OpenAI Codex team builds with minimal specs, 8-week horizons, and designers shipping code¶
Insight: OpenAI's Codex team operates without medium-term roadmaps: ~10-bullet specs, 8-week planning horizons, and designers writing and shipping code via parallel agent workstreams instead of sequential handoffs to engineering. The argument is that conventional planning cycles cannot keep up with how fast underlying model capabilities shift, so the team optimises for short loops and direct authorship.
Detail: Corroborates existing entries in this file on spec-first development and the planning bottleneck in agentic teams. Single first-party account from a vendor team about its own product workflow — credible as a description of practice but potentially self-serving as a generalisable recipe.