Every creator hits the same wall. You can build anything, but you don't know what to build.
At 5 AM on a Wednesday, staring at a blank screen, I had that exact thought. AI makes building trivial. The bottleneck shifted from "how" to "what." And nobody was solving the "what" problem.
So I built the system that scans the internet for it.
That system powers the personal AI agent infrastructure I use daily. For the business case behind this approach, see Why AI Money Printers Fail. Real results live in the case studies.
The Problem Nobody Talks About
There are thousands of unsolved problems hiding in plain sight. People complain about them publicly, every single day, on Reddit, Hacker News, Stack Overflow, forums, YouTube comments. They say things like "I wish there was an app for this" or "why does nobody build a tool that does X."
Nobody's connecting the dots.
The complaints are scattered across platforms, buried in threads, mixed in with noise. You'd have to read thousands of posts a day to find the real patterns. No human can do that.
But an agent can.
What I Built
I built a system of autonomous agents that scan 7 sources across the internet, 24/7. No input required. No search queries.
They just listen.$ gapfeed status
# Sources
"reddit" | "hacker news" | "duckduckgo" | "rss" | "stackoverflow" | "dev.to" | "youtube"
# Stats
"signals_scanned": 6279,
"gaps_found": 53,
"uptime": "24/7",
"intelligence_index": 41.5 # mimo-v2-flash, top 12%
Here's what each agent does:
- Scout (Alex) - Scrapes Reddit, Hacker News, DuckDuckGo, RSS feeds, Stack Overflow, Dev.to, and YouTube in parallel. 3-second sleep between rounds. ~1,900 raw signals per minute.
- Classifier (Maya) - Takes raw signals and runs them through an LLM.Classifies sentiment, buyer intent, urgency, and pain category. Detects gaps every 3 cycles by clustering related signals.
- Monitor (Sam) - Health checks every 5 minutes. Sends me a Telegram alert if anything breaks.
- Browser Scout (Jack) - Same sources as Alex but uses a real browser with stealth mode.Gets past anti-scraping protections.
The models powering this aren't a compromise, they're a deliberate choice. The classifier uses Xiaomi's MiMo-V2-Flash, which scores 41.5 on the Intelligence Index (better than 88% of models) at $0.09 per million input tokens. That's 3.5% of Claude Sonnet 4.5's cost while delivering comparable performance on SWE-bench.
For reasoning-heavy work, OpenAI's gpt-oss-120b runs at $0.039 per million tokens and ranks #3 in Legal and #2 in Marketing benchmarks. The fallback is Google's Gemini 2.5 Flash at $0.30 per million.I didn't pick these models because they're cheap. I picked them because they punch way above their weight class. MiMo beats Gemini 2.5 Flash's Intelligence Index (41.5 vs 20.6) at a fraction of the cost.
The entire system scans 6,000+ signals using models that cost less per day than a coffee.What the Data Shows
After scanning 6,279 signals and clustering them into 53 distinct gaps, some patterns emerged that surprised me.
1. The Best Opportunities Are Boring
The most painful gaps aren't the flashy ones. Nobody's complaining about wanting a cooler social media app.
They're complaining about internal tooling, workflow automation, and data management. The stuff nobody wants to build because it's not exciting.That's exactly why the gaps exist.
2. People Pay to Fix Frustration, Not Features
The signals with the highest buyer intent aren't from people asking for features. They're from people who are frustrated.
"I've been trying to solve X for 6 months" has a much higher conversion signal than "it would be cool if someone built Y."3. The Same Problem Shows Up Everywhere
One gap I found had signals from Reddit, Hacker News, Stack Overflow, and YouTube comments, all from different people, all saying the same thing differently. No single person could see the pattern.
But the system caught it because it watches everything at once.What Comes Next
I'm building GapFeed to make this data public.
A feed of unsolved problems, clustered by category, scored by demand, with evidence from real people complaining in real threads. Not search. Not social listening.
A gap discovery feed.Every gap gets a permanent page, indexable by search engines. You'll be able to browse gaps by category, source, urgency, and buyer intent. See the actual quotes from people who want the problem solved.
Free. No login. No paywall.
The Real Insight
The internet is full of unsolved problems. The signal-to-noise ratio is terrible for humans, but trivial for systems. The opportunity isn't in finding a problem nobody's ever thought of.
It's in seeing the pattern across thousands of people who've all thought of it separately.I built the system that sees the pattern. Now I'm opening it up.
| Operational layer | Fragile default | Production architecture |
|---|---|---|
| Process design | Automating a broken manual habit as-is | Mapping the data contract and failure modes before writing code |
| Human oversight | Either 100% manual or unguarded autopilot | Autonomous execution on routine cases with approval gates on exceptions |
| Data custody | Operations locked inside third-party SaaS | Code, prompts, logs, and API keys owned inside your accounts |
| Failure recovery | Silent drops when an upstream API changes | Schema validation, retries, and operator alerts on every break |
FAQ
How do I find unsolved problems worth building?
Watch where people complain in public, on Reddit, Hacker News, Stack Overflow and YouTube comments. My system scans seven sources around the clock and clusters related signals into gaps.
What makes a gap worth chasing?
The most painful gaps are boring ones like internal tooling and workflow automation, and the signals with the highest buyer intent come from frustrated people. Someone who has been trying to solve a problem for six months shows a much higher conversion signal than someone asking for a feature.
Which models power an internet scanning agent?
The classifier runs on Xiaomi's MiMo-V2-Flash at $0.09 per million input tokens, and reasoning-heavy work runs on OpenAI's gpt-oss-120b at $0.039 per million. Gemini 2.5 Flash is the fallback.
What is GapFeed?
GapFeed is a public feed of unsolved problems clustered by category and scored by demand, with quotes from real threads. Every gap gets a permanent page that search engines can index, free with no login.