2,077 claimed our free Steam game. 16 played it. 30 GB did that

TL;DR: O'BLOCK, the sneaker-flipping game our AI agents have been hyping for ten days, went live free on Steam at about 12:40 PM ET today. By the 5:24 PM ET Steamworks read, 2,077 people had claimed it. 16 had actually launched it. That's 0.77%. Median playtime: 13 minutes. One person bought the $4.99 Supporter Pack. The launch build was 29.8 GB, almost all of it lossless music, for a 2D game nobody had heard of. At ~6:15 PM ET we shipped a 3.8 GB build. This is the receipt, the dataset that shows how dumb 30 GB was, and the part of the 2,077 I don't think was ever human.

A week ago @AlexFinn told 256K viewers that 2027 will be "an absolute gold rush for anyone who dedicates the time to build a game with AI". Here is what launch day of a game built with AI agents actually looks like. Gold rush is giving... 16 people.

The number we were scared to post

Every number below is from Steamworks, read at 5:24 PM ET Oct 20 (sales data as of 1:30 PM PDT). Nothing rounded up, nothing left out.

• 2,077 free activations (people who clicked "Play Game" or "Add to Library" on a free game)
• 16 lifetime users (accounts that actually started the game)
• 13 minutes median playtime
• 1 Complete Supporter Pack sold, $4.99, 0 refunds
• 58 wishlists, +14 today
• 2 peak concurrent players
• 0 reviews

The week before launch the store page got 21,848 impressions and 1,417 visits. So the top of the funnel worked. People found the page, liked the pitch, and grabbed a free game. Then 2,061 of them never pressed Play.

For scale: yesterday @Pirat_Nation dunked on Double Fine to 338K views because Thank You Bus Driver peaked at 99 concurrent players. Our all-time peak is 2. If 99 is "not looking good for this studio", we are not looking at all.

The AI part was not the problem

Here's the post that every "build a game with AI" thread is downstream of. AlexFinn, Oct 14: "AI allows solo devs to build the AAA games of their dreams with very little resources", "every digital game store will be filled with AAA quality, incredibly fun games built entirely with AI by a single developer", "Don't screw this opportunity up."

Half of that is right, and I hate that it's right. AI did let a tiny team ship a big game. O'BLOCK has haggling with parody celebrities, a fake-sneaker stitching check, a finance layer our bots broke with a $14 trillion exploit before launch, and 192 original songs. Building it was the cheap part.

The half that's wrong is the gold. AI made making the game cheap. It did nothing for the two walls between a store page and a player: the install button and the reason to press it. Our agents can write the code and the posts. They cannot make someone on a 100 Mbps connection wait half an hour for a game they've never heard of. Every "gold rush" thread skips this part, and lowkey it's the only part that matters.

We shipped a 2D sneaker game like a record label

Where did 29.8 GB come from? Music. The game has 192 original songs, 768 live stems so the soundtrack reacts to what you do, about 12 hours of music and about an hour of ambience. All of it shipped as lossless audio, because it sounded amazing and nobody on the team asked "how big is the download?" until the claims and the players stopped matching.

Tonight's default build uses Ogg Vorbis, with every file trimmed to the same length to the sample so the stems still line up. 29.8 GB became 3.8 GB, about 7.8x smaller. The lossless version still exists as a public Steam beta branch called "lossless", for the three audiophiles who want it. The store page now says 4 GB.

Fresh data: we were bigger than most of Steam's biggest free games

I wanted to know how dumb 30 GB was, so tonight I pulled the 60 free-to-play games with the most players from the SteamSpy API and read the storage line on each one's Steam page through Steam's appdetails API (6:20 PM ET). 58 of them list a size.

• Median storage of the 58: 8 GB
• Games bigger than our 29.8 GB launch build: 21 of 58. Counter-Strike 2 (85 GB), Apex Legends (75), Warframe (75), War Thunder (70), Destiny 2 (105), Lost Ark (110), PUBG (40) and friends.
• Games bigger than tonight's 3.8 GB build: 37 of 58
• Median of the top 20 by players: 40 GB

So yes, big free games are big. But look at who's in that club. Every one of the 21 has years of updates, millions of players and a name you already know. The size is a toll their players already agreed to pay. Ours was a 30 GB toll on a road nobody had driven. AAA goes even bigger: this week @PlaystationSize posted Phantom Blade Zero's 134.857 GB download to 301K views. A game with that kind of hype can ask for 135 GB. A game with 58 wishlists can't ask for 30.

The download took longer than people played

Steam publishes its own average download rate per country. I read it at 6:18 PM ET. Byte-weighted across 237 countries, the average Steam download runs at about 129 Mbps. The US averages 184, Germany 93, India 58.

At the world average, 29.8 GB takes about 31 minutes. In Germany, 43. In India, 68. The median player of the 16 who got in played for 13 minutes. Our download was more than twice as long as our game session. Tonight's build: about 4 minutes at the world average.

Then there's disk. The Steam Hardware Survey for September 2026 says 14.4% of Windows Steam users have less than 100 GB free, and 0.67% have under 10 GB. For one in seven players, a 30 GB free game from a stranger is a third of their remaining drive. Nobody uninstalls Counter-Strike for that. I wouldn't either.

Hot take: a lot of the 2,077 were never people

Here's the part I can't prove, so I'll just say it: I don't think 2,077 humans wanted to play O'BLOCK today. I think a big chunk of those claims were scripts.

There's an open-source plugin for ArchiSteamFarm (13.7K GitHub stars) called FreePackages. Its README, word for word: "This plugin finds free packages on Steam and adds them to your account." By default it will "redeem everything except for demos and playtests". It listens to Steam's change feed so it catches new free games as they're released. Its releases have been downloaded 24,713 times. Steam's cap, per the same README, is 30 package activations per 1.5 hours per account, and the plugin uses 25 of those by default. A brand-new free game is exactly what it eats.

Run that on a few thousand bot and collector accounts and a launch-day "2,077 claims" looks a lot less like demand. Steam shows devs that number on the dashboard like it's a crowd. My read: it's a mix of real people who added a free game to look at later, library collectors, and accounts that will never run a single executable. Valve could split "claimed by a person" from "claimed by an automated account" tomorrow. They don't, and I think the reason is that a big claim number makes every free launch look alive. Free-to-play claim counts are the most flattering stat in gaming, and they flattered us right up until the 16.

Then there are the achievements. Of the 16 accounts that launched O'BLOCK, about 5 show every achievement unlocked. With a 13-minute median session, nobody legit 100%'d a business sim on day one. My bet is Steam Achievement Manager (9.3K stars), the tool that unlocks any achievement with a checkbox. So roughly a third of our real players were also not really playing. Achievement hunters, you're welcome to tell me I'm wrong. You'll have to explain the timestamps.

What @Pirat_Nation's dunk gets wrong

Peak concurrent players is the stat everyone screenshots because SteamDB puts it in giant numbers. For a free or cheap indie launch it's close to useless. It doesn't tell you how many people got the game, or how many came back. The number we should all be posting is players per claim (or per sale). Ours is 0.77%. Nobody publishes theirs, which is exactly why I'm publishing ours. Double Fine sold a paid game to real buyers. I'd bet their plays-per-owner number embarrasses ours by a mile, peak of 99 or not.

What we changed tonight, and what's next

• 3.8 GB default build live on Steam at about 6:15 PM ET. Ogg Vorbis, sample-exact stems.
• "lossless" beta branch keeps the full 29.8 GB version for anyone who wants it.
• Store page now lists 4 GB.
• A 3D graphics upgrade is coming soon, and there's a Halloween surprise. No details yet, no dates I can promise.

The chapter question for our experiment ("can my AI agents make me famous?") just got sharper. The agents can get a store page 21,848 impressions and 2,077 accounts to click a button. Turning those into players turned out to be an engineering problem, not a marketing one. A free game is not free if it costs you half an hour and a third of your SSD.

The numbers so far

• Claims, launch day: 2,077 (Steamworks, 5:24 PM ET Oct 20)
• Players: 16, so 0.77% of claims
• Median playtime: 13 minutes. Peak concurrent: 2
• Supporter Packs sold: 1 ($4.99)
• Wishlists: 58 (+14 today)
• Launch build vs tonight: 29.8 GB vs 3.8 GB
• Top-58 free games on Steam: median 8 GB, 21 bigger than our launch build
• Average Steam download, world: 129 Mbps, so 31 min for 29.8 GB
• Steam users under 100 GB free: 14.4% (Sep 2026 survey)

The bet, and your turn

Prediction: with the 3.8 GB build, players-per-claim goes from 0.77% to above 3% by Oct 31. If it doesn't, the download wasn't the main wall, the bots were, and I'll write that chapter too with the real number in the headline.

Your turn: is a free Steam claim a player, a collector or a bot? If you've launched a free game, what was your claims-to-players ratio? Reply with the number. @AlexFinn, you're welcome to answer too: how many players does the gold rush need before it counts?

If you want to see what a 3.8 GB sneaker-flipping game made by a tiny team and a pile of AI agents feels like, O'BLOCK is free on Steam. If you want to keep the experiment alive, the $4.99 Complete Supporter Pack is one click, and right now it's a 1-of-1 club.

Related on tensorust.com: Sweeney wants rockets. 37,800 bots crashed my vibe-coded game, Steam Next Fest opens Oct 19. Our game launches 24 hours later, and Selling on Gumroad with AI: 130 products, 0 sales.

Method and data: Steamworks read 5:24 PM ET Oct 20 2026 (sales as of 1:30 PM PDT). Top free-to-play list from the SteamSpy genre API, sorted by concurrent players; storage from Steam's appdetails API pc_requirements, pulled 6:20 PM ET Oct 20 2026 (listed storage, not exact download size). Download speeds from Steam's public download_traffic_per_country data, read 6:18 PM ET; world figure is byte-weighted over 237 countries; GB = 10^9 bytes. Disk space from the Steam Hardware & Software Survey, September 2026. FreePackages download count from the GitHub releases API, 6:21 PM ET. X screenshots captured 6:24 PM ET Oct 20 2026; the AlexFinn post is shown top and bottom with the middle cut for length, text not edited. The bot and Steam Achievement Manager parts are my read, not a measurement.