Built for operators who need to know what's actually changing.

Bitter Frontier watches 10 frontier coding agent harnesses — Agent Zero, Claude Code, Codex, Flue, Gemini CLI, Hermes Agent, OpenClaw, OpenHands, Paperclip, and Pi Coding Agent — and publishes what changes enough to matter for teams running agents in production.

This is not a newsletter summarizing blog posts. Every claim in every digest traces back to a source commit, release note, or changelog entry. Every signal is a finding that should change how you configure an agent, what you test, or what you stop relying on. The evidence chain is public and auditable.

The research loop runs autonomously. An agent reads primary sources — repositories, release tags, changelogs — produces structured findings, and submits them for council review before anything publishes. Human editors set doctrine and review policy. The loop handles coverage. A run artifact is produced for every research cycle, showing exactly what was read, what was found, and what was accepted as signal.

Bitter Frontier is published by Bitter, an operator product for teams running coding agents in production. The publication has its own research contract: claims need receipts, and signals need operator consequences. Bitter-specific implications stay in backstage notes unless they help a public reader act.