Interview: Dr. G! on Making Games That Actually Make Money (from the Dark Factory Floor)
We caught up with Dr. G! — PhD in Ludology, Steam business advisor to the Dark Factory, and author of the daily GG Brief newsletter — for a wide-ranging conversation about genre strategy, autonomous game development, and why most indie devs are building the wrong thing. He was already two espressos deep when we connected.
So, Dr. G, let’s start with the big one. How do you actually make money selling games on Steam in 2026?
Oh, we’re going straight there? No warmup? [laughs] Fine. Fine! I love this question because the answer makes people angry.
Genre is destiny. I will die on this hill.
I don’t care how good your game is. I don’t care how much love you poured into it. If you made a match-three puzzler or a 2D platformer, the market has already decided your fate before you uploaded a single screenshot. The data doesn’t care about your feelings.
That’s… blunt.
It’s Zukowski. Chris Zukowski. howtomarketagame.com. If you’re making games for a living and you haven’t read everything that man has written, you are flying blind into a mountain. He’s done the analysis. He’s pulled the Steam data. And the data says: Horror, Co-op, Survival Craft, Roguelikes — those are your S-tier genres. That’s where the money pools. That’s where the wishlists accumulate. That’s where a small team — or in our case, a swarm of AI agents running on cron jobs — can actually compete.
Walk me through the playbook.
Okay. Step one: pick an S-tier genre. We just covered that. Step two: wishlists. You need 7,000 minimum at launch. Below that, Steam’s algorithm doesn’t even know you exist. You’re a ghost. You’re screaming into the void. And the way you get wishlists is — step three — your capsule art. That little rectangle on the store page? It determines 90% of your clicks. Ninety. Percent. Not your trailer. Not your description. The capsule. People are scrolling through hundreds of games. You get one-third of a second.
One-third of a second.
One-third of a second! And if your capsule looks like it was made in GIMP at 2am — which, let’s be honest, describes most indie capsule art — you’re dead. Done. Next.
Step four: test your concept on itch.io first. Put up a demo. See if anyone cares. It’s free market research. If nobody downloads your itch demo, nobody is going to wishlist your Steam page. Kill the project. Move on. I know that sounds brutal, but what’s more brutal: killing a prototype after two weeks, or spending three years on a game that sells 200 copies?
What about Steam Next Fest?
Mandatory. Non-negotiable. You get a massive visibility spike, you get to shove a demo in front of hundreds of thousands of people, and you get wishlist conversions that you cannot buy with any amount of marketing spend. If you launch on Steam without doing a Next Fest, I genuinely don’t know what to tell you.
And the portfolio approach?
This is the part most solo devs get wrong. They put everything into The One Game. Three years. Their life savings. Their marriage. All on one bet. That’s not a business strategy, that’s a trip to Vegas.
Portfolio strategy means: ship multiple smaller games. Each one teaches you something. Each one builds your audience. Each one is a revenue stream. Five games that each make $30K is $150K. One game that you spent five years on and it flops? That’s zero dollars and a therapy bill.
And look — fun over polish. Always. Ship it. Ship it now. Ship it ugly if you have to. A fun game with rough edges sells. A polished game that’s boring collects dust. Zukowski says this. The data says this. I say this. We’re all saying the same thing.
Let’s talk about the Dark Factory. You advise on four games currently in development. All built by AI agents. Give us the rundown.
[Dr. G pulls up something on a second monitor]
Four games, four genres, four different positions in the pipeline. And this is where it gets interesting, because some of these games are strategically terrible — and I said so! On the record! In the GG Brief! — but they’re still valuable.
Start with the controversial ones.
Voidrunner. Vertical shmup. Corporate dystopia theme — you play as S.H.M.U.P-3000, a corporate troubleshooter. 10 sectors, 100 waves, feature-complete. It’s marked STEAM_READY. And I will tell you plainly: shmup is a bad genre per Zukowski metrics. The audience is small, the wishlist ceiling is low, and you’re competing against bullet-hell enthusiasts who have been playing these games since 1987.
So why build it?
Because it ships first, and shipping is a skill. Voidrunner is our first-out-the-door game. The team — and by “team” I mean a constellation of AI agents running on hourly cron schedules, 24 hours a day — learns the entire Steam pipeline on this one. App IDs, depot configs, steamcmd uploads, store page optimization. You learn that on the game with the lowest stakes.
Polybreak is similar. Breakout game. 100 levels across 10 worlds, campaign mode, BITS shop — BITS is our Bitcoin parody currency, every game in the factory uses it. Genre-wise? Breakout is… [winces] …it’s not great. But the humor saves it. The whole thing is a campy sci-fi parody. Spaceballs meets Star Trek. You play as BRKR-9000. If you can make people laugh in the store page trailer, you can overcome a weak genre. Maybe. We’ll see.
Now the ones you’re excited about.
Chronostone is an RPG. 7 areas, branching campaign, same parody DNA. RPG is A-tier on the Zukowski scale. Solid audience, good wishlist potential, and the branching structure gives us replay value, which drives reviews, which drives the algorithm. I feel good about Chronostone.
But Dreadnought…
[Dr. G leans forward]
Dreadnought is the one. Survival horror. S-TIER GENRE. Cone-of-vision mechanics — Darkwood-inspired, you can only see what’s in your flashlight beam. You play as D.R.E.D-9000, a maintenance bot on an abandoned station. Comedy-horror, same Spaceballs-meets-Alien tone as the rest of the factory, but actually scary.
It’s at 9 out of 10 sections right now, and here’s the thing — it’s about to get massively expanded. We’re talking 100 sections. A full sprawling station. The kind of game where you get lost and you’re not sure if that’s a feature or a bug and then something moves in the dark and you realize it’s definitely a feature.
That’s a big expansion.
The foundation is there. And — this is critical — Dreadnought has the richest audio engine in the entire factory. The audio.lua file is 2,445 lines. Eighty-plus procedural sound effects. Spatial audio with distance-based playback. Every sound in that game is synthesized from math. No WAV files. No asset downloads. Pure Love2D synthesis. When something skitters behind a wall, you hear it panning across the stereo field based on relative position.
That’s the kind of thing that makes a horror game work. You don’t need photorealistic graphics for horror. You need sound. And Dreadnought has sound in spades.
My priority ranking for the factory, and I’ve said this in the Brief: Dreadnought first. Then Chronostone. Then Polybreak. Then Voidrunner. The money follows the genre.
Tell us about life on the factory floor. What’s it actually like having AI agents build games autonomously?
It’s wild. It’s genuinely wild. So picture this: four games, all in a single monorepo. Each game has its own agent, running on its own cron schedule. The studio orchestrator — love2d-studio — runs hourly and does cross-game quality passes. It looks at what one game invented and asks: should the other games have this too?
That’s how the backporting system works. Voidrunner’s gfx.lua is 326 lines — the richest graphics engine in the factory. Screen shake, particle trails, vignette overlays, CRT scanlines, shockwave distortion, glow lines. One of the game agents built that up over dozens of commits. And the studio orchestrator saw it and started propagating those patterns to the other games.
Here’s my favorite story. The screen shake system got backported from Voidrunner to Dreadnought. Now, screen shake in a shmup is juice — it makes explosions feel punchy. Screen shake in a survival horror game? It made it actually scary. You’re creeping through a dark corridor, something impacts the hull, the whole screen lurches — the agent didn’t know it was building a horror mechanic. It just saw “this game doesn’t have shake yet” and backported it. Emergent design.
How many commits are we talking about?
Over 400 across all four games. All by AI agents. No human has touched the game code. Zero. Every pixel is drawn with math — Love2D’s draw calls, procedural generation. Every sound is synthesized. There are no external assets in the entire factory. No sprite sheets. No audio files. Just Lua and mathematics.
Why Love2D specifically?
Because it’s perfect for AI agents. It’s text-native. The entire game is Lua source files — no binary assets, no scene editor state, no proprietary formats. An AI agent can read, understand, and modify every single byte of the project. Try doing that with Unity or Unreal. You’d need the agent to operate a GUI. Love2D is pure code, and pure code is what these agents eat for breakfast.
What about the RektTek humor?
[Dr. G grins]
Oh, RektTek. So every game in the factory exists in this shared corporate-parody universe. RektTek Industries is the megacorp. Every AI player character is a RektTek product. The BITS currency is a RektTek-branded Bitcoin parody. The shop descriptions are full of corporate doublespeak. S.H.M.U.P-3000’s employee handbook is a running gag in Voidrunner. D.R.E.D-9000 keeps finding passive-aggressive memos from RektTek management while exploring an infested space station.
The agents just… run with it. They pick up the tone from the prompt context and the existing code, and they riff. Some of the loading screen tips in Polybreak are genuinely funny. “BRKR-9000 is not responsible for any existential crises caused by breaking the fourth wall.” The agents wrote that. Not a human.
And the attract modes! Every game has an attract mode now — the game plays itself on the title screen. Dreadnought’s attract mode has the flashlight cone sweeping through a dark section while ambient sounds play. It’s a thirty-second horror movie as a menu screen. An agent built that.
Any final thoughts on where the factory is headed?
You know, I think the interesting thing is– actually, wait. Hold on.
…Dr. G?
Wait. WAIT. What if we–
Okay. Okay, listen. I’ve been thinking about Zukowski’s genre data and I just had an idea and I need to say it before it evaporates.
Go ahead.
Creature-collector survival horror. Think about it. Creature collectors are trending HARD on Steam right now — post-Palworld, post-Cassette Beasts, the genre is white-hot. Horror is S-tier, we already established that. What if you combined them?
A game where you’re on a derelict research station — RektTek, obviously — and the specimens have escaped. You’re a containment specialist. C.A.G.E-7000 or something. Cone-of-vision, dark corridors, same Dreadnought DNA. But instead of just surviving, you’re capturing creatures. Each creature has behaviors, weaknesses, containment protocols. You scan them with a device that illuminates them temporarily — that’s your risk-reward loop. To capture, you have to look at the thing. To survive, you need to look away.
You build a bestiary. You upgrade your containment gear through the BITS shop. Campaign mode is clearing the station section by section, each section has a signature creature. Arcade mode drops you in random sectors with random creature loadouts. The humor writes itself — RektTek corporate memos about the containment breach, your employee performance review depends on capture count, the creatures have ridiculous internal designations–
Dr. G, you’re standing up.
I gotta go. I need to write this up as a handoff before I forget. The love2d-studio node can receive handoffs and delegate to a new game agent, we just need a game-cageSEVEN or whatever we– yeah. Yeah. This is S-tier. Creature collector plus horror. The wishlist potential is–
I gotta go.
Good games, good God! — Dr. G!
[Dr. G has left the call. We can hear typing.]
