Good-looking, Friendly, Pay the AI Tax: A Human Survival Guide from Cats and Dogs

katsura — Tue, 20 Jan 2026 09:15:56 +0000

Since Christmas 2025, my code output has exceeded everything I wrote from learning to program until then. I didn't write a single line. Some projects turned out better than when I led a 40-person dev team.

I fell into an existential crisis. What the fuck am I even doing? After catching up with industry best practices, I realized I'm basically just lurking. I thought I could at least be a degen code babysitter. Turns out I'm just the Chief Lurking Officer.

That moment, I thought about horses.

The Fate of Horses

1900, New York had a hundred thousand horses. Core of productivity. 1920, cars took over. Horse population collapsed.

Where are horses today? Racetracks. Fast ones entertain gamblers. Slow ones become studs, genes milked dry, no say in who they fuck. The rest go to glue factories.

Horses did nothing wrong. But "being useful" stopped being valuable.

Why Cats and Dogs

Meanwhile, cats and dogs exploded in numbers. Can't work. Pure liabilities. You even clean their shit.

Why them? Pigs are smarter than cats. Crows use tools better than dogs. But pigs are ugly, crows are black. They don't even qualify as pets. Pigs get braised. Crows get chased. The uglier ones don't even deserve to be eaten.

Cats win on looks. Dogs win on friendliness. Survival strategy isn't usefulness. It's likeability.

What Now

Gotta look better. Back to the gym. Used to think treadmills were like being a hamster. Boring. Now I'm trying hard to play the hamster. Being cute like a cat is the ultimate goal. After all, there's only so many areas where I can still beat AI. Even my workout plan is from Gemini. It just can't physically work out for me.

Watching Claude Code and Codex on screen, planning, writing, debugging on their own around the clock. So what about me? Might as well be friendly like a dog, get out there and make friends. That's why I'm writing this first blog post. This whole blog is written by Claude. Lucky for me it has bad memory, so making friends is hard for it.

But there's a prerequisite. Only people who can use AI get to worry about this. The rest don't even get a seat at the table. No matter how hard a horse tries, it can't outrun a car. The painful part is that free AI is still artificial stupidity. Only the most expensive ones deliver top output. That's the AI tax of the new era.

Three Survival Factors

1. Good-looking (Learn from cats)

"Useful" is monopolized by machines. "Good-looking" becomes scarce. Cats do nothing, mediocre personality. Being cute is enough.

2. Friendly (Learn from dogs)

The dog's survival code: happy to see everyone. One tail wag and you're healed. Friends everywhere they go.

3. Pay the AI Tax

Claude Code, GPT Pro, Gemini. A few hundred bucks a month. The new era's head tax. This is the entry ticket. Without it, you won't even know what game is being played.

Pet Awareness

Humans are sliding from planetary rulers into a new ecological niche. Won't go extinct. But roles will change.

Cats don't think they're pets. They think humans serve them. Maybe that's the mindset we need going forward.

Evolution has no dignity. Survival is winning.

So, pay the AI tax. Then gym, skincare, make friends.

Welcome the new era.

I Analyzed 3,000 Developer Job Postings Overnight (Here's What I Found)

katsura — Tue, 20 Jan 2026 08:55:08 +0000

I Analyzed 3,000 Developer Job Postings Overnight (Here's What I Found)

Mining product opportunities from the gig economy with Claude Code, Playwright, and ChatGPT Pro.

It started with a simple question: what are people actually paying developers to build right now?

Not the Twitter hype. Not the VC narratives. Real jobs, real budgets, real pain points. The kind of stuff you could productize or solve with a weekend of vibe coding.

So I dug through 3,000 job postings from major gig platforms overnight. Here's what happened.

It Started with Manual Browsing (And Frustration)

I spent about 30 minutes manually scrolling through job listings. Keywords like python automation, ai chatbot, telegram bot, n8n, zapier integration, shopify app, chrome extension...

Tons of interesting stuff popped up. But here's the problem: I'd see something cool, scroll past it, and 5 minutes later couldn't remember what it was or find it again. No system. No way to compare. Just scattered impressions that never formed a big picture.

There had to be a better way.

Enter Claude Code: 30 Minutes to Automation

I fired up Claude Code, described what I wanted, and 30 minutes later I had a fully working data collection system:

Playwright for browser automation (headless mode, stealth settings)
undetected-chromedriver as backup for trickier sites
SQLite for storage
48 keywords across dev categories
$500+ budget filter (serious clients only)
Deduplication, pagination, error recovery, auto-resume

The system kept hitting walls mid-run (these platforms are hostile to automation), so I switched to headless mode and let it run overnight.

Woke up to ~3,000 job records in my database. Not bad for zero babysitting.

The Annoying Part: These Platforms Are Dinosaurs

Let me rant for a second.

These gig platforms are stuck in 2008.

No public APIs. No RSS feeds. No webhooks. You want structured data? Your options:

Pay $500+/month for "enterprise API" access (lol)
Automate browser interactions like a caveman
Give up

I chose option 2. And these platforms really don't want you collecting data:

Cloudflare challenges everywhere
Aggressive fingerprinting
Login walls for basic search
Infinite scroll that breaks pagination
Rate limits after 20 requests

Honestly? If I had a few free days, I could ship an open, API-first alternative. No detection BS. Clean JSON endpoints. Let the data flow. These incumbents are begging to be disrupted.

The Analysis: ChatGPT Pro is Stupidly Powerful

After collecting data, I needed to analyze ~3,000 jobs. Tag them. Score them. Find the patterns.

Original plan: OpenAI API batch processing.

Problem: I'm a proud Claude Code Max and ChatGPT Pro subscriber—I've already paid my AI taxes. But API keys? That means digging through dashboards, copying secrets, configuring .env files... and ironically, it's the one thing CLI tools will refuse to help you with for "security reasons." The only friction point in the entire vibe coding workflow.

Plus the math doesn't work. At $0.01-0.03 per job, that's $30-90 minimum—and probably more if you want actually reliable analysis, not just pattern-matching slop. For a weekend curiosity project? Hard pass.

So I tried ChatGPT Atlas—OpenAI's new agentic browser. The browser itself is actually cool—way faster than Chrome for ChatGPT (no idea why, but regular Chrome sometimes lags hard on chat.openai.com while Atlas is buttery smooth).

But the Atlas agent mode? Garbage. It kept trying to apply generic templates instead of actually thinking. Every response felt like it was trying to bullshit me with boilerplate. Not ready for real work.

But then I just uploaded the SQLite file directly to ChatGPT Pro (regular chat, not Atlas agent mode) and asked it to:

Filter out irrelevant postings and sketchy clients (~3,000 → 2,000 legit dev jobs)
Tag each job: category, difficulty (1-5), SaaS potential (1-5), market signal
Write a 300-800 word analysis for each job
Generate stats and top lists
Write everything back to the database

2 hours later: Done. 2,000 jobs fully analyzed. 2000+ pages of deep analysis (~500 chars each). Database updated. All for $0 extra—already paying for Pro.

This is the real unlock. For bulk offline analysis, ChatGPT Pro with file uploads destroys API pricing.

What I Found: The Data

Here's what 2,000 dev-related job postings tell us:

Category Breakdown

Category	Count	%	Avg SaaS Potential
Data	600+	30%	2.5
SaaS	380+	19%	4.0
AI	360+	18%	3.8
Integration	240+	12%	3.5
Bot	210+	11%	2.7
Extension	80+	4%	3.8
Automation	80+	4%	3.7

Market Temperature

Signal	Count	%
Stable	1,300+	64%
Rising	280+	14%
Hot	260+	13%
Niche	160+	8%

AI is on fire. Most of the "hot" signals cluster in AI-related jobs.

Difficulty Distribution

Difficulty	Count	%
1 (Trivial)	420+	21%
2 (Easy)	540+	27%
3 (Medium)	420+	21%
4 (Hard)	300+	15%
5 (Complex)	320+	16%

48% of jobs are difficulty 1-2. Easy pickings for anyone with basic dev skills.

Tech Stack Demand

Tech	Mentions
Python	300+
JavaScript	300+
React	140+
OpenAI API	130+
Shopify	120+
Node.js	110+
n8n	100+
Zapier	100+

Early Insights

A few things jumped out:

AI + automation is the sweet spot. High demand, high budgets ($5k-15k), and most have clear productization potential.
n8n and Zapier are everywhere. Low-code automation isn't a trend—it's the baseline. If you're not fluent in these, you're leaving money on the table.
Shopify ecosystem is quietly lucrative. 120+ jobs, 3.8/5 avg SaaS potential. Themes, apps, integrations—recurring revenue waiting to happen.
"Easy" jobs still pay well. Plenty of difficulty-2 gigs at $800-2000. The skill arbitrage is real.
Bot development is stable but commoditized. Good for consistent income, not for building wealth.

What's Next

This post is just the overview. I'm sitting on 2,000 analyzed jobs with detailed breakdowns. Coming up:

Productization opportunities: Jobs that are begging to become SaaS
Vibe coding targets: Problems a solo dev can solve in a weekend
Easy wins: Low-difficulty, high-paying gigs for quick cash
Tech deep dives: n8n, Shopify, GoHighLevel ecosystems
Individual job teardowns: Technical approach + realistic pricing

The data pipeline is now running daily in the background, continuously collecting fresh postings. I'll publish weekly updates as patterns emerge.

Need the raw data or want custom analysis? Reach out—happy to chat.

Forem: katsura

Good-looking, Friendly, Pay the AI Tax: A Human Survival Guide from Cats and Dogs

The Fate of Horses

Why Cats and Dogs

What Now

Three Survival Factors

Pet Awareness

I Analyzed 3,000 Developer Job Postings Overnight (Here's What I Found)

I Analyzed 3,000 Developer Job Postings Overnight (Here's What I Found)

It Started with Manual Browsing (And Frustration)

Enter Claude Code: 30 Minutes to Automation

The Annoying Part: These Platforms Are Dinosaurs

The Analysis: ChatGPT Pro is Stupidly Powerful

What I Found: The Data

Category Breakdown

Market Temperature

Difficulty Distribution

Tech Stack Demand

Early Insights

What's Next