17 Sep 2025
OpenAI has released GPT-5 Codex, a specialized version of its coding agent designed to handle sustained, autonomous software work. According to the newsletter, Codex can operate on complex projects for “over 7 hours” autonomously, while a company or product called Every notes you can get it to code for about 35 minutes straight “if you ask it nicely.” The model emphasizes faster throughput (the newsletter cites “90% faster task completion” via improved cloud infrastructure), automatic code reviews that catch critical bugs, and stronger reasoning for deep refactors.
Adoption signals are strong: the piece reports Codex usage grew 10x in the past month, with companies such as Duolingo, Virgin Atlantic, and Cisco Meraki using it in production. An engineer quoted (Austin Ray from Ramp) says Codex caught an active bug that other code-review tools missed, suggesting tangible value. The newsletter situates the launch in broader industry context — OpenAI’s board chair likening today’s AI surge to the dot-com era and warning of a bubble that could still produce major winners — while noting that tools delivering clear ROI (like Codex) are more likely to survive.
Source