Four hours. Every week. One curated news article.
That was my life before Paxelo. And it’s the reason this product exists.
The news curation feature wasn’t an add-on I thought of later. It was one of the primary reasons I started this whole experiment. Not a hypothetical customer need I read about in a market report. A problem I was living through every single week.
Watch the full episode: YouTube. Day 15
What the Manual Process Looked Like
In my day job, I publish industry news content regularly. Curated articles that pull from multiple sources, synthesize the key developments, and deliver them in a readable, structured format. I was doing that process entirely by hand.
Here’s what the time actually broke down to. Research alone: an hour per article, sometimes longer. Draft writing, source gathering, fact-checking: another two to three hours. WordPress formatting, SEO tagging, post creation: one more hour. Four hours of work for one curated news article. Every week.
My setup was a tear folder. A curated link library of RSS feeds, industry news sources, and publication pages I’d built up over time. I’d go through it by hand. Click. Read. Decide what’s worth including. Copy. Paste. Synthesize. Format. Publish.
It worked. But it was the kind of process that works right up until you run out of time. Which, when you’re also running a VP-level day job, is constantly.
Why This Was the Perfect Candidate for Automation
The whole point of this experiment with AI was to identify exactly this kind of problem. Something I was doing manually, repetitively, and time-intensively. And automate it.
News curation was the perfect candidate. Repetitive? Every week. Time-intensive? Four hours per article. Producing real value? Yes, published content that drives traffic and engagement. The trifecta.
Briefing the AI Like a New Hire
I described it to Abacus the way I’d describe it to a new hire on their first day. I didn’t prescribe the method. I described the outcome.
Here’s what I want: a news feed curation service with full automation. Pull from three major news aggregation services through NewsAPI. Deduplicate the results so the same story doesn’t appear from multiple sources. Package the final output as a clean, publishable article.
I also asked for a fallback search mechanism. Because some niche industry topics don’t have enough primary source coverage, and the service needs to still deliver something useful when the main feeds come up thin.
This is the same principle I’ve used for thirty years managing people. Tell them what done looks like. Let them figure out how to get there. Works the same with AI as it does with a new marketing coordinator.
The Three-Layer Build
The architecture is straightforward once you see it:
Layer one: NewsAPI pulls from three major aggregation services simultaneously. Broad coverage, multiple perspectives on the same story.
Layer two: Deduplication logic. Same story from three different sources gets filtered to one clean output. No repeated headlines, no redundant coverage.
Layer three: Fallback search. For niche topics where primary source coverage runs thin, the service goes looking instead of coming back empty. It still delivers.
The output is a clean, structured, publishable article. Ready to go.
What Went Live
Fourteen builds over about a week. The feature set didn’t just ship. It shipped complete.
Sales page. Tutorial. Resources page. Hero image. Marketing article. Documentation. All of it built, all of it live. The news curation feature went from concept to finished product in the time it used to take me to produce three articles by hand.
The Principle Worth Stating Plainly
If you are already doing something by hand. Something you do regularly, something that takes significant time, something that produces value for you or your organization. That is your next feature.
Not a hypothetical customer need. Not a market gap you read about. Your own genuine, recurring problem.
The best product features come from the friction you personally experience. Because then you know exactly what the solution needs to feel like. You know which steps are waste. You know what the output should look like. You know because you’ve been the user.
Four hours of manual work per article. Now: a prompt, a wait, a clean deliverable.
That’s the experiment working.
I’m Stephen Sowinski. I’m 65. I’m building Paxelo in public, one episode at a time.
Watch Day 15 on YouTube | Follow the full Build Series