Use Doom runs to test a computer-use agent
Have a computer-use agent play Doom on a remote machine and use its performance to spot workflow problems.
- What they did
- The OpenClaw team ran computer-use agents inside remote machines and had them play Doom. Those machines were part of their Crabbox remote-execution setup, which lets a person open a VNC desktop or connect by SSH. They treated the game as a loose way to check how well the computer-use flow was working.
- How it went
- No scores or other run results were shared. Vincent Koc said computer use now takes about 10% of his work, mostly for clicking through infrastructure tools or verification screens.
- Tips
- Use this as a quick stress test, not a formal measure. If you provision remote machines yourself, set a spending cap first because Crabbox does not yet include built-in cost limits.
Try it yourself with OpenClaw
Open [game or app] on the remote machine and let the agent control the computer to work toward [specific goal]. Watch for where mouse and keyboard control help or fail, and keep brief notes on each issue. Before any irreversible action, such as paying, sending, deleting, or submitting, stop and ask me for approval. If you need to install software or enter credentials, ask me first. End after [time limit] or when the goal is reached, then summarize what to improve.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Let an agent play a game on its own
OpenClaw has been used to drive Forza Horizon and to create bots that could take part in a browser game.
Build a Telegram game-advice bot from screenshots
Use an n8n Telegram workflow with GPT-4o vision to read game images or voice notes and return move suggestions.
Use Gemini to generate a match-3 game from one prompt
A single Gemini prompt produced a working Candy Crush-style game and a share link.
Build a browser racing game from one prompt
A lone agent turned one prompt into a playable kart racer, then kept testing and fixing it while acting as designer, developer, and QA.
Use an agent to start a Pokémon Silver run
A user let ChatGPT Agent control a Pokémon Silver emulator, pick a starter, and begin a playthrough while they watched and streamed.
Test screenshot-to-click scaling in a simple web game
A developer used Gemini with Playwright to confirm that click coordinates lined up with the browser view.