Summary
In this short OpenAI customer video, a Notion engineer named Quinn shares his early experience using GPT-6 Astra inside Codex for real engineering work, including cost optimization tasks like improving cache reuse to cut token spending. He describes Astra as feeling less like a smart-but-unclear assistant and more like a ‘staff engineer collaborator’ — a model that can be handed an open-ended, previously unsolved problem rather than a narrow ticket.
Quinn contrasts Astra with prior models, noting that earlier coding agents were good at finding interesting patterns in code but poor at communicating them clearly. He says Astra explains findings crisply, without unnecessary back-and-forth, which shortens the time he spends actively supervising the model. He walks through a concrete example: asking Astra to audit past optimization work and identify gaps the team had missed, which it did by connecting patterns human reviewers had overlooked.
The conversation closes with Quinn’s advice for how engineering teams should prompt Astra going forward — favoring high-level goals like ‘get this metric one nine closer to its target’ over granular step-by-step instructions. For engineering leaders evaluating how frontier coding models change day-to-day workflows, it’s a brief but concrete look at GPT-6 Astra in production use at Notion.
📺 Source: OpenAI · Published September 21, 2026
🏷️ Format: Workflow Case Study
