OpenAI confirms 2.5x weekly growth across its agent products: Codex (the enterprise code editor) and ChatGPT Work. Sam Altman stated that demand for GPT-5.6 Sol is “insane” and could cause infrastructure hiccups as facilities scale.

The ecosystem reacted immediately: JetBrains added Codex as a recommended agent in its IDE, while OpenAI published operational guides showing how data science and sales teams use ChatGPT Work on real tasks—we documented these on July 14.

Why it matters to you. If you work with code or manage teams that write it, this signals that coding agency is no longer an experiment: it’s production infrastructure, with endorsements from mainstream IDEs and verifiable operational guides. The inflection point isn’t the technology—GPT-5.6 Sol was already available days in—but the integration into daily workflows and the fact that teams are documenting what actually works.

In detail

Where the growth comes from.

The 2.5x weekly growth in Codex + ChatGPT Work arrives in a specific context: GPT-5.6 Sol hit the market ten days ago (launched July 10) and is demonstrating concrete performance on complex coding tasks. The Ultra model even proved a mathematical conjecture open for 50 years on July 12.

But the growth number isn’t just the model: it’s the ecosystem moving. JetBrains, which dominates enterprise IDEs, chose Codex as a recommended agent—an endorsement worth as much as a certification. And OpenAI published two operational guides on data science and sales, showing verifiable workflows instead of vague promises.

Infrastructure under pressure.

Sam Altman added a significant detail: demand for GPT-5.6 Sol could cause “scaling hiccups” as infrastructure adapts. It’s not an absolute capacity problem—OpenAI has the hardware—but one of geographic distribution and latency. When a million daily users start using a model for long sessions (coding agents work for hours, not minutes), cluster pressure is different from quick chat.

What it means for coding agent users.

If you use Claude Code, you’ve seen rapid releases over the past two weeks (v2.1.212, v2.1.211, v2.1.210): Anthropic is racing to keep pace. If you use Codex in your company, you now have JetBrains integration and OpenAI’s operational guides to show your team.

The emerging pattern is clear: coding agents are no longer toys for early adopters. They’re tools teams use every day, which mainstream IDEs recommend, and on which companies document repeatable workflows. The competition isn’t “which model is more powerful” anymore, but “which ecosystem lets you work better”.

Open limitations.

OpenAI has reset usage limits multiple times in recent weeks, a sign that calibration between demand and capacity is ongoing. And as we reported July 12, token overhead before even reaching your prompt varies heavily across tools: Claude Code sends 33k tokens, OpenCode 7k. With 2.5x weekly growth, these efficiency details make the difference between a usable service and one that makes you wait.

The lesson: if you’re evaluating a coding agent for your team, look at documented workflows (OpenAI just published two), try the IDE integration where you already work (JetBrains now recommends Codex), and measure token overhead beyond output quality: the commit graphs of those who measure it systematically show the real impact.