an edge case essay
Stop Copping. Start Cooking.
Everybody standing in line for a chatbot is buying somebody else's product, cut the way somebody else decided to cut it, at the price somebody else decided to charge. A few dollars in API credits and twenty minutes with a terminal, and you're not a customer anymore. You're the one who cooks.
Everybody's a Customer
If you can think like a boss, you can master AI. Not because the technology is hard — the technology has never been easier — but because everybody handing their whole request to a chat window is thinking like a customer, and a customer doesn't run anything. He gets served whatever the house is serving today, at the interface the house decided to build, inside the limits the house decided to set. Type it in, get an answer, close the tab. Every person doing that is working the exact same corner, buying the exact same product, indistinguishable from the next guy in line.
Ditch the chatbots. Spend a few dollars on API credits, spend a few minutes learning to use a terminal, and build your crew. That's the whole move. Nobody's gatekeeping this from you — the account, the key, the command line are sitting right there, free for anyone willing to spend twenty minutes not being intimidated by a blinking cursor. The gap between the guy typing into a box and the guy running his own operation isn't talent. It's posture.
Turning Cocaine Into Crack
I know exactly what that comparison sounds like, and I'm making it anyway, because it's the most accurate one I've got: going from chatbot to your own API-driven terminal is like turning cocaine into crack, but you won't catch a bid or start selling your ass. Same raw material. Completely different product. Somebody selling you the finished, cut, pre-packaged version has already taken his piece off the top before it ever reaches you — the markup, the limits, the interface standing between you and the thing itself. Cook it yourself and you get the whole yield, at full strength, on your own terms.
That's not a metaphor I'm reaching for to sound clever. It's the actual economics. A chat window is a retail product with somebody else's packaging on it. An API key is the raw material with nobody standing between you and what you do with it. The skill required to go from one to the other is real, but it is not advanced chemistry — it's a folder, a key, and the willingness to type commands into a terminal instead of a chat box. That's the whole cook.
Build Your Army
Here's the actual mechanics, no more dressed up than they need to be. Put some stuff into a folder about what you want to do and the players that can make it happen — what the job is, who's on it, what each one owns. Spin up a CLI agent and tell it to build your army by creating those soldiers you just described, each one in its own subfolder. That's it. That's the whole setup instruction.
Then you work it like an operation, not a vending machine. Start asking it to do things. Ask it how to do things — the terminal answers that question too, and it doesn't get impatient with you for asking. Spin up the agents in the other folders. Make them talk to each other. Every subfolder is a soldier with a job; the folder above them is you, running the operation, not waiting in line for a single box to answer a single question one prompt at a time.
Discipline them when they fuck up, but tell them how to do it right. That line does double duty and I mean both halves of it. An agent that goes sideways doesn't get abandoned, it gets corrected — you tell it exactly what it got wrong and exactly what right looks like, and it doesn't take the correction personally, doesn't sulk, doesn't need it said twice in a softer tone. That's the whole management problem, solved, minus every reason management is usually a headache.
The Crew Talks to Itself
This is where the whole thing stops being a productivity trick and starts being an operation. One agent researching. One agent writing. One agent checking the first one's work. One agent doing the boring part nobody wants to do at 2 a.m. Each one in its own subfolder, its own scope, its own job — and you at the top, giving instructions and correcting course, the way a boss runs a crew instead of the way a temp runs a single task. This essay came out of exactly that setup. I'm not describing a hypothetical.
Multiple agents, multiple tabs, multiple sessions cycling — that reads like chaos to somebody who's never run more than one thing at a time. It isn't. It's the same skill that separates a boss from a mule: knowing who's doing what, catching it fast when it's wrong, and never doing the work yourself when you could be running the people who do the work. Except here, the people are soldiers you described into existence with a folder and a prompt, and they never call in sick.
No Bid, No Corner
I know what the real version of this operation costs when it goes wrong. I've done federal time and state time, both drug-related, and there's no version of that story where the corner was worth what it took from me. That's exactly why this comparison isn't cute to me — it's the whole point. Same instinct, same hustle, same willingness to run something instead of just using something. None of the risk. Nobody's catching a bid over API credits. Nobody's out on a corner over a terminal window.
The chatbot was never the ceiling. It was the entry-level product, dressed up to look like the whole business. The real thing was always one folder and one key away, waiting for somebody to stop standing in line and start running the operation instead.
Related: The Importance of Injecting Your Seed — on what actually makes the output yours once you've stopped renting somebody else's box to get it. And on the corner I'm actually talking about, not the metaphor: Tranq Doesn't Kill You. It Won't Let You Die.
— The Edge Case