How I Built an AI News Editor to Cut Through the Insurance Information Flood

My Morning Routine Was a Black Hole

Every morning, I did the same thing: grabbed my phone and checked what happened overnight. Dozens of unread messages in WeChat, a rotating cast of topics on Weibo hot search, a string of “important alerts” from news apps, and then the tech forums with their “just launched” and “major update” posts.

After ten minutes of scrolling, a weird feeling settled in. I knew a lot of stuff, but I couldn’t tell you what actually mattered today.

This isn’t just a tech problem. It’s an insurance problem. Policy updates, regulatory shifts, actuarial tables, reinsurance deals, climate risk models — each one looks worth clicking. But if you read them all, you’re left with almost nothing. We used to worry about not having enough information. Now we’re drowning in it, and the flood is stealing our judgment, not just our time.

So I ran an experiment: I built an AI agent to curate my own news feed and take back control of “what to read today.”

Step One: Stop Scraping the Bottom, Plug Into a Real Intelligence Network

My first attempt was simple. I set a daily 8 a.m. task in an agent: collect five insurance news items and send them to me. That gave me my first briefing assistant.

The results were okay. Before, I’d jump between news sites, apps, and social feeds to piece together what happened in the insurance world overnight. Now, while I made coffee, a briefing would be waiting on my phone.

But the novelty wore off in days. The same regulatory ruling, rewritten three different ways, would take up three slots. A story I’d already seen yesterday would resurface as “latest” because another outlet had rehashed it.

To be fair, the agent wasn’t slacking. A web-wide search just returns a bag of “loose info”: press releases, news reports, secondhand analysis, reposts, and clickbait, all looking like fresh news. Feed that to an AI and it can summarize fast, but it can’t judge what’s actually worth your attention.

The fix starts with the sources. Over the years, I’ve collected 161 RSS feeds. But quantity isn’t quality. To stay sharp in this information flood, I needed a clear filtering system.

Action One: Build a Tiered Source Pool by Credibility

In an age where information gets recycled and AI-generated content is everywhere, the core principle is “traceability.” Every retelling and aggregation loses context, sometimes distorting the original meaning. Only by getting as close to the original node as possible can you see the true, accurate picture.

So when picking sources, I focus on the ones that actually produce content, ranked by credibility:

• Primary sources — official announcements, earnings reports, regulatory filings. For insurance, that means state insurance departments, NAIC bulletins, company press releases, and SEC filings.
• Authoritative media — Reuters, Bloomberg, WSJ, The Information, and specialized trade press like Business Insurance or Insurance Journal. They have real reporting and cross-verification.
• Quality secondhand and aggregators — sites like Axios Pro, InsuranceNewsNet, or even Techmeme for tech-driven insurance news. They turn raw info into readable analysis.
• Analysts and experts — actuaries, risk modelers, and consultants who post on LinkedIn or niche forums. They give you real-world perspective.

Step Two: Organize the Flow with a Tree Directory

When you have over a hundred sources, managing them is a chore. If you dump them all into one list, you’ll get lost. You need a tool that lets you organize. I use Folo, an RSS reader.

RSS might sound old-school in the age of algorithmic feeds, but it’s still the best way to actively aggregate and control your information. With Folo, you can flip through a custom magazine of your own making. I’ve split my feeds into six core sections: insurance regulation, P&C, life & health, insurtech, climate risk, and industry analysis.

Step Three: Open the Interface — Folo CLI Becomes the Agent’s External Brain

What matters more than organization is that Folo is agent-friendly. It offers a CLI tool that turns those 161 feeds from a simple reader into a knowledge base the agent can call directly.

Once configured, the agent can read unread items from my Folo subscriptions. It’s no longer grabbing random headlines from the whole web; it’s reading a batch of sources I’ve already vetted. Plus, each item has a direct link, so the AI can’t fabricate sources.

I upgraded my morning briefing with this prompt:

“Use Folo, read the skill doc, and follow the instructions. Read my Folo subscriptions’ unread items from the past 24 hours. Apply editor judgment: merge duplicate coverage, cross-verify details across sources, and prioritize primary sources. Output five deep briefs, each with a title, core facts (multi-source), and why it matters.”

A Good Assistant Is Trained by Your Complaints

Even with a quality pool, a new problem emerged: what’s hot industry-wide isn’t necessarily what I care about. For a while, large language model releases dominated the tech pages. Day one, I’d click. Day two, I’d see parameter breakdowns. Day three, I knew it didn’t affect my day. What I actually wanted were the hardware changes — a new laptop spec leak, a phone release date.

Humans swipe away based on current interest. Agents don’t. They just know a topic is hot and sources are writing about it. So I told the agent directly: “Too much AI news today. I want more consumer electronics and hardware. Remember that for future pushes.”

The agent wrote that preference into a MEMORY.md file. And it worked. The next morning, the AI model stories were gone, replaced by hardware and consumer electronics news.

That’s what I love about agents: they don’t magically understand you, but they remember what you dislike and what you care about. A good assistant is often trained by your complaints.

Stitching Fragments into a “Cyber Newspaper”

Dialog boxes are fine for text, but they’re not great for reading. Since AI can write code, why not turn scattered clues into a personalized “cyber newspaper”?

I upgraded again: “Take the final five deep briefs and format them as a single HTML page: minimal card layout, with a title, core facts (multi-source), why it matters, and links as buttons at the bottom for source tracing.”

Now, instead of dense text, I get clean white cards. Each brief has verified facts and commentary, with buttons to jump to the original source.

Going Deep: Tracking a Long-Running Story

The same method works for long-running topics. For insurance, think about something like the FAA’s drone insurance rules or the ongoing saga of flood insurance reform. Each update looks like big news, but strung together, it’s often the same issue rephrased.

I challenged the agent to track foldable iPhone rumors — an analogous case. It built a self-contained HTML page that tracked every leak, categorized by type, with a timeline, source frequency, and credibility tags. It distinguished “confirmed” from “multiple sources” from “single rumor.”

That same approach applies to insurance. I could ask the agent to track, say, the evolution of parametric insurance for wildfires. It would pull every mention, cross-verify, and present a structured report with a timeline, key claims, and source links. You’d see how a concept moves from a single analyst’s note to a multi-source consensus, and you’d know exactly what’s confirmed and what’s just noise.

Why This Matters for Insurance Professionals

In 2025, Merriam-Webster chose “slop” as its word of the year. It’s used to describe low-quality, AI-generated digital content. The insurance industry isn’t immune. You’ve got AI-generated articles about policy changes that are subtly wrong, or “expert” opinions that are just repackaged press releases.

Building your own reliable source network is the best defense against AI slop. The AI can collect, deduplicate, and organize, but it can’t outsource your judgment.

Information will keep flooding in. Trying to swim faster is a losing game. Instead, build a small dam upstream: subscribe to sources you trust, keep diverse voices, and always check the original when you see a conclusion.

Whether you’re filtering sources by hand or letting an agent follow your preferences, you’re just setting a gate upstream. What reaches you is already filtered. You can tell what deserves a slow read and what deserves a quick scan.

Being able to extract what’s truly useful from the noise — that’s the whole game. And it’s worth playing.