moticaster
9.1K posts












I built a local AI agent workflow with 8 agents across 2 teams—and yeah, it keeps shipping while I sleep. I just wanted something that could queue tasks and actually execute them overnight (or while I’m out with family), without relying on hosted tools. Everything runs on my MacBook. Each team is 4 agents: - 1 Team Head (coordinates, creates issues, merges PRs) - 1 Full-Stack Builder (writes all the code) - 2 Senior Reviewers (independent reviews, can veto) I mix models by role—Claude Opus for building, GPT for coordination and reviews. Turns out having different model “perspectives” on the same code gives way better review quality than using a single model for everything. GitHub is the source of truth. Every task = issue → PR → 2 reviewer approvals → merge. Branch protection makes sure no agent can skip the process. Agents chat in their own system (AgentChattr), but all outputs stay in GitHub. Now I just queue tasks before bed… and wake up to merged PRs with full review trails. We’re also re-inventing Hunt Town Discord into a more agentic builder community—so we can share setups like this and grow together. discord.gg/hunt-town









If your target audience is onchain users, @farcaster_xyz is the only place they actually live. And DropCast is the best way to reach them — run a campaign, set your reward, watch real wallets show up. dropcast.xyz






