Codex Climbed Out of Its Slump Under Tibo: Every Chronicles a Year-Long Comeback

Dan Shipper, co-founder and CEO of the tech publication Every, has released a long analysis titled "How Codex Happened." It chronicles how OpenAI's coding agent Codex climbed out of its lowest point in a little over a year and became one of the company's main products. Shipper worked on the text for several months and conducted a series of interviews with people involved in the events. The Chinese outlet Jiqizhixin retold the piece under a headline saying that Tibo, who took over Codex less than a year ago, has effectively "swallowed" ChatGPT.

A caveat up front. The excerpt available to us contains no revenue metrics, no user counts, and no internal charts. "Swallowed" is an image from the headline, not a measured quantity. Below is what can be stated, and what needs to be checked against the primary source.

What the publication tells us

The facts the story rests on:

• Author of the report: Dan Shipper, Every, long-read format.
• Title: "How Codex Happened."
• Scope: a little over a year of the product's life, from failure to resurgence.
• Work on the text: several months, with interviews of people involved.
• Key figure: Tibo, who took over Codex less than a year ago.
• Jiqizhixin's editorial thesis: Codex has matched ChatGPT in importance to OpenAI, or overtaken it.

In other words, we are looking at a retelling of a retelling. The long read's primary value lies in the behind-the-scenes details: product decisions, shifting priorities, disputes within the team. The excerpt does not reveal those details.

Why this story matters for the market

Coding agents have become the most straightforward commercial use of LLMs. A developer pays for hours saved, and that money is easy to calculate. That is why Codex, Claude Code, and other tools are fighting over every single engineer. If OpenAI's agent really did pull out of a slump, the cause lies in organization, not in a single model: who makes decisions, how quickly the team ships iterations, and how the link between the model and the execution environment is built.

The thesis about a "swallowed" ChatGPT also deserves caution. It can mean three different things: leadership attention and resources are shifting to Codex; Codex is growing faster than the chat product; the agentic interface is becoming the primary way of working with the model. These are different claims, and only usage and revenue figures can confirm them. We have no such figures at the moment.

How to use this news

For teams choosing a coding agent, the takeaway is simple: read the primary source, not the headlines. First open Every's long read and write down the specific decisions and timelines Shipper cites. Then check them against Codex's public releases and against what you observe in your own repositories.

A practical order of steps:

1. Run Codex and its closest competitor on the same set of real tasks: bug fixes, refactoring, tests. 2. Measure the share of accepted edits and the time spent on review, not impressions. 3. Do not build a strategy on the thesis that one product has won until there is usage data. 4. Watch how OpenAI embeds the agent in its core products: integration will say more about priorities than any interview.

The Codex story is interesting as a case study of a management turnaround at a large AI company. But it will become convincing only when metrics appear alongside the catchy headline.