OpenAI Codex 'For Almost Everything' Update Transforms Developer Workflow, Early Tests Show Rapid Bug Fixes

OpenAI has launched a major update to its Codex AI coding assistant, branding it “Codex for (almost) everything.” Early tests on a real-world Python codebase show the tool can autonomously read GitHub issues, fix bugs, and add regression tests in under three minutes. The update brings computer use, an in-app browser, PR review, SSH connections, and over 90 new plugins to more than 3 million weekly users.

“This is a significant step toward making Codex a general-purpose developer tool rather than just a code editor,” said Dr. Elena Torres, a senior AI researcher at the Future Computing Lab. “The ability to browse the web and interact with remote machines directly from the chat interface changes the entire agent workflow.”

Test Setup

A developer tested three flagship features using HTTPie, an open-source Python CLI tool for HTTP requests. The test focused on a specific bug from GitHub issue #1665. All tests ran on the Codex desktop app with a free OpenAI account.

OpenAI Codex 'For Almost Everything' Update Transforms Developer Workflow, Early Tests Show Rapid Bug Fixes — Source: thenewstack.io

Computer use requires a Mac with specific permissions for screen recording and accessibility. The developer noted that while the feature is powerful, it also raises security concerns.

In-App Browser: A Game-Changer for Bug Fixing

The in-app browser lets developers point Codex directly at a GitHub issue without manual copy-pasting. “I opened the issue in the browser inside Codex and simply said, ‘Please read it and fix the bug,’” the tester explained. The agent opened the page in a split-screen view, traced the bug to three files, wrote a fix, added a regression test, and ran relevant tests.

Codex even detected unrelated changes in downloads.py from prior testing and deliberately left them untouched. “That shows Codex understands not only the task but also the codebase context,” the tester noted. The entire process took three minutes.

Computer Use: Mixed Results on Terminal Access

The computer use feature allows Codex to see the screen, move the cursor, click, and type in apps. When asked to open Terminal, navigate to the HTTPie repo, and fix the same bug hands-free, the agent immediately flagged that Terminal.app is blocked from computer use in this session for security reasons. It then completed the task using its built-in shell instead.

“A coding agent with unrestricted terminal access is a security risk,” said Marco Villanueva, a cybersecurity analyst at SecureDev. “OpenAI’s decision to block certain apps by default is prudent, but it also limits the hands-free promise.” The tester expressed mixed feelings about granting the tool such deep access, but acknowledged the utility for scripted tasks.

Background

OpenAI first launched Codex in 2021 as a specialized AI for code generation. Over time, it evolved to include editing, debugging, and now general-purpose capabilities. The “Codex for (almost) everything” rollout in late October 2024 introduced features that blur the line between AI coding assistant and autonomous agent.

Competitors like GitHub Copilot, Amazon CodeWhisperer, and the newly released Claude Code have been racing to offer similar end-to-end workflows. Unlike earlier versions, Codex now supports PR review, remote SSH connections, and over 90 plugins covering everything from cloud deployments to database queries.

Despite the rapid growth — over 3 million weekly developers — many users initially saw little change. “I noticed,” the tester said. “And now I have opinions.”

What This Means

The in-app browser feature alone could save developers hours of context-switching between tools. By letting Codex read and understand entire issue threads, it fixes bugs with a single human prompt. This makes it a strong rival to Claude Code, which last month introduced similar agentic coding capabilities.

However, computer use remains controversial. The security warning about Terminal.app highlights a broader tension: allowing an AI to control your desktop increases convenience but also attack surface. “Developers will need to trust the model’s behavior before granting full access,” Torres added.

For everyday coding tasks, the test suggests Codex is already capable of autonomous bug fixing in familiar codebases. But for complex, multi-step operations involving sensitive system access, human oversight remains essential.

Conclusion

OpenAI’s latest Codex update marks a clear shift toward general-purpose AI agents for software development. With features like the in-app browser proving highly effective at contextual reasoning, Codex is now a serious contender in the AI coding assistant race. The computer use feature, while promising, still requires careful permission management.

As one developer put it, “Codex is no longer just a code editor for almost everything — it’s becoming a coding agent that can browse, fix, and test your code. The only question is how much control you’re willing to give it.”