I'm launching Hive on Product Hunt today: the open-source multi-agent harness that takes you from a one-sentence idea to a merged PR:
producthunt.com/products/hiv…
It was a solid start to the day to learn that multibillion-dollar company (@databricks) is chasing the same idea I built into Hive. At least now I know I'm not alone in this direction. But they built it in a War Room. And I don't have the War Room!
The question my peers ask me the most is why I chose Ruby for Hive. Ruby is a scripting language that was literally made for this kind of job -- working with files. It’s token-efficient, I understand it well, and there are wonderful Ruby bindings for the Crush TUI library
Recently, I had a very bad experience working on a project in a language I didn't know and a framework I didn't admire. My desire to experiment with the new technologies quickly vanished.
Ok, you can add Hive to your OpenClaw via ClawHub and you can select Codex as the main agent for development and review
> openclaw skills install hive-cli
than in openclaw chat run:
/hive setup
-- it will guide you through setup and project configuration process.
Hive is my agentic orchestartor, that runs multiple SOTA coding agents to implement features from idea to PR. Now it can improve your projects without any input from you. github.com/ivankuznetsov/hiv…
I built a tool for Claude Code / Codex that clones my team's best code reviewers from their PR history.
Then I built an eval to check if it actually reviews like them. And the eval told me — twice — that I was confidently wrong. 🧵
Plugin, public benchmark, and the eval harness — all
open source. Full write-up with the data:
ikuznetsov.com/posts/i-clone…
Reproduce it, or point it at your team and argue with me using my own eval.