Orchestrate parallel AI workers to build products from a spec
A developer has a lead orchestrator agent assign modules to sub-orchestrators and workers from a large spec, building a consumer shop, a SaaS and Steam games. ·
Done withClaude CodeChatGPT
- What they did
- Starting from a large spec, a lead orchestrator agent (Claude Fable) reviews it, tracks work in git, and hands well-scoped independent modules, each with its own repository, to sub-orchestrators. Those spawn workers, using cheaper ChatGPT and Google agents. A self-hosted CI/CD and git, plus a custom MCP for swarm communication, run on Hetzner servers using each provider's harness.
- How it went
- A Rust-engine Steam game is nearly ready to release, and the shop and SaaS are planned for year end. A two-hour DeepSeek test burned $40. The author is unsure whether it is truly worthwhile or just burning tokens.
- Worth knowing
- Subscriptions cost about $1k a month combined; test suites run slowly because the lead agent writes huge ones, so a fast self-hosted CI/CD build server matters.
Try it yourself with Claude Code
Here is my spec for [product, e.g. small web shop or indie game]: [paste spec or file path]. Break it into independent modules, write a short brief for each with inputs, outputs and acceptance tests, then work through them using sub-agents in parallel where they don't depend on each other. When all modules pass their tests and are integrated, give me a summary of what was built and what is left, and ask before deleting any files or deploying anything.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: HN · Nov 15, 2026
Related use cases
Turn tickets into merged pull requests
Pulls tasks from GitHub Issues or Linear, then runs Claude Code or Codex in isolated Kubernetes pods to carry each one through to a PR.
Create social media visuals with Canva MCP
The poster tried Canva MCP in Claude Code to produce app marketing visuals and found it a disaster.
Calculate US taxes from 1099s and brokerage statements
A poster saw a GPT Agent handle tax work and built a case study where ChatGPT accurately computed US taxes from 1099s, a self-employed spreadsheet and brokerage statements.
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Build and publish a BaZi chart reading agent
Built solo with coding agents, an agent computes BaZi charts, interprets them, detects language and takes on-chain payments for readings.
Build a first app with no coding experience
A commenter used Replit's agent with Claude Sonnet to make an app in about five minutes despite never having coded anything functional before.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.