Video summary
I Turned My AI Coding Agents Into a Video Game
Main summary
Key takeaways
Overview
An AI coding agent “software factory” is presented as a 3D, game-like environment. The creator’s system runs real coding agents (e.g., Claude Code/Codex, and references OpenCode) that automatically turn GitHub-style work items into code changes and PRs, while a single “CEO agent” orchestrates hiring, task assignment, and coordination across projects.
Core Product / Tech Concepts
Software factory in 3D
- Projects are represented as “floors” in a browser-based 3D world.
- Each agent has:
- a visible live terminal UI
- an “agent browser” for viewing agent details.
Two workflow modes
-
Direct GitHub integration
- Log issues → assign to coding agents → a swarm implements changes → creates pull requests.
-
Paperclip-like approach
- A self-hosted web/CLI interface manages agents (described as less “sexy”).
- Org structure is handled via the CEO-like agent.
CEO agent orchestration
- The user primarily talks to a CEO agent rather than managing sub-agents directly.
- The CEO agent can:
- hire new agents
- manage their workload
- operate across multiple projects
Model / harness configuration
- The app supports multiple coding-model backends:
- Claude Code, Codex, OpenCode
- Users can choose setup defaults such as:
- reasoning effort
- default coding harness
Key Features Shown
Project setup options
- Create new projects
- Link existing GitHub repositories
- Use a local workspace/folder
Backlog ↔ GitHub issues mapping
- New backlog items become GitHub issues.
- Agents:
- update issues with comments
- report progress in real time
- Tasks are auto-assigned to the right swarm agents based on role.
QA agent review loop
- After developer agents complete changes:
- a QA agent tests/reviews
- If problems are found:
- the task is reassigned back to the developer agent.
Auto-merge with safeguards
- Merge requires:
- QA passing
- all checks succeeding
- Merge gating includes:
- GitHub CI/CD checks
- optionally checks from CodeRabbit and Vercel
- Users can disable auto-merge.
- The app can pull the QA report (including evidence like test screenshots) and link it to the corresponding GitHub PR.
Live app preview
- A built-in “monitor” starts a dev server so changes appear immediately in a live website preview.
- Includes a responsive “phone view.”
Mobile / interactive management
- Mobile UI shows:
- who’s working on what
- planning/milestones
- chat with the CEO
- Users can review “candidates” in a lobby/waiting room and approve:
- hire → move to project
Real-time visible agent activity
- Agents can be selected in the 3D space to view:
- the terminal UI
- role/job descriptions
- current work
Build / Implementation Notes (Tutorial-Style)
3D environment generation prompt
- The creator used Opus 5.5 (noted as strong at 3D/game environment work) to generate a browser-based 3D environment.
- The prompt instructed:
- cartoonish first-person 3D
- software-factory behavior
- use of Claude Code + Claude Agent SDK
Cost / time claim for initial build
- Ran for ~1 hour 16 minutes
- Cost about $35 for the initial session/build
Setup instructions
- Open-source project
- “Single command” setup installs dependencies and opens the browser.
- First-run flow includes:
- entering company name
- configuring a CEO agent (name, attributes/rules)
- choosing manual vs automatic hiring
Using the system
- Wait for the CEO to hire initial sub-agents.
- Ask the CEO to create tasks (example: improve visual design / add variety).
- Navigate to project floors to view live agents and see GitHub-backed issues/PRs updating.
Mentioned Additional Learning Resources
- The creator promotes Agentic Labs (Skool) with hours of videos:
- “agentic coding fundamentals,” dev environment setup
- design systems/animations
- projects to follow along
- weekly sessions (coffee Tuesdays, deep dives Thursdays)
Sponsor (Relevant Technical Angle)
- Oracle Agent Studio for Fusion Applications is sponsored:
- agents run inside Oracle Fusion for security/rules/tracking
- developers build using familiar tools (e.g., VS Code/Cursor/terminal/Git + coding agents)
Main Speakers / Sources
- Video creator (main speaker; name not provided in subtitles)
- CEO agent (in-demo persona):
- “Morgan” (mobile/board-controlled CEO)
- “Luna” (example setup name)
- Sponsor: Oracle
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.