Codex Adds Usage Breakdown by Task, Subagent and Chat

OpenAI added an Analytics tab to the usage settings in Codex, its coding agent, that shows where a subscriber's plan allowance goes. OpenAIDevs' post says it breaks usage down by tasks, by subagents (helper agents that a main agent starts for parts of a job) and by individual chats. The screenshot in the Codex engineer's post that OpenAIDevs quoted shows the detail: a 7-day or 30-day usage history covering Work, Codex and other agentic tasks, split by feature into categories such as tasks, subagents, auto review, proactive suggestions and an unknown bucket, plus a Top chats table listing each chat's share of the weekly limit and credits used. The screen notes that ordinary Chat conversations are not included. In the screenshot's 7-day view, tasks took 90.6 percent of usage and subagents 4 percent. He says the tab requires the latest version of the Codex app.

The tab follows a promise OpenAI made in late August. Thibault Sottiaux, a Codex leader who posts as Tibo, reset usage limits for paid users after the team found bugs that burned allowances, including goal-driven runs that kept going past their stop condition and smaller models calling more capable helper agents without being asked, and said OpenAI was working on showing users in the app where their usage goes. The breakdown does not stop such bugs, but it lets a user spot a category or a single chat taking an unusual share of the weekly limit and decide which workflow to change.