good Assistant Gets 'Scolded' Into Shape

Even with a quality pool, there's another problem: industry hot topics aren't necessarily my interests. For a while, every news blast was about a new AI art generator. Day one, I clicked. Day two, another breakdown of the same model. By day three, I knew it wouldn't change my day. What I cared about were the concrete hardware details—like a new tablet for drawing or the next camera sensor.

So I told my agent directly: 'Too much AI news today. I want more about painting techniques and new pigment releases. Remember that for future updates.'

The agent created a MEMORY.md file, logging my preference into its long-term rules. And it worked. The next morning, the feed was full of painting-related news—new oil mediums, brush reviews, and pigment science. That's why I love agents: they don't intuit your taste, but they remember what you dislike and what you care about. A good assistant is often 'scolded' into shape.

Step Three: Stitch Fragments Into a 'Cyber Newspaper'

But chat windows are ugly. So I asked the agent to turn the final five deep-dives into an HTML page—a minimal, card-based layout with a headline, core facts (multi-source verified), and a 'why it matters' note, plus buttons linking to original sources. Clean white cards, no clutter. Click through to the full story if it interests you.

I use the same method for long-running stories. Take the 'foldable iPhone' of the art world—say, the upcoming renovation of a major museum. Rumors, denials, leaks—it's a loop. So I gave the agent a bigger challenge: create a self-contained HTML page that tracks the 'foldable iPhone' rumors as a living document.

The page had to include a 100-word status summary, a tree diagram of all claims, a timeline of key leaks, a keyword frequency chart based on independent sources (not reposts), and cards for each clue with credibility tags like 'Confirmed,' 'Multiple Sources,' 'Single Rumor,' and 'Unverified.' It also needed filters by category, credibility, and status, plus a source list. Dark theme, mobile-friendly, no external dependencies.

The result was stunning. Two years of scattered rumors were dissected into structured data. The top summary cut straight to the point: production status and the core mystery. Then a parameter tree showed every claim—screen ratio, hinge material, price—all in one view. The timeline and frequency chart revealed how a rumor evolved and where consensus formed. Each card was tagged with its credibility, and clicking through led to the original analyst report or media piece.

That level of cross-referencing saves you hours of piecing together truth. It hands you a logical investigation report.

Regaining Calm in a Sea of Slop

After seeing that tracking page, I felt less anxious about the actual release date. Not because I lost interest—a rumor chart can't replace holding the real thing. But when you see which claims are backed by supply chain consensus and which are just clickbait, the fear of missing out disappears.

In 2025, Merriam-Webster chose 'slop' as its word of the year. It used to mean 'swill,' but now it describes the low-quality, AI-generated content flooding our feeds. The more information there is, the harder it is to judge.

Building your own reliable source list 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 your judgment. The flood of information will keep coming. Instead of swimming faster, build a small dam upstream. Subscribe to sources you trust, keep diverse voices, and when you see a conclusion, go back to the original.

Whether it's manually filtering sources or letting an agent follow your preferences, what we're doing is installing a gate at the top of the stream. What flows through is already settled. You can tell what deserves a deep read and what deserves a glance. And extracting something useful from the noise—that's meaningful enough.