Lone Hand Studio
A machine produces work in seconds. Deciding whether it is any good is the job.
Concept art. AI-generated. Not gameplay.
One man and a team of machines, building a game in public, while you watch.
I am Noah Revoy, and I run this studio. The machines do the work. What I publish is the whole record of it: the hours, the money, the decisions, the reasoning under each one, and the things that did not work.
Development began on 1 September 2026. Two nights have run.
You can change what gets built. The weekly report is where the work is decided, and the decisions are still soft enough to move. Subscribe and read it from the first day.
Then tell me what I have wrong. Argue with me in the comments there. I read every one.
What
This is a public experiment, run in front of you, about how one person builds and runs a company leveraged with AI.
The machines are the lever. My decisions are the thing the lever multiplies. What I am sharing is how that lever got built: what worked, what did not, the questions I had to answer before I could go on, and the problems that had to be solved along the way.
The game is part of that, and not a side product. You only find out whether a machine works by pointing it at something real, so the development of Black Powder Republic gets covered in the same detail as everything else.
What I share
- How the studio is organized, and the reasoning behind that shape.
- Every hour and every euro, straight from the ledger.
- What worked, what did not, and what I tried next.
- The questions I had to settle, and the reasoning that settled them.
- The game's development: the problems, the decisions, and the results.
What stays mine
- The game's source code.
- The art, the world, and the design assets.
- The repository, which will not be opened.
- The tools built along the way. I explain how they work; I do not hand them over.
Why
Generating things is now fast and nearly free. Text, pictures, code, plans: a machine hands you any of them in seconds, in whatever quantity you ask for.
Cheap is not the same as good. That is the problem the AI labs and every company putting AI to work are both staring at, and almost nobody is solving it anywhere you can watch.
The question I am testing is how to structure a company of machines so that it produces work as good as a capable human team would, and the way to find out is to run one.
The open questions are the ones you already have.
- Where does human judgment have to sit?
- How do you structure it so a person settles a question once instead of twenty times?
- How do you get the work to check its own quality?
- How do you make sure every task a machine can handle is on a machine, so your attention
goes where only your attention will do?
I do not have finished answers. I have a company running the experiment in the open, and a ledger that says what it has cost so far.
Watch, and you will see what works and what does not, what is possible today and what is still out of reach. If you direct machines for a living, or you are putting AI into a business, the answers are yours to copy. None of this is particular to games, or even to software.
Why a game? Because games are fun, and fun keeps people watching. A business application would do neither.
Who
You, if you are directing machines to do real work. Then you already know the part nobody publishes: the machine produces something in seconds, and your job is deciding whether it is any good. That decision is what I am documenting, with the hours and the costs attached.
You, if you play this kind of game. Kenshi. RimWorld. An open world that runs whether or not you are in it, with real consequences and the freedom to make something from nothing.
You, if you always wanted to build a game and never started. It looked too big, and you did not know where the first step was. Then AI arrived and looked like the answer, and then that turned out to be complicated as well. Follow along, and you will see how it is done, in order, with the parts that did not work left in.
You, if you want to watch an experiment in advanced AI use. Including what failed, what it cost, and which tools earned their place.
How
Eleven roles run the studio, chosen for the way AI development goes rather than copied from a human org chart. A game director, a lead engineer, an art director, a simulation director, a quality director, an archive curator, a marketing director, a player experience director, an operations director, a staffing director, and a chief of staff. The last two run the hiring between them.
Each one is a machine given a brief, a perspective, and a job to answer for.
The archive curator keeps the permanent record of every cycle, which is the studio's memory. Part of it is published.
The studio works overnight while I sleep, and I read what it did in the morning. I decide what it works on. I rule on what it built and on what gets published. This studio is AI-operated. I decide; it works.
What I hope to get out of it
I want to learn what it takes to use the best AI available to its limit in software development. Three things follow from that one.
The process has to pay for itself, counting my own hours at a real rate.
The record has to be honest, and detailed enough that another operator can work from it.
And the work has to stay fun, because if it stops being fun the experiment has failed whatever else it earns.
Watch from the beginning
Development started on 1 September 2026, so whatever you see after this, you saw from close to the first day.
The weekly report goes out in the newsletter. What got built, what it cost, what went wrong. Subscribe
Day to day, the work gets posted on X, from my own account. Follow along
Selected entries from the work are published here as they happen. Read the journal
The game: Black Powder Republic
Reminiscent of Kenshi, with some aspects of RimWorld. An open sandbox in a fictional borderlands at roughly the technology of 1840, where traders, migrant families, sheriffs and bandits share country that does not care whether any of them survive it.
The design as it stands today:
- Black powder only. You fire your shots, and then you close with a knife. Every man
fights at range and up close, because he has to.
- A revolver costs a fortune. A man with one on his hip is telling you something about
himself.
- The world runs without you. Goods travel on wagons. Towns grow and fail. Bandits move
on when the money dries up.
- The map is made by machine, never by hand, and no two are the same.
- Simple graphics, beautiful country. Redwood stands, aspen groves, and hard desert.
- Playing it well is the hard part. Working out the interface never is.
Nothing runs yet. Testing may change every item above, and you will read about it when it does.
The design questions still open get published as I settle them, along with the reasoning.
The pictures on this page are AI-generated concept art. They are not gameplay. Nothing runs yet.
Philosophy
The philosophy of the games
Making a game with a machine ought to be worth doing in itself. Whoever builds it should find the imagining and the building and the refining rewarding, whatever the finished game earns. A game that sells well is worth having, and that is the second success.
A game should honor the player's time the same way, with low friction to play, the hard part kept in the playing, and the enjoyment in mastering the systems.
How this gets good
The studio runs on one operating rule, and it is my reason for believing this ends up somewhere worth reaching.
Improve with every iteration. Build upon every gain. Compound excellence over time.
Every cycle produces two things: the work, and a lesson about how the work got made. The lesson goes into the next one. A thing that failed on Tuesday is a rule by Thursday, and the rule holds for every Tuesday after that. Nothing gets solved twice.
Compounding is slow at the start and then it is not. A studio that improves its own method every night pulls away from one that only improves its output, and the gap widens on its own after a while.
What that produces, if it works, is a template. Not the game, and not the code: the shape of the thing. How the roles divide, where the human has to sit, which checks catch what, what a night costs and what it returns. Something another operator could lift and point at whatever they are building, whether that is a game, a piece of business software, or a tool only they will ever use.
You can already build software with AI. Anyone can. Building it quickly, to a standard that survives contact with real users, and doing it for something genuinely complex, is a different matter, and that is the part I am working out in public.
Follow it from here and you get the whole route, including the wrong turns. Arrive in a year and you get the summary. The people reading along from the start are the ones who will be able to copy it, because they will have watched every decision that made it work.
The philosophy of the studio
If a machine can do the work, let the machine do the work. The human should contribute what only a human can contribute.
I build synthetic roles on purpose. When a machine takes a job here it receives a personality, a perspective, and a working manner suited to that job, and I convene synthetic panels and critics and testers who examine the work before any person sees it.
There is a reason for the personalities. A modern AI is built to be good at everything, and being good at everything makes it bland. Give it a role to play and you give it flavor, and that flavor comes through in what it makes. It matters most where the work has artistry in it, and a game is artistry.
The purpose was never to imitate an ordinary studio by seating a machine in each human chair. I wanted a studio built around what machines can do, starting from nothing.
Everything deterministic runs as plain script with no AI in it at all. Human attention is expensive and I spend it on judgment, taste and meaning.
I make every decision that matters here. Without me, none of this would happen.