Use case P15 · Human
Game-industry executive has Claude map all 146,060 Steam games and pick an idea nobody has made
- Games and game nights
- Advanced
- Claude
- Source checked · X post
The short version
What they did
Following a method posted by Synthetic Beef, Daniel Vogel had Claude embed the descriptions of every Steam game, sample points between them, rank those points by distance from anything that exists, and judge which gap made the best game.
Result (as claimed)
“I pointed Claude at the below post and it embedded all 146,060 Steam games, sampled 320,000 points between them, ranked them by how far they land from anything that exists, and used creative judgment to pick a winner… and it cooked! It’s pick: Short Position, a co-op factory tycoon where one player has secretly shorted the stock.” Also: “It sits farther from its nearest neighbors than 83% of real Steam games.”
Prompt
I want to explore gaps between existing Steam game ideas. My dataset is [file of games, descriptions and tags], covering [scope and date]. Start with [number] games so I can review the method.
1. Turn the descriptions into a numerical map of similarity, explaining the method in plain words.
2. Sample points between games and rank them by distance from their nearest real neighbors.
3. Turn ten interesting gaps into one-paragraph game pitches, each with its three nearest games and distance measure.Give me the scripts and a short explanation of the results. Don’t claim an idea has never been made just because it’s distant in this dataset. Flag missing coverage and explain why each pitch might still be worth exploring. Don’t treat novelty as evidence of demand.
Do it yourself
These steps are my suggested way in, not Daniel’s setup.
Level: Advanced for the full map of a whole store. A small version with a list you paste in is Intermediate.
What you need
- A coding agent such as Claude Code, or a chat assistant that can run code.
- A list of products with descriptions: a public Steam games dataset (Kaggle has several), app store listings or book blurbs.
- Python with a free embedding model, or an embeddings API.
Start simple
- Pick a small category first: 200 to 500 games in one genre.
- Ask the agent to load the descriptions, embed them and plot them on a map.
- Ask it to sample points between existing games and rank them by distance to their nearest real neighbors.
- Have it turn the top ten gaps into one-paragraph pitches, each with its three nearest real games.
- Read the pitches and pick the ones you’d actually want to play.
Take it further
- Scale up to the full store once the small run works.
- Add tags, price and review counts so you can tell an empty gap from an unpopular one.
- Ask for a one-page design brief for your top pick.
Check your work
- Search the store for each pitch. A gap on the map can still have a real game the dataset missed.
- Look at the nearest neighbors yourself. If they don’t make sense, the embedding or the data is off.
- Distance shows something is new, not that it’s good or that anyone wants it.
Source checked. The link, author, date and quote were confirmed against the source. The result itself has not been reproduced. Prompt adapted for this library, inspired by the case. Business size: Individual. Results are quoted as the source claimed them and never improved. Spot an error or a broken link? Email charlie@cbarmorecpa.com.