OpenAI Codex lead promises a daily improvement or a full quota reset for 28 days

On October 5, QbitAI reported that Tibo Sottiaux, the OpenAI executive in charge of Codex, said on X that over the next 28 days he will make a daily either/or choice: ship an improvement clearly useful to most Codex and Work users, or trigger a full usage-limit reset.

The word "improvement" is deliberately loose. It could be a new feature, or simply a bug fix. Either way, the arrangement gives users a floor: if something improves, they get to try it; if nothing does, their quota resets.

The report notes that this is not the first time Sottiaux has framed things this way, and that the either/or has become a recurring formula for him. In recent weeks he has been asking users what Codex still lacks, then narrowed the work to four directions: simplifying the product, improving efficiency so it supports more usage, breakthrough features, and new models.

The pledge also drew cross-company banter. Lauren Tan of Grok Bot mocked OpenAI for shipping "slop" and relying on quota resets to hold its audience, while Sottiaux shot back that she should do better herself. The exchange read partly as mutual audience-building.

Beyond the noise, users' frustration is not only about running out of quota. The article groups the complaints into three areas. First, DevDay shipped more than 20 updates at once, so many that the focus was hard to find, and the one highlight, the personal agent Dot, currently looks like it may have fallen short.

Second, resets were meant to serve real work by hiding the moment quota runs dry, but today's frequent and unpredictable resets make it hard to plan a working schedule in advance. On top of that, quota tiers keep shrinking while a newly introduced $500 tier is seen as too expensive.

Third, the model itself is said to drop mid-task, so no number of resets fixes a job stuck at the same point. The article compares it to having unlimited bullets but a gun that jams on every shot.

The report's conclusion is that the most important word in Sottiaux's promise is not "daily" but "clearly useful to most users". Fewer disconnects, fewer rounds of rework and one less interface to learn all count as improvements, and a big reveal is not required. It also leaves an open question: is OpenAI this eager to please because its service mindset is maxed out, or because it is less confident in its own product?