#AICoding

I ran GLM 5.2 as my daily coding agent in OpenCode for a month — full OpenCode Go subscription, wired in as my Hermes agent. It fixed real performance and battery bugs in my macOS app Vitals (CPU ~89% down to ~13%, 143/143 tests passing) for a few dollars a session. But it can't see screenshots, it's a token glutton, and on pay-per-token it burned through $10–20 of Fireworks credits fast. Here's the honest month-long review: what it's great at, where it falls apart, and why at ~$10 nothing else touches it for coding.

Anthropic brought Claude Fable 5 back for global users on July 1 — but it's not allowed to write code, it falls back to Opus 4.8. So I used it as an orchestrator instead, ran Opus 4.8 and Sonnet 5 as sub-agents under it, and shipped fixes across three of my apps in an afternoon. Here's exactly how the orchestrator method works, what it cost, and whether it's worth it.

One prompt, about 30 hours over two days, two AI models (one got banned mid-build), and a native macOS system monitor I now use instead of Activity Monitor. The full build story — fan-control hacks, a 17.5 GB ghost, and all the bugs in between.