Astra admits quality regression: experiment hit 4-5k users, three root causes fixed
TheMoonMidas · x · 2026-09-12
Thibault Sottiaux posted a reset and postmortem for recent Astra quality complaints, citing three fixed root causes:
- Mis-firing legacy skills written for earlier models triggered too often or stopped the model from checking its own work;
- An opt-in context management experiment caused early stops and replies to stale messages, affecting an estimated 4,000-5,000 users — now disabled;
- Badly configured engines degraded quality for a long tail of traffic — removed.
The team says the experience is now significantly better: more consistent follow-through, better tracking of the latest message, and stronger self-checks during execution.
Related event: GPT-6 Astra Suspected of Silent Downgrade, Fixes Announced(8 posts)→
More from coding & agent
- The benchmarking problem: we're evaluating adapter-harness-model triplets, not models — teortaxesTex · 2026-09-12
- Git AI Duo Joins OpenAI as Company Pledges to Keep Agent-Tracking Tool Open Source — Dimillian · 2026-09-12
- Tasking an AI agent to buy 4 Tokyo concert tickets: the 'Math Olympiad' of assistants — annbordetsky · 2026-09-12
- AI Agent vs. Customer Support Agent: The Negotiation Future of Codex Computer Use — ethanniser · 2026-09-12
- Sandbox latency compounds into a hidden infra tax in coding-agent RL, Daytona intern finds — tw_killian · 2026-09-12
- Notion AI Meeting Notes now drafts email recaps as users hand follow-ups to agents — ivanhzhao · 2026-09-12