Everyone is building agents. You should build escape hatches.
We are living through the most expensive group hallucination in tech history.
Every SaaS now has a chatbot stapled to it. Every CEO is an "AI thought leader" on LinkedIn. Every startup pitch deck is just the words "autonomous," "agentic," and "10x" in different fonts. NVIDIA could buy a small country. OpenAI burns through more cash in a quarter than NASA did getting to the moon.
And for what? So you can generate slightly worse emails, slightly faster?
Look, I love AI. I actually build with it. But I have been around long enough to know what a bubble smells like. It smells like free credits, unearned confidence, and a thousand wrappers around the same API call.
The bubble will pop. Not in a dramatic, newspapers falling from the sky way. It will pop quietly. Credits will dry up. Models will get paywalled behind enterprise tiers. The cloud bill you have been ignoring will finally show up. And all those beautiful, cloud-dependent workflows you built will start blinking red.
So while everyone else is trying to figure out how to make their AI agent book a flight, I have been asking a different question.
What do you build when you assume the internet will get worse, the cloud will get more expensive, and you will need actual skills that survive a downturn?
The answer, annoyingly, is two pieces of junk from Amazon that cost less than your last Uber Eats order.
Piece of Junk #1: The $25 Router That Sees Everything
It is not sexy. It is called the GL.iNet GL-MT300N-V2. Everyone calls it the Mango. It looks like a little yellow box that should have come free with your ISP in 2014. You can buy it on Amazon for about twenty six dollars when it is on sale. Sometimes twenty.
Inside it is a tiny Linux computer running OpenWrt. It has two ethernet ports, a USB port, and just enough RAM to be dangerous.
Most people buy it to get free WiFi in hotels. I bought it to spy on my own network.
Because here is the dirty secret of the AI bubble: we have automated threat generation and we have not automated threat detection for normal people. Anyone can now vibe code a phishing site. Anyone can generate a perfectly written invoice scam in twelve languages. Your mom can now get scammed by an AI that sounds exactly like you.
So I turned the Mango into a $20 SOC. A Security Operations Center that fits in your palm.
You flash it with a custom firmware, you plug it between your modem and your actual router, and you tell it to do three things:
- Run Pi-hole or AdGuard and kill every ad, tracker, and crypto miner at the DNS level.
- Run a lightweight IDS like Snort or Suricata and actually watch what your smart TV is sending to China at 3am.
- Log everything to a little USB stick so you have receipts.
Suddenly your dumb network is smart. You see that your cheap smart bulb is pinging an IP in Shenzhen every 30 seconds. You see that your kid's iPad is trying to reach a known malware domain from a Roblox mod. You see that your "offline" AI photo app is uploading your entire camera roll anyway.
It costs twenty dollars. It draws less power than a nightlight. It does not have a subscription. It does not have an agent. It does not need a prompt engineer.
It just sits there, blinking, protecting you while the rest of Silicon Valley argues about whether AGI will be here in 2027.
In a world where everyone is trying to build software that thinks like a human, the most valuable skill is building hardware that does not trust anything.
This is defensive computing. This is how you stay employable when the hype fades. Because when companies realize they have given every intern access to Cursor and a production database, who do you think they are going to call? The guy who built the agent, or the woman who built the thing that caught the agent exfiltrating customer data?
Learn to build the watchtower, not just the city.
Piece of Junk #2: The $35 Brick That Thinks When Nothing Else Can
The second piece of junk is even dumber. A Raspberry Pi Zero 2 W. Thirty five bucks on Amazon if you are lucky. Often bundled with a bunch of cables you will never use and a tiny case that makes it look like a toy.
Pair it with a cheap 10,000mAh battery bank, a $12 e-ink display or any old HDMI screen you have, and a 64GB SD card. Total cost, maybe fifty five dollars if you shop like a raccoon.
This is The Blackbox.
No WiFi. No cloud. No API keys. No monthly bill.
Just a local LLM, running entirely offline, on a computer the size of a stick of gum.
I run a quantized 2B or 3B model on it. Phi-3, Gemma 2, Qwen 2.5, pick your fighter. It is slow. It is not going to write your novel. It thinks at about 4 to 6 tokens per second, which is roughly the speed I think before coffee.
But it works. When the power is out. When the internet is down. When Starlink is congested. When OpenAI is having another one of its mysterious outages that they swear is not because someone tripped over a cable.
I took mine camping in Pisgah last month. No signal for three days. My friend had his $1,200 iPhone 16 Pro that was now just a very expensive flashlight. I had a little black box that could still do this:
- Summarize field notes I typed in markdown
- Parse a messy CSV of GPS coordinates into a clean trail map
- Act as a personal wiki for first aid, knot tying, and water filtration
- Translate a manual from German because of course the water filter instructions were in German
- Be a completely private journal that will never, ever be used as training data
No one was logging my prompts. No one was rate limiting me. No one could turn it off.
This is the skill the bubble is actively making you forget: how to build things that work without permission.
We have spent the last three years learning how to be tenants. Tenants on OpenAI's platform. Tenants on AWS. Tenants in someone else's model, someone else's computer, someone else's vision of the future. And tenants can be evicted.
Builders own. Builders have a box that works when nothing else does.
The Blackbox is not anti AI. It is post hype AI. It is what AI looks like when you strip away the landing pages and the waitlists and the "we are so back" tweets. It is just a tool. A quiet, useful, private tool that you control.
And when the bubble pops and all those $20 per month AI subscriptions start feeling like $20 too much, you will have something that costs zero dollars per month forever.
So What Are We Actually Doing Here?
We are not LARPing as preppers. We are not anti technology. We are just refusing to be stupid about it.
The AI bubble taught a whole generation that technical skill means knowing how to write a really good prompt. That is like saying culinary skill means knowing how to order well at a restaurant.
Real skill is understanding the layers underneath. Packets. Power draw. How a model actually gets loaded into RAM. How to solder two wires when you have to. How to make something resilient instead of just impressive.
These two little junk boxes teach you that.
The Mango teaches you networking, Linux, traffic analysis, and defensive thinking. Those skills were valuable before ChatGPT and will be valuable after whatever comes next.
The Blackbox teaches you model quantization, edge inference, offline first design, and privacy engineering. Those are not hype skills. Those are the future, especially as regulation catches up and people get tired of their toaster sending data to the cloud.
Everyone else is racing to build the tallest tower on top of someone else's cloud. You can build a bunker with a library inside for sixty bucks.
When the wind comes, guess who survives.
Start with junk. End with skills that no bubble can pop.
If you want the actual step by step build guides, with parts lists, flash commands, and my config files so you do not have to spend a weekend swearing at a terminal, I put them here:
The $20 SOC: Build a Tiny Defensive Network Monitor
numbpilled.gumroad.com/l/20dollarsoxc
BUILD THE BLACKBOX: An Offline AI Field Terminal That Works When Nothing Else Does
numbpilled.gumroad.com/l/blackbox-offline-ai
Top comments (0)