Announcing new improvements to Codex

If you use Codex in your daily development workflow—whether at your desk, on the go, or switching between devices—this update matters to you. OpenAI's coding agent has been refreshed with significant improvements to speed, dependability, real-time teamwork, and its ability to handle tasks on its own. In this article, we'll walk through what's changed and why it makes Codex a more practical companion across every environment where you build software: the terminal, your IDE, the web, and even your phone.

What's New: Speed and Reliability Come First

The headline improvements are straightforward: Codex now runs faster and breaks down less often. For anyone who has waited on a sluggish response or watched a long task fail halfway through, these changes address real pain points. Quicker turnaround means shorter feedback loops when you're iterating on code, and better reliability means you can hand the agent longer, more complex jobs with greater confidence that it will actually finish them.

For developers, this is the difference between treating an AI assistant as a novelty and treating it as genuine infrastructure. When a tool is fast and stable enough to trust, it stops being something you demo and starts being something you depend on.

Real-Time Collaboration Gets Better

Coding with AI is increasingly a two-way conversation rather than a one-shot request. This update strengthens Codex's real-time collaboration features, meaning you can work alongside the agent more fluidly—steering it as it goes, correcting course mid-task, and reviewing changes as they happen instead of waiting for a finished result.

This matters for how teams actually work. Pair programming with an agent that responds in real time feels much closer to working with a colleague than firing off prompts and hoping for the best. If you've found earlier versions awkward to collaborate with, the improved interaction model should feel noticeably smoother.

More Capable of Working Independently

The other big theme is autonomy. Codex has improved at tackling tasks on its own—independently working through assignments without needing constant hand-holding. That means you can describe what you want done and let it run, checking back when the work is complete rather than babysitting every step.

Autonomy and collaboration might sound like opposite goals, but they complement each other well. You can let Codex run unsupervised on well-defined tasks, then jump into a live collaborative session when the work needs refinement or when you're exploring something new. Having both modes available makes the tool flexible enough for different working styles.

One Agent, Every Surface You Develop On

Perhaps the most practical aspect of this update is its emphasis on availability everywhere you develop. Codex now works consistently across:

• The terminal — for developers who live in the command line
• Your IDE — for those who prefer an integrated editor experience
• The web — for browser-based access from any machine
• Your phone — for checking in on tasks, reviewing work, or kicking off jobs when you're away from your keyboard

This multi-surface approach changes how you can fit an AI agent into your day. Start a task from your laptop, monitor progress from your phone during your commute, and pick up the results back at your desk. The agent travels with you rather than being tied to a single environment.

What This Means for Your Workflow

Putting it all together: a faster, more dependable Codex that collaborates in real time, works independently, and follows you across every device you use. That combination lowers the barrier to making an AI coding agent a genuine part of your daily routine rather than an occasional helper.

If you've been on the fence about integrating Codex into your workflow—or if a past experience with slow or flaky performance turned you off—it may be worth giving the updated version a try. The improvements target exactly the issues that made earlier versions frustrating for sustained, everyday use.