OpenAI Product Head Tibo: Daily Improvements or Full Usage Reset for 28 Days
Thibault Sottiaux, the core product head at OpenAI, announced that over the next 28 days, the team will either deliver a clear improvement to most Codex and ChatGPT Work users each day or execute a full usage reset. This commitment covers the shared usage quota of Codex and ChatGPT Work, without specifying a list of features or announcing which day will trigger the reset.
Earlier the same day, Sottiaux stated that the team has entered a lock-in phase, focusing only on four types of work: simplifying products, improving efficiency for more usage, breakthrough features, and new models. He also mentioned that user feedback has clearly requested simpler products, and the team will invest ahead of the curve rather than continue stacking complex capabilities.
Sottiaux currently oversees OpenAI's core products and platforms, including APIs, agent infrastructure, enterprise products, ChatGPT (including ChatGPT Work), and Codex, reporting to President Greg Brockman. In May 2026, OpenAI merged ChatGPT, Codex, and APIs into the same core product group, for which he is responsible. He joined OpenAI in 2024 after working at Google on Maps and later at DeepMind on machine learning infrastructure.
Codex and ChatGPT Work share a usage quota, which includes a 5-hour quota and a weekly quota. Resets are categorized into automatic refreshes and banked resets, the latter requiring manual use by the user or expiring. Public tracking records show that since September 2025, there have been approximately 53 to 58 resets confirmed by his posts, with an average interval of about 6.9 days; the most recent confirmed full reset was completed on October 2, 2026.
On October 3, he publicly solicited features still missing from Codex and stated that he is investigating reports from Pro 500 users who did not receive resets as expected, promising to address this. After a brief service interruption on September 26, he announced a reset of usage for all paid users of Codex and ChatGPT Work, mentioning that there was a backup Codex for internal use during the outage. In public interviews, he stated that early resets were used to compensate for outages or degraded experiences, while later resets were used to mark milestones and allow users to try new models, without coordinating with the market or finance departments.
This is a supply commitment aimed at paid subscription users, not a new pricing or financing event. The buyers are developers and white-collar users on paid tiers like Plus, Pro, and Business, while the seller is OpenAI's inference and agent computing power. If improvements are realized, usage efficiency will increase, keeping users subscribed; if no delivery occurs on a given day, a full reset effectively releases already used quotas, temporarily shifting usage pressure back to OpenAI's inference costs. The beneficiaries are high-consuming Codex and Work heavy users, while OpenAI faces pressure from weekly quotas and 5-hour window subscriptions, along with the additional inference volume brought by resets.
At the time of posting, the interaction count was approximately 5042 likes, 907 replies, and 135,697 views. OpenAI has not publicly disclosed the specific reset dates or feature lists for these 28 days in its help documentation; whether quotas are credited will still depend on client status and subsequent completion statements.
Source: Public Information
ABAB AI Insight
Thibault Sottiaux's trajectory has shifted from quota announcer to core product controller. After joining in 2024 and leading the Codex project, the early team consisted of about 30 mixed research and product engineers, with Codex covering almost all software engineers internally and reviewing production code. In May 2026, after Greg Brockman merged ChatGPT, Codex, and APIs into a single core product group, Sottiaux took on unified responsibility for APIs, agent infrastructure, enterprise products, ChatGPT, and Codex. The public controversies are also quite specific: resets have long been tied to personal accounts rather than an official status page, and the community directly refers to quota replenishment as the reset named after him; incidents like Pro 500 missing resets require him to issue compensations afterward.
Capital has not flowed into a new round of external acquisitions but has treated subscription quotas as controllable inventory. Codex initially served engineers who could tolerate faults, then was packaged as ChatGPT Work and pushed to white-collar positions on mobile and web. Resets, in his words, are divided into two phases: early ones compensate for outages and degraded experiences, while later ones provide trial quotas for new models and milestones, without coordination with market or finance. The 28-day commitment frames "feature delivery" and "quota replenishment" as a daily choice, effectively using inference costs to buy retention rather than purchasing share through price reductions.
A parallel can be drawn with Slack treating free message history limits as conversion gates, and cloud vendors using sudden credits instead of lowering directory prices to quell capacity complaints. OpenAI is currently in a control phase, transitioning from coding agents to general work agents: Codex has already become an internal production tool, while Work is responsible for selling the same capabilities to non-engineering positions. The industry position is not about zero expansion but managing demand peaks with quota rhythms after a subscription model has already been established.
Structurally, this represents a transfer of pricing power. While model capabilities remain under OpenAI's control, the price perceived by users is no longer just the monthly fee but a combination of the 5-hour window, weekly quotas, banked resets, and temporary full resets. The mechanism disconnects inference costs from seat fees: when features are not ready, releasing quotas is cheaper than admitting capacity shortages; when features are ready, efficiency improvements can convert the same monthly fee into more tasks. Quotas thus transform from billing attachments into regulators of the product itself.
ABAB News · Cognitive Laws
If features cannot be delivered, then deliver quotas.
Subscription fees are the listed price; quota windows are the real price.
When computing power is tight, resets are cheaper than price reductions.