Got laid off. Started building. Three SaaS already failed, more cooking. 馃殌 Current Nicheloom- https://nitter.cf/t.co/dOOShrHLNy
Toronto, Canada
Joined March 2026
- Tweets6.3K
- Following2K
- Followers3K
- Likes2.2K
The most starred brand new GitHub projects of October (we track about 10,000 of them):
17,590 stars: removes ads and bloat from your Android TV
13,581: a math library
10,428: DeepSeek inference framework
5,458: an AI agent skill for drawing diagrams
4,832: Microsoft Word rewritten in Rust
4,312: an ad-free TV home screen
4,223: removes ads and push junk from Android phones
3,806: another ad-free TV home screen
Count the pattern: 4 of the top 12 are about getting ads out of your own TV or phone, 37% of all the stars in the top 12.
Only 2 of the top 12 are AI native.
The biggest thing developers loved this month was not an AI tool. It was an Android TV with no ads.
More on nicheloom 馃憞馃憞馃憞馃憞馃憞馃憞
Niche of the day: specialized calculation tools, 17 launches today.
Vertical of the day: Media & entertainment, 144 launches today.
Vibe-codeable today: 61% (671 of 1099 assessed launches).
I tracked about 18,500 new GitHub repos. A scan turned up about 235 that look fake, roughly 1 in 80.
Two kinds:
1. Repos named after paid software (Pro, Suite, Full, Unlocked). 78 of them. Several have 200 to 700 stars, which puts them near the top of any list sorted by stars.
2. One page copied over and over. 37 repos with the exact same description, "AI API gateway / relay: 307 models behind one OpenAI-compatible key", with names ending -fr, -ja, -ko, -cn, -de, -ru. All posted in 2 days.
The template alone was 23% of an entire "niche" in our data.
Others: 27 repos about one video game's save files, 28 "best software for [game] yet" repos.
Lesson: if you spot "the fastest growing category" on GitHub, check the repo names first. Some trends are one person pressing copy.
More tracking on nicheloom 馃憞馃憞馃憞馃憞
New GitHub and Hacker News launches, first half of October vs second half (same total volume, 4,964 vs 4,809):
AI coding agents: 23 to 36 (+57%)
Model context protocol tools: 73 to 51 (-30%)
AI agent platforms: 57 to 37 (-35%)
LLM inference engines: 30 to 16 (-47%)
Developer command line tools: 33 to 12 (-64%)
Agent skill registries held flat at about 77 a fortnight.
The shift: fewer people building the plumbing (MCP, inference, platforms), more building the thing on top (coding agents).
Software engineers earn more than 300k in silicon valley but with AI in picture bot software engineers and their salaries are decreasing now
Niche of the day: ai video generation tools, 13 launches today.
Vertical of the day: Media & entertainment, 192 launches today.
Vibe-codeable today: 63% (774 of 1223 assessed launches).
I looked at 1,152 Y Combinator companies from the last year.
35% are AI agents. 401 of them.
10% are hardware.
11% are software for one industry.
By industry (when they name one):
-Healthcare: 97
-Fintech: 96
-Manufacturing: 89
-Sales: 42
-Security: 39
YC backs more manufacturing startups than sales tools and security combined.
I sorted 24,000 product launches from the last 4 weeks by industry and checked how many are built around AI.
Overall: 29%.
Most AI:
-Sales: 43%
-HR: 39%
-Marketing: 36%
-Media: 33%
Least AI:
-Travel: 11%
-Fintech: 15%
-Commerce: 16%
-Manufacturing: 15%
Sales and marketing tools are mostly writing, outreach and reports, which suits AI. Travel, money and physical goods are slower to adopt.
If you want less AI competition, build for travel or fintech. If you want the most demand to build around AI, go where it is already crowded.