Hey Product Hunt 👋 I kept hitting the same wall with AI coding agents: either I was babysitting them line by line, or I'd hand over a big task and get back a PR that quietly missed the point. Neither felt like leverage. So I built Son of Anton around a simple rule: AI should do the implementation, but a human should own the decisions. There are exactly three of those — what to build, how to break it down, and whether it's actually done. Those are real stops. Everything between them (worktrees, implementation, verification, an adversarial review pass, stacked PRs) the orchestrator runs on its own. A few things I'm proud of: • It installs as a git subtree — real tracked files in your repo, no npm package, no submodule, nothing leaving your machine. • An adversarial subagent reviews every ticket assuming the first pass cut corners, and hard-blocks the PR if the review ledger disagrees with git history. • It's agent-agnostic — Codex, Cursor, Copilot, Claude, OpenCode, anything that reads AGENTS.md. Swap mid-project if you want. • Bonus: it integrates with Codogotchi, so a little desktop pet reacts to each gate event as your pipeline runs. It's free and open source. Would genuinely love your feedback — especially on the three-gate model. Does it match where you actually want to stay in control?

Son of Anton

AI should do the implementation. You own the decisions.

AI should do the implementation. You own the decisions.