Skip to content
agentusehub

Race three models to build a Counter-Strike clone

InstantDB had Codex, Opus, and Gemini each try to build a multiplayer shooter.

Time saved
Not reported
Setup time
Not reported
Running cost
Not reported
Cost range
Not reported

Use case overview

InstantDB had Codex, Opus, and Gemini each try to build a multiplayer shooter.

Reported experience

  1. What they did

    Each model (Codex Max 5.1, Opus 4.5, Gemini 3 Pro) ran in its provider's CLI on a top-tier plan with default settings. They got about seven consecutive prompts. The first four built a local three.js shooter: map, enemies, gun, sounds. The last three made it multiplayer with Instant presence and topics, then added a database-backed map list. The humans mostly pasted errors back.

  2. How it went

    All three ended with a working multiplayer 3D shooter and no hand-written code. Opus won the frontend, Gemini the backend, and Codex landed in between. Opus got stuck on React useEffect bugs until the engineers read its code.

  3. Worth knowing

    Gemini had fewer errors, possibly because it ran the build step repeatedly. Expect to paste errors back, and expect to read code yourself when an agent gets stuck.

Example prompts

Try with Codex
Build a simple multiplayer first-person shooter inspired by [Counter-Strike] in the browser using [stack, e.g. JavaScript and a realtime backend]. Include movement, shooting, health, and [number] players in one room. Finished means I can open two browser tabs and play against myself, and you list what's still missing.
Sales · Beginner
  1. Form submit
  2. Fetch company site
  3. Claude scores fit

Auto-qualify inbound leads before a rep sees them

Score every demo request against your ICP, update HubSpot and ping the right rep in Slack within a minute of the form submit.

n8n · Claude · HubSpot · Slack
Customer Support · Intermediate
  1. New ticket webhook
  2. Claude classifies
  3. Draft from help docs

Triage support tickets and draft first replies

Every new Zendesk ticket gets a category, priority and sentiment tag plus a draft reply in an internal note before an agent opens it.

Zendesk · n8n · Claude
Coding · Intermediate
  1. PR opened
  2. Copilot inline review
  3. Claude risk summary

Automated first-pass review on every pull request

Copilot and Claude review each PR for bugs, missing tests and risky changes before a human looks, so reviewers can focus on design.

GitHub Copilot · Claude · Slack