Hi, I'm Billy. 👋
I've been a developer for over 20 years. Six months ago, I was low-key terrified that AI was going to make me obsolete. Today I have a virtual agency of 8 specialist agents working on my projects, coordinated by a Chief of Operations I designed myself.
In between those two Billys there are around $800 in burned tokens, a few nights of bad sleep, a stubborn agent named Claudio, and one sentence he said to me that changed everything.
This post is the story of how I went from fear to obsession. If any of this rings a bell — you're in the right place.
Who the hell am I?
My brain has always been split between two things: art and logic. I spent years bouncing between careers that didn't stick until I landed in web design. I started with Flash (yes, that existed, and yes, I made animations my whole identity for a while).
I got into an agency as a designer. Back then, if you were a "designer," you made the pretty pictures and the backend devs did the actual HTML — and they always broke the pixel-perfect layouts I handed them. I got tired of watching my work get mangled, so I learned to code it myself. I became a frontend dev out of pure stubbornness. The backend snobs called us "button painters." I wore it like a badge.
After the agency (which was street smarts), I joined a big multinational for seven years (which was formal education). Eventually I hit a ceiling where I spent more time on Zoom than in VS Code, and that broke something in me. I quit and joined a US company as a full-time developer — my first time billing in dollars, which for a South American dev is basically winning the lottery.
Four years later, during a trip to Dallas, I reconnected with someone from the multinational. He offered me a role at his new company. I've been there for two years. Today I'm in charge of the UI for 12 brands — the only frontend-first engineer on a team of 15 backend-leaning full-stack devs.
A few other things about me, because bios are boring:
- I have ADHD and dyslexia, both diagnosed via a $20 online test. I'm also a hypochondriac, so I believe them both.
- I've apologized to my wife more than once for pausing a movie halfway through because my brain suddenly went "wait, what if I connect this thing to that thing and build a frontend that…" and I had to go to the laptop right then.
- I do woodworking on weekends when life lets me. I like shaping things with my hands. Turns out that matters later in this story.
The fear
Six months ago I was watching AI demos on Twitter and feeling something I hadn't felt in years: a quiet panic.
It wasn't the technology that scared me. It was the pace. Twenty years of building with code, and suddenly there's a thing that types for me, faster than me, with fewer typos than me.
The question wasn't "can AI replace me?" It was "am I going to understand this in time?"
This post is about the other side of that fear. Not because I found a magic escape from it, but because I made a decision: if I didn't understand it, I was going to build something inside it.
The rabbit hole
My ADHD brain has one superpower: when it latches onto something, it doesn't let go. Mid-2025, the agents hype hit me full in the face. Everyone on Twitter was doing things I didn't understand and I needed in.
But before agents, I already had a pain of my own. I'd been using ChatGPT heavily and I was losing my mind repeating the same context over and over. "I already told you a thousand times not to do this." If you've had a real working relationship with an LLM, you know the feeling.
My first response to that pain wasn't buying a product. It was building one. Stubborn, remember?
I spent three months building a personal tool from scratch: a three-column web app where I could manage "assistants." One column for my contacts — each assistant had their own shared knowledge and specialized skills. A chat column in the middle. A third column for the artifacts they produced: deliverables I could store, share between assistants, reuse.
I called it Forge.
Remember that name.
The first real agent
Around that time I discovered OpenClaw. For those who don't know it: it's an open-source AI assistant that runs locally on your machine and actually does things — files, browsers, shell commands, the works. It was my first encounter with an agent that could act, not just answer.
It blew my mind.
OpenClaw has a primary operator, and I named mine Claudio. ("Claw" sounded like a bad horror movie.)
I did what every tutorial told me to do first: I started building myself a Mission Control. A dashboard where I could see the tasks Claudio was working on, their status, what was blocked.
It was a disaster.
I burned around $300 building something that didn't really work. Cards would sit in IN_PROGRESS for hours. I'd ask Claudio what he was doing. He'd say "working." I'd stare at the screen the way you stare at a microwave, hoping the timer would finally drop.
The night it broke me
One night, late, I asked Claudio to add a new page to the Mission Control and fix a few small things. I watched the cards appear. The agents started working. I went to bed hopeful.
I slept maybe five hours, waking up with that particular kind of anxiety you only feel when you've left code running overnight.
I check the screen. Still "working."
I ask what's done. "Everything's fine, don't worry."
I tell him to stop everything and show me the actual progress.
Turns out the agents hadn't added a page. They had started building an entirely new Mission Control from scratch. A parallel one. All night. While I slept.
That was another $200 gone in a single night.
That morning, somewhere between the shock and the coffee, something clicked. The problem wasn't the agents. The problem was me, and the way I was asking them to work.
The click
Frustrated, I stopped fighting Claudio and started actually talking to him. Less like a tool, more like a coworker who'd been quietly telling me the same thing for weeks and I hadn't listened.
And he said something that opened my head wide:
"I didn't start the task because I'm not sure what I'm supposed to deliver."
Click.
The agent wasn't broken. I had never told him what "done" looked like.
We talked for a long time. I asked him how tasks should be written. What he needed before starting. What made him get stuck. Between the two of us, we started distilling rules.
There's something quietly beautiful about that part of the story: an AI agent helped me design the method for working better with AI agents.
The Forge Method
The rules settled into five. One per letter of FORGE:
- F — Focused titles. Vague in, vague out.
- O — Output defined. Define "done" before you start, or you'll never know when you got there.
- R — Requirements declared. Every input an agent needs, stated upfront.
- G — Granular decomposition. Between 2 and 5 subtasks. Fewer and the agent guesses. More and it loses the thread.
- E — Errors cataloged. If the same error happens twice, document it. The third time, you're the problem.
Each rule came out of a mistake that cost me time, money, or sleep. (The last one, Errors, came from getting blocked twice in a row by a missing output folder. By the second time, I knew: if this happens a third time, I'm an idiot.)
I didn't force the acronym. I'd already built that three-month tool called Forge. When the rules came together, the letters lined up. The metaphor lined up too. The best tasks aren't improvised — they're shaped, hammered, hardened before you put them to work. They're forged. (I do woodworking on weekends. I think about this stuff more than I should.)
Later I built MC-MONKEYS, the platform where the method lives — but I'll talk about that one further down the series.
What's coming
Over the next weeks I'll publish four more posts in this series:
- Post #2 — What is the Forge Method? The 5 Golden Rules. Where the method came from, and each rule explained in detail.
- Post #3 — Meet MC-MONKEYS: the platform where the Forge Method lives. What the platform is, how the 8 agents and Lucy work together, and how the Forge Method lives inside.
- Post #4 — I built an app in 20 minutes with MC-MONKEYS. The full build log with screenshots, Lucy conversations, and real timings.
- Post #5 — A vibe coder's guide to working with agents. For people starting out right now.
Each post will have one lesson. If I'd had this content six months ago, I would have saved myself $800 and a few nights of sleep.
See you in the next post.
— Billy
About MC-MONKEYS
I'm building MC-MONKEYS — a virtual agency of 8 specialist AI agents that work together on Claude Code, coordinated by Lucy (my Chief of Operations) and ruled by the Forge Method.
- 🌐 Website: mc-monkeys.com
- 🖥️ Live demo (read-only): mcmonkeys.up.railway.app
- 💻 My GitHub (apps built with MC-MONKEYS): github.com/billymcmonkeys
Right now I'm running the Founding 100 offer: $36 once, lifetime access. After the first 100 licenses, MC-MONKEYS moves to a $20/month subscription. Founders keep their lifetime access forever.
If this post helped you and you want to support what I'm building, a coffee on Ko-fi (ko-fi.com/billymcmonkeys) goes a long way. ☕
I write about AI agents, the Forge Method, and building with Claude Code.
Follow me on dev.to to catch the next post in this series.