What If AI Coding Agents Worked Like CI Pipelines?

odyth — Wed, 06 May 2026 22:10:03 +0000

Most AI coding tools have a weird hidden cost:

They use an LLM to decide what another LLM should do next.

That means you are spending tokens on routing, not just on the actual coding.

I built RedQueen to flip that model.

What is RedQueen?

RedQueen is an open-source, self-hosted orchestrator for AI coding agents.

Instead of asking an LLM, “what should happen next?”, RedQueen uses a deterministic state machine to move work through a software development pipeline:

spec → code → review → test → human review

No token spend for orchestration.

No mysterious agent loop.

No “trust me bro” routing logic.

Just a debuggable workflow you can read, edit, and run yourself.

Why this matters

Autonomous coding gets messy when one giant prompt is responsible for everything.

RedQueen breaks the work into isolated AI workers:

one worker writes the spec
one writes the code
one reviews the code
one tests it
humans approve at configurable gates

Each phase has a focused job.

And the workflow itself lives in config, not vibes.

Want to add a security review step? Add it.

Want to remove spec review? Remove it.

Want the agent to stop and wait for a human before touching production code? Make that a gate.

The opinionated part

I do not think the future of AI coding is one giant autonomous blob.

I think it looks more like a tiny dev team:

focused workers
explicit handoffs
retry limits
human review gates
issue tracker sync
source control integration
boring, readable orchestration

Boring is good.

Boring is how you debug things at 2 a.m.

Quickstart

npm install -g redqueen
redqueen init -y
redqueen start

Open the dashboard, label a GitHub issue, and watch it move through the pipeline.

RedQueen currently dispatches Claude Code workers and supports GitHub Issues and Jira, with the adapter pattern designed so more integrations can be added cleanly.

Why I made it

I wanted AI coding agents that felt less like slot machines and more like build systems.

A build system does not spend tokens deciding whether to compile.

A CI pipeline does not ask an LLM whether tests should run.

So why should an AI coding workflow?

RedQueen is my attempt at making autonomous coding more:

deterministic
inspectable
self-hosted
configurable
human-controlled
cheap to orchestrate

It is still early, rough around the edges, and very much a v0.1 preview.

But the core idea is simple:

Stop orchestrating with LLMs.
Start commanding them.

Check it out here:

https://redqueen.sh