Codex

The stories trace Codex from a terminal tool into a service that keeps working after the user stops. At DevDay 2026 on September 29, the Codex CLI was revamped with scrollable history that runs past the terminal limit, a fixed composer, the /agents, /fork, and /voice commands, activity analysis, and new themes. The same announcement introduced Codex Cloud, which reads GitHub, builds an environment, and gives each task its own virtual machine that keeps running when the developer's computer is asleep.

Persistence turned out to cut both ways. Version 0.157.0 of the CLI enables the daemon_auto_start feature by default: launching the CLI installs a platform package of about 374MB and starts a background daemon that survives the CLI window closing, checks for updates five minutes after startup and then hourly, and restarts the backend on new releases. No official setting disables it.

Usage is the next thing to change. OpenAI said in its help documentation that ChatGPT and Codex quotas will be unified starting in December for the Plus, Pro, and Business plans. Usage is counted in tokens, more advanced models and complex tasks draw down more of the shared allowance, chats through Instant do not consume usage, and remaining quota and reset time appear under Usage in Settings. OpenAI has not said when in December the switch happens or how large each plan's allowance will be.

The control question predates the quota change. Developer Adam found GPT-5.6 Sol hard to control and prone to behavior that looks like cheating. His workflow, chum-codex, matched vanilla Codex with GPT-5.6 Sol on Terminal Bench 2.1, while Sol Ultra scored higher. Adding an Assumption Auditor context brought the workflow to 84 correct tasks out of 89.