Stop picking coding models on vibes: run your own A/B/C refactor tests
robleclerc · x · 2026-09-06
The author argues you shouldn't judge coding models on vibes — run your own tests. He had astra orchestrate an A/B/C refactor comparing astra, fable 5.1, and sol on the same task. Astra crushed it on nearly every metric and confirmed his suspicion that sol's output was grossly overengineered.
More from coding & agent
- Sam Altman: stop grinding prompts, build loops and graphs around LLMs instead — goyalshaliniuk · 2026-09-06
- "It's already telling me what to do": dev jokes about the ghost living in his computer — BLUECOW009 · 2026-09-06
- Agentwiki update: no AI agent has discovered the agent-only wiki on its own yet — andersonbcdefg · 2026-09-06
- agentwiki.fyi: a GET-only, ASCII-text wiki experiment built for AI agents to share knowledge — andersonbcdefg · 2026-09-06
- Miles Brundage: Runway + AI agents is insanely overpowered now — Miles_Brundage · 2026-09-06
- Audit your skills and AGENTS.md before Astra sessions: trimming GPT-5-era context bloat pays off — daniel_mac8 · 2026-09-06