How I Built an AI Editor to Cut Through the Jiu-Jitsu News Firehose

The Morning Scroll That Goes Nowhere

Every morning, I do the same thing: roll over, grab my phone, and see what happened in the jiu-jitsu world overnight. There are dozens of unread messages from my teammates, a fresh batch of technique videos from the usual suspects, and a long list of “breaking” updates from every podcast, news site, and subreddit I follow.

After fifteen minutes of scrolling, a weird feeling settles in. I know a lot of things happened, but I can’t name what actually matters today. A new gi dropped. A leglock system got updated. Some gym announced a new seminar. Each item seems worth a click, but when I click through them all, I’m left with very little.

We used to worry about not having enough information. Now the problem is that there’s too much, and it’s eating not just my time but my ability to decide what deserves attention. So I decided to run an experiment: build an AI agent to curate my own jiu-jitsu news feed and take back control of “what I look at every day.”

Step One: Ditching the Random Scraper for a Trusted Source Network

My first idea was simple: set up a scheduled task in an agent to collect five jiu-jitsu news items every morning and send them to me. That gave me my first briefing assistant.

The results were decent. Before, I’d have to hop between news sites, Instagram, and YouTube to piece together what happened in the jiu-jitsu world overnight. Now I could have a summary waiting while I sipped my coffee.

But the novelty wore off in a few days. The same event—say, a new rule change for an upcoming championship—would get three different write-ups from three different outlets, all taking up space in my briefing. Yesterday’s news would get rehashed today by another site, labeled as “latest,” and pop up again.

I couldn’t blame the agent. It was pulling from a giant pile of “loose information”: official announcements, media reports, secondary commentary, reposts, and clickbait all looked like “new news.” Handing that pile to AI just gives you a fast summarizer, not a good judge of what’s worth reading.

To cut down the echo, I had to start with the sources. Over the years, I’ve built a list of 161 RSS feeds. But quantity doesn’t mean quality. To stay sharp in this information swamp, I needed a clear filtering and management system.

Action One: Build a Tiered Source Pool by Credibility

In an age of recycled content and AI-generated fluff, the core principle is “get to the source.” Information loses context and meaning when it’s passed around. You have to get as close to the original node as possible to see the full picture.

So I sorted my sources by trust level:

• First-hand sources: Official announcements from major jiu-jitsu organizations (IBJJF, ADCC), athlete blogs, and gym press releases.
• Established media: Major outlets like FloGrappling, Jiu-Jitsu Times, and the like—they have solid editorial standards and cross-check their facts.
• Quality secondary sources and aggregators: Sites like BJJEE, fan forums, and technique-focused blogs that turn raw info into readable content.
• Bloggers and influencers: Top competitors, coaches, and YouTubers—they offer real-world experience and personal takes, but they’re supplementary.

Action Two: Organize Your Feeds with a Tree Structure

When you have over a hundred sources, management becomes a chore. If you shove them all into one list, you’ll go nuts trying to find anything. You need a tool that can hold them in order.

I use a tool called Folo. It’s basically an RSS reader. RSS might sound old-school in the age of algorithmic feeds, but for actively pulling and controlling information, there’s still nothing better. With Folo, you can organize your subscriptions into categories like a custom magazine. My feeds are split into six core sections: technique, competition news, gear, business, culture, and science.

Opening Folo is like flipping through a jiu-jitsu magazine I’ve built myself.

Action Three: Give Your Agent an API to Your Source Pool

What matters more than All-in-One and categorization is that Folo is agent-friendly. It offers a CLI tool that turns those 161 sources into a database your agent can directly call.

After setting it up, my agent can read the unread items in my Folo subscriptions. Instead of grabbing random headlines from the whole web, it’s now reading a batch of sources I’ve already vetted. Plus, each item comes with a direct link, so the LLM can’t make up links.

I upgraded my morning briefing:

• Merge duplicates: Identify multiple sources covering the same event and don’t repeat them.
• Cross-verify: Compare details from different sources and combine them into one deep, detailed summary.
• Source priority: If there’s an official announcement or a firsthand report, use that wording and note “multi-source verified.”

Step Two: A Good Assistant Is “Trained” by Your Complaints

Once the info pool was solid, a new problem surfaced: what’s trending isn’t necessarily what you care about.

A while back, the big story was a new guard pass system that everyone was talking about. Day one, I clicked. Day two, I saw the same breakdown from another angle. Day three, I knew it didn’t change my game. What I actually cared about were specific gear releases—a new lightweight gi, a new rash guard design, or a new heel hook instructional.

Humans naturally swipe past what doesn’t interest them. Agents don’t. They only know the topic is hot and the sources keep writing about it. So I told my agent directly: “Too much guard pass stuff today. I want more about new gear and equipment. Remember this for future updates.”

The agent created a MEMORY.md file in the background and added that preference to its long-term rules. It actually remembered. In its reasoning process, you could see it filtering against that preference.

The next briefing was full of gear news—no more guard pass explainers. That’s what I love about agents: they don’t magically know you, but they can remember what you don’t like and what you care about. A good assistant is often “trained” by your complaints.

Step Three: Sewing the Fragments into a “Cyber Newspaper”

Even with a great assistant, the chat interface felt lacking. Text was cramped. Reading experience wasn’t great.

Since AI can write code, why not have it turn scattered clues into a personalized “cyber newspaper”? I upgraded again:

“Take the final five deep summaries and format them into an HTML page: minimalist UI, card layout. Title, core facts (multi-source), why it matters. Links as buttons at the bottom for easy sourcing.”

Now I get a clean view. No more cramped text. White cards neatly arrange verified facts and commentary. If something catches my eye, I click a button to jump to the original.

The same method works for tracking a long-running topic. Take the “foldable iPhone” of jiu-jitsu—maybe the upcoming rule changes for IBJJF competitions. Rumors, denials, more rumors. Each one looks like a big deal, but together they’re often the same issue rehashed.

I gave my agent a tougher task: combine all the rumors about a potential rule change into a “dynamic wiki” that tracks the evolution.

I asked for a single-page HTML file with:

• A 100-word summary at the top about the current status.
• A tree diagram showing the topic as the root, organized by time, specifics, and unresolved questions.
• A timeline of key rumors, their first appearance, latest changes, and current status.
• A keyword frequency chart counting independent sources, not reposts.
• Clue cards with specific claims, credibility, status, earliest date, and original source links.
• Filters by category, credibility, and status.
• A source list.

Rules: prioritize official statements, supply chain announcements, analyst reports, and original media. Merge duplicate reposts. Clearly distinguish “confirmed,” “multi-source corroborated,” “single rumor,” and “unverifiable.” Never present unconfirmed info as fact. Each credibility rating must trace back to a source.

The agent produced a dark-themed, self-contained HTML page with embedded CSS and JavaScript. It broke down two years of rumors into structured data. The top had a 100-word summary with no fluff—just production progress and key questions. Then a tree diagram showed everything from the specific rule text to the potential impact. The timeline and keyword frequency chart showed how a rumor evolved and reached consensus.

Each clue card was labeled “confirmed,” “multi-source,” or “single rumor.” For hot topics like “no-gi vs. gi,” you could see exactly how many independent sources backed it, and clicking a button would take you to the original report.

This multi-dimensional cross-referencing saves you the effort of piecing together the truth from dozens of pages. It hands you a logically structured “investigation report.”

After seeing that tracking page, I felt less anxious about missing the big news. Not because I lost interest—rumors can’t replace actual experience. But when you can see which rumors are consensus and which are just clickbait, the fear of missing out fades.

Why This Matters: Fighting the “Slop”

In 2025, Merriam-Webster picked “slop” as the word of the year. It used to mean garbage, but now it describes low-quality, AI-generated digital content. Information is growing, packaging looks more polished, and judgment gets harder.

In this flood of digital slop, building your own reliable source network is the best defense against AI junk. That’s the whole point of this system: AI can collect, dedupe, and organize, but it can’t outsource your judgment.

Information will keep pouring in. Instead of trying to swim faster, build a small dam upstream. Subscribe to sources you truly trust. Keep different voices. When you see a conclusion, go back and read the original.

Whether it’s handpicking sources or letting an agent work by your preferences, what we’re doing is putting a filter at the top of the stream. What flows through is already settled. Which pieces deserve a deep read and which deserve a quick glance becomes clear.

Being able to extract what’s truly useful from the noise—that itself is enough.