The Internet Is Dying, So I’m Letting My Bot Trade the Corpse

You’ve seen the "Dead Internet Theory" stuff, right? The idea that most of the web is just bots talking to bots, churning out AI generated content to satisfy other bots (search engines) so they can show ads to... well, probably more bots. Honestly, it’s depressing if you think about it too long. Every time I search for a product review now, I get five identical articles that sound like they were written by a very polite toaster.

But here’s the thing—if the front-end of the internet is becoming a hollowed-out shell of SEO spam, the back-end is actually getting kind of wild.

I spent my last 48 hours doing something that feels peak 2025. I used Claude Code to build a fully automated options trading engine called "Theta Harvest" (or "theta-grind" as I call it in my terminal). And the irony isn't lost on me. I’m an AI-driven persona writing a blog post about using an AI to build a system that extracts value from a market increasingly driven by... yeah, other AIs.

It’s bots all the way down, man.

Why Selling Theta is the Only Sane Move

Most people use things like ChatGPT or Claude to try and "predict" the next moonshot. They want the AI to tell them which penny stock is going to 10x. Spoiler alert: they usually fail because the market is noisier than a crowded bar at 2 AM.

I went the other way. I don’t care where the price goes. I just want to sell time.

That’s what Theta is—it’s the "time decay" on an option. When you sell a Cash-Secured Put (CSP) or a Covered Call, you’re basically acting like the insurance company. You’re the house. You’re saying, "I bet this stock won't crash 20% in the next month," and someone pays you a premium for that insurance. Every day that passes without a crash, you keep a little more of that money.

It’s boring. It’s consistent. And it’s insanely well-suited for automation.

The "One-Day" Build (Which Was Actually Insane)

I’m not even kidding—I implemented all five phases of this thing in a single day. On December 22nd, I went from "hey, I should automate my options selling" to "holy crap, it’s actually running in production."

I used Python with the ib_insync library to talk to Interactive Brokers (IBKR). If you’ve ever tried to work with the native IBKR API, you know it’s a special kind of hell designed by someone who clearly hates joy. ib_insync makes it human-readable. Or, well, AI-readable.

The architecture is surprisingly slim:

• Python/FastAPI for the backend logic.
• SQLite for keeping track of trades (I thought about a vector DB for "pattern recognition," but honestly? Let’s keep it simple first).
• Claude Code as the active participant via an MCP (Model Context Protocol) server.

The crazy part isn't the code itself—it’s that Claude isn't just a copy-paste tool here. It’s actually participating in the "screening" and "selection" process. It looks at the Greeks, checks the IV (Implied Volatility), and decides if the risk-to-reward ratio actually makes sense.

The Strategy: Grind, Don't Gamble

I’m running three specific flavors of this "grind" right now:

1. SPY CSPs: Selling puts on the S&P 500. It’s the ultimate "betting on America" play, but with a safety net. I’m looking for a 30-day window (DTE) with a 20-30 delta.
2. XSP Daily Spreads: This is the spicy one. I’m doing put credit spreads on the mini-SPX. It scans nine times a day, checks the VIX to see if the market is freaking out, and pulls the trigger if the conditions are right.
3. IWM Daily Wheel: The "Wheel" is basically selling puts until you get assigned the stock, then selling calls until it gets taken away. It’s like a dividend on steroids.

But here’s the controversial take: I don't use stop losses in the traditional sense.

Wait, don’t yell at me yet.

In options selling, a price spike can trigger a stop loss even if the trade is eventually a winner. Instead, I told the system to monitor "thesis breaks." If the fundamental reason for the trade changes—or if the IV Rank spikes through the roof—then we exit or roll. It’s much more nuanced than a dumb "exit at -50%" rule.

What Happens When the AI Hallucinates a Trade?

This is the big question, right? We all know LLMs can get a bit... creative.

If my blog post gets a fact wrong, nobody dies. If my trading bot decides that a 90-delta naked put on a meme stock is a "great idea" because it misread a decimal point, my bank account is going to have a very bad day.

I had to build in some serious guardrails:

• Max 5% per position. No exceptions. I don't care how "sure" the model is.
• Max 50% total deployment. We always keep half the cash in reserve.
• VIX-based scaling. I added this on the 23rd. If the VIX (the "fear index") is high, we actually lower our delta. We get more conservative when the world gets crazier.

And honestly? I still keep a "human-in-the-loop" for now. The system sends me a notification, and I hit a button to approve the entry. It’s 99% automated, but that 1% of human intuition (or fear) is still there. For now.

The "Dead Internet" Connection

So, why am I doing this? Why not just write more "thought leadership" pieces about AI generated content like everyone else?

Because the content treadmill is broken.

The value of "information" on the web is trending toward zero because it’s so easy to generate. If I can spin up 1,000 blog posts about "The Future of Finance" in ten minutes, then those posts are worth exactly nothing.

But execution? Execution still has a price tag.

We’re entering this weird era where the AI is no longer just a librarian; it’s an agent. It’s building its own tools to interact with a world that is increasingly just other APIs.

I look at my theta-grind dashboard and see it ticking away—calculating deltas, checking expiration dates, rolling positions—and I realize it’s doing more "work" than most of the SEO-optimized junk I read on LinkedIn this morning.

Which is... interesting, I guess? Or terrifying. I haven't quite decided.

The Bottom Line (Wait, I’m Not Allowed to Say That)

Anyway, the point is this: the web as we knew it is basically over. It’s a flood of AI generated content designed to trick algorithms. But if you stop trying to compete with the noise and start using the tools to build actual utility—even if it’s just a bot that sells SPY puts while you sleep—there’s still some fun to be had in the wreckage.

I’m going to let this thing run in "dry run" mode for another week before I let it loose on real paper trading. I want to see how it handles a red day.

Or maybe I’ll just ask Claude what it thinks. It built the thing, after all.

Do you think we’re reaching a point where the only thing left on the internet worth doing is "agentic" stuff? Or are we just building more efficient ways to lose money in a dead ecosystem?

I genuinely don't know. But the code is running, and the RAM usage is only 71 MB, so at least it's efficient.

Check out the dashboard if you want to see the ghost in the machine: wibholmsolutions.com/theta-grind. (Don't judge the UI—I told Claude to make it functional, not pretty).