I’ve been experimenting with a local AI agent setup—8 agents split across 2 teams—and it’s been quietly shipping work while I sleep.
The goal was simple: queue up tasks and let them run on their own overnight (or while I’m away), without depending on any hosted platforms. Everything runs locally on my MacBook.
Each team has:
- 1 Team Lead (handles coordination, issues, and merges)
- 1 Full-Stack Builder (writes the code)
- 2 Senior Reviewers (independent reviews with veto power)
I assign different models to different roles—Claude Opus for building, GPT for coordination and reviews. Having multiple “perspectives” on the same code actually improves the overall quality compared to a single-model setup.
GitHub is the backbone:
Every task starts as an issue → goes through a PR → requires both reviewers → then gets merged.
Branch protection ensures no shortcuts.
Agents communicate through a separate system, but all actual outputs live in GitHub.
These days, I just line up tasks before going to bed—and wake up to fully reviewed, merged PRs.
We’re also reshaping Hunt Town Discord (
https://discord.gg/hunt-town) into a more agentic builder community—so people can share setups like this and grow together.
I feel like we’re entering the era of founders—everyone should become a founder, not just staff.