ironbyte-rgb for crescevo

Posted on Jun 11 • Originally published at ai.crescevo.com

I Tracked 592 Signals Across 6 B2B Niches for 3 Weeks. AI Won — But Not How You'd Expect

#ai #machinelearning #llm #programming

TL;DR

I run a scanner that watches Hacker News, arXiv, Reddit, and industry feeds across six B2B niches and scores what's breaking through. Over three weeks (May 22 – June 11, 2026) it logged 592 distinct signals.
AI was 45% of everything — 264 of 592 signals — more than tech, marketing, finance, work, and business combined-ish. AI isn't a story; it's the weather.
The twist: of the ten AI stories that broke through hardest, only one was a model launch (Claude Fable 5). Five were about AI fatigue, backlash, or culture — "Please Use AI," "I'm Tired of Talking to AI," "Tech CEOs are apparently suffering from AI psychosis."
Honest caveat up front: 71% of signals came from Hacker News, so this is the builder/technical internet's attention, not the whole market.

Most "state of AI" pieces are written from vibes. I wanted to write one from a log file. For the last three weeks I've been running a content signal scanner — it pulls from Hacker News, arXiv, Reddit, and a handful of industry feeds, sorts each item into one of six B2B niches (AI, tech, marketing, finance, work, business), and scores it by velocity and topical fit. This is what 592 signals actually showed, numbers first.

How this was measured (so you can discount it correctly)

Between May 22 and June 11, 2026, the scanner logged 592 unique items. Every signal is a real URL with a computed attention score. Two honest limitations before any conclusions:

Source skew. 420 of the 592 signals (71%) came from Hacker News — 274 from the front page and 146 from Ask/Show HN. Another 107 were marketing RSS feeds and 64 were arXiv papers. So "what breaks through" here means what breaks through on the technical internet, weighted toward developers and founders.
Scores aren't comparable across source types. A Hacker News item is scored by upvote velocity; an RSS item gets a flat base score. So I compare within a source, not across. When I say a signal "broke through," I mean it ranked high among comparable items.

Finding 1: AI is 45% of the entire conversation

The niche split was lopsided. AI accounted for 264 signals (45%), followed by tech at 121 (20%) and marketing at 115 (19%). Finance (43), work (27), and business (22) made up the long tail. Every single arXiv paper that surfaced was AI/ML. If you work in B2B and feel like AI has eaten the agenda, the log agrees: it is nearly half of everything that moved.

Finding 2: What breaks through in AI isn't capability — it's exhaustion

Here's the part I didn't expect. I ranked the AI signals by attention and looked at the top ten. Exactly one was a frontier-model launch: Claude Fable 5, which topped the entire dataset. After that, the list turns human:

"Please don't spam people looking for employment" — backlash at AI-generated application spam.
"Please Use AI" — satire about mandated AI adoption.
"I'm Tired of Talking to AI" — fatigue, stated plainly.
"Tech CEOs are apparently suffering from AI psychosis" — the culture turning on its own hype.
"If you're an LLM…" — the prompt-injection meme that won't die.

Five of the top ten AI breakouts were about fatigue, backlash, or absurdity — not benchmarks. The remaining four were a security scare (an Instagram account-hack story), a search-behavior shift (DuckDuckGo gaining after a Google AI move), a product-market-fit take, and a markets story. The launch everyone expected to dominate did dominate — once. Then the feed filled with people who are tired.

If you're building or marketing in AI, that's a signal in itself: the audience that decides what spreads is no longer impressed by "state of the art." It's negotiating its relationship with the tools. Content that names the fatigue is, right now, outperforming content that sells the capability.

Finding 3: the technical internet still sets the agenda

The source breakdown is its own finding. With 71% of signals originating on Hacker News, the stories that "break through" in these niches are overwhelmingly the ones that win a developer-and-founder audience first. For B2B teams, that's a practical routing note: if you want a piece to travel, it has to survive the most skeptical room on the internet before it reaches everyone else.

What I'd do with this

If you make AI content: stop leading with capability. The data says the conversation that spreads is about adoption friction, fatigue, and trust. Write the honest version.
If you're in the other five niches: AI is 45% of attention, so the highest-reach angle is almost always "what AI means for your niche," not generic niche news.
If you distribute content: assume the technical internet is the first gate. A piece that can't earn a skeptical Hacker News read probably won't travel far in B2B either.

This is three weeks of one scanner's view, skewed toward the technical web. It is not the whole market. But it's a log file, not a vibe — and the log says the AI story in mid-2026 is less about what the models can do and more about how tired everyone is of hearing about it.

Frequently asked questions

How many signals and over what period?

592 unique signals between May 22 and June 11, 2026, collected by an automated scanner across Hacker News, arXiv, Reddit, and industry RSS feeds, sorted into six B2B niches and scored by velocity and topical fit.

Why is Hacker News overrepresented?

71% of signals (420 of 592) came from Hacker News, which is one of the scanner's primary sources. That makes this a view of the technical/builder internet's attention, not the entire B2B market — a limitation worth keeping in mind when reading the findings.

What was the single biggest signal?

Anthropic's Claude Fable 5 launch topped the entire dataset. It was the only frontier-model launch in the top ten AI signals; the rest skewed toward AI fatigue, backlash, and culture.

Can I compare scores across niches directly?

Not cleanly. Hacker News items are scored on upvote velocity while RSS items get a flat base score, so comparisons are most reliable within a source type rather than across them. Niche volume (how many signals) is the more apples-to-apples number.

Before you go — a free checklist

Shipping an AI feature? The LLM-in-Production Checklist covers the 27 things to verify before you ship — evals, RAG, cost, fallbacks, and safety. Subscribe free to unlock it.

Originally published on AI at Crescevo — subscribe free for more.

DEV Community