Solo dev asks how to route coding tasks across Grok, Mistral and local Qwen
newsbypenny · reddit · 2026-10-12
A self-taught solo developer building a news/publishing project asks Reddit for help designing a multi-model AI coding workflow. He currently juggles Grok, Google Antigravity, Mistral and Xiaomi's MiMo, and wants:
- Task routing rules: repo-level agent rules that decide which provider, model and reasoning level handles each task — Mistral Large 4 for general work, Mistral's FIM model for completion, a local Qwen for trivial jobs.
- Repo governance: strict agent-enforced library allowlists, GitHub Actions deploys, lint and vulnerability checks pre-push, and a local skill that strips secrets before commits.
- Org-wide standards: a master repo for his nonprofit that auto-applies all standards to new repos.
- Plus auto-generated commit docs and changelogs via Mistral, polished READMEs, and Docker/Cloudflare Workers testing — while wrestling with self-hosted annoyances like GPU-based AV1 video conversion.
After hundreds of failures over a year, his newsroom project is finally working and he wants to systematize the workflow. The post is mostly goals rather than proven solutions, but it's a clear snapshot of a solo dev's real multi-model routing pain points.
More from coding & agent
- Two-week course teaches you to build an agentic software factory you can trust — hugobowne · 2026-10-12
- SoftwareGym synthesizes 182 app environments, lifting CUA OSWorld Pass@3 to 56.97% — Tian Jin · 2026-10-12
- Claude created a mysterious 30GB 'blind' directory no agent supposedly reads — Sauers_ · 2026-10-12
- AWS+HSBC paper co-evolves agent graders from just 10 labeled examples, matching ground truth — rohanpaul_ai · 2026-10-12
- Haiku 5.5 costs 75% less than 4.5, but shines only as a subagent under Opus 5.5 — Scobleizer · 2026-10-12
- The sneakiest AI coding failure: looping quietly while sounding productive — chich2017 · 2026-10-12