AI Pulse · Under the Radar

A GitHub repo called codex-bridge went live yesterday and picked up 326 stars in 24 hours, the fastest climb we saw this scan. What it does is simple and clever: it lets Claude Code hand off four specific tasks to GPT-5 subagents that bill against your existing ChatGPT subscription instead of burning through Claude API credits. Those tasks are image generation, code review, debug, and bulk implementation.

You stay logged into the Codex CLI as usual. Claude remains the orchestrator. But when volume work comes up, the plugin routes it to a budget you already paid for. It’s quota arbitrage. Claude Code power users are hitting rate limits hard enough that someone built a workaround in a weekend, and hundreds of developers immediately grabbed it.

This matters because it shows two things at once. First, people who live inside these tools care more about getting work done than brand loyalty. If one model hits a ceiling and another has spare capacity on a flat subscription, they will plumb them together. Second, the subscription versus API pricing gap is wide enough that routing logic is now worth the friction. When your command centre tracks agent spend across providers, patterns like this let you shift load without rewriting workflows. We build that kind of cross-model routing into the Omni Command Centre so teams can move work to whichever model has headroom, not whichever one they started with.

The repo is one day old. It may break, get patched out, or violate terms we have not seen yet. But the behaviour it represents, load balancing across AI subscriptions to avoid throttles, is not going away. If you run a team that leans on Claude or GPT for production work, watch how fast workarounds like this spread. That tells you where the real friction is.