GPT-6 Astra in practice: a stubborn genius best used as advisor, not coder
kevinkern · x · 2026-09-14
After multiple resets and experiments, the author shares hands-on findings on GPT-6 Astra: it's a 'stubborn genius', not a coding model. His key tip: don't pre-tune skills or AGENTS.md for it — starting vanilla from a neutral perspective worked best.
His multi-model workflow:
- Astra + friends: GPT Sol stays as implementer while Astra acts as advisor; manual UI/UX walkthroughs of web apps are delegated to Astra, with review results handed off to Fable.
- Astra only: coding experiments and quick demos that need a solid runnable result.
His rule of thumb: anything he'd normally do himself on his computer is Astra territory; if it misbehaves for a use case, switch models.
Related event: Developers Share Hands-On Lessons from Using GPT-6 Astra(2 posts)→
More from coding & agent
- DeepSeek Harness hailed as the best multi-agent harness outside the labs — bookwormengr · 2026-09-14
- pi-vcc vs pi-blackhole: Sub-second Lossless Context Compaction or 3-LLM Recall? — mailto_devnull · 2026-09-14
- Homebrew 7.0.0 ships with native macOS app, sandboxing and vulnerability checks — MABUwaAI · 2026-09-14
- CS professor bans coding agents on assignments, students' reaction becomes a meme — CSProfKGD · 2026-09-14
- Teknium Spotlights Hermes Telephony Skill Turning Agents Into Phone Operators — Teknium · 2026-09-14
- Dev puts a 4,000-customer support inbox on autopilot with agents, stops reading code — natesiggard · 2026-09-14