After Trying Every Coding Agent Setup, What Workflow Actually Survived — and What Does It Cost Per Month?

philzxx · reddit · 2026-08-28

A user deep in the AI-coding-workflow rabbit hole opened a discussion laying out the full decision tree people face:

His goal isn't zero-supervision perfection, but post-spec autonomy with deterministic gates catching most failures — the human shifts from programmer to project manager. For internal tools, prototypes and automations, "good enough really is good enough."

He's candid about the friction: harnesses lock you into providers, model-agnostic ones need more config, skills solve planning but not execution, and orchestration adds yet another layer. His key questions: are people running this spending hundreds to thousands of dollars a month? Does the tiered approach (strong model for architecture, cheap/open-weight for implementation, deterministic gates, strong independent reviewer) actually save money, or do retries eat the savings? He wants the architecture and economics of workflows that survived, not model-vs-model comparisons.

Original post →

More from coding & agent

coding & agent channel →