OpenAI announced a major expansion of Codex this week, transforming it from a code-focused assistant into a full-spectrum development partner that can operate your computer, browse the web, generate images, and remember your preferences across sessions.

With over 3 million weekly users already, this update positions Codex to compete directly with GitHub Copilot Workspace, Anthropic's Claude Code, and other emerging AI development tools. But does it deliver?

Let's break down what's new, what works well, and where the gaps remain.

What's New in Codex

1. Computer Use: Your AI Can Now Click and Type

Codex can now operate your Mac independently, seeing your screen, moving its own cursor, clicking, and typing—all while you work in other apps. Multiple agents can run in parallel without interfering with your workflow.

Why this matters: This is particularly useful for:

• Iterating on frontend changes in browsers or design tools
• Testing apps that don't expose APIs
• Working across multiple applications simultaneously

Platform note: Computer use is currently macOS-only, with EU/UK rollout coming soon. Windows and Linux support wasn't mentioned.

2. In-App Browser for Frontend Development

The Codex desktop app now includes a built-in browser where you can comment directly on pages to give precise instructions. Think of it as "inspect element" meets AI pair programmer.

Use case: Perfect for frontend and game development on localhost. OpenAI plans to expand this so Codex can command the full browser beyond local development.

3. Image Generation with gpt-image-1.5

Codex can now generate and iterate on images using OpenAI's gpt-image-1.5 model. Combined with screenshots and code, this enables rapid prototyping of:

• Product concept visuals
• Frontend designs and mockups
• Game assets and UI elements

4. 90+ New Plugins and Integrations

OpenAI released more than 90 new plugins that combine skills, app integrations, and MCP (Model Context Protocol) servers. Developer-relevant additions include:

Plugin	Purpose
Atlassian Rovo	JIRA management
CircleCI	CI/CD pipeline integration
CodeRabbit	Code review automation
GitLab Issues	Issue tracking
Microsoft Suite	Office document handling
Neon by Databricks	Database operations
Remotion	Video programmatic generation
Render	Deployment automation
Superpowers	Unknown (likely internal tool)

5. Enhanced GitHub Workflow Support

Codex now includes deeper GitHub integration:

• Address review comments directly in the app
• Run multiple terminal tabs simultaneously
• Connect to remote devboxes via SSH (alpha)
• Open files with rich previews (PDFs, spreadsheets, slides)
• Summary pane to track agent plans, sources, and artifacts

6. Automations and Memory

Two features that could be genuine differentiators:

Automations: Codex can schedule future work for itself and wake up automatically to continue long-term tasks across days or weeks. Teams are reportedly using this for:

• Landing open pull requests
• Following up on tasks
• Monitoring Slack, Gmail, and Notion conversations

Memory (Preview): Codex remembers useful context from previous sessions—personal preferences, corrections, and hard-won information. This reduces the need for extensive custom instructions and speeds up repetitive tasks.

Personalization rollout: Context-aware suggestions and memory are rolling out to Enterprise, Edu, and EU/UK users soon.

How Codex Compares to Alternatives

vs. GitHub Copilot Workspace

Feature	Codex	Copilot Workspace
Computer use	✅ Yes (macOS)	❌ No
In-app browser	✅ Yes	❌ No
Image generation	✅ Yes (gpt-image-1.5)	❌ No
GitHub integration	✅ Deep	✅ Native (GitHub-owned)
Memory	✅ Preview	⚠️ Limited
Pricing	ChatGPT subscription	GitHub subscription

Verdict: Codex has more ambitious autonomy features, but Copilot Workspace has deeper native GitHub integration and may feel more seamless for GitHub-heavy workflows.

vs. Anthropic Claude Code

Feature	Codex	Claude Code
Computer use	✅ Yes (macOS)	⚠️ Limited
Multi-agent parallel work	✅ Yes	❌ No
Image generation	✅ Yes	❌ No (text-only models)
Context window	Undisclosed	200K+ tokens
Safety focus	Standard	Heavy emphasis

Verdict: Claude Code excels at careful, safety-conscious code review and complex reasoning. Codex is more action-oriented and multimodal.

Getting Started

If you've been using Codex in the terminal or editor, these updates are rolling out today to the Codex desktop app for ChatGPT subscribers.

To get started:

1. Download the Codex app
2. Sign in with your ChatGPT account
3. Explore the new sidebar with file previews and summary pane
4. Try commenting on a localhost page in the in-app browser
5. Set up plugins for your most-used tools (JIRA, CircleCI, etc.)

Limitations and Considerations

Before diving in, keep these constraints in mind:

• macOS only for computer use (Windows/Linux users wait)
• EU/UK rollout pending for personalization features
• Memory is in preview—expect bugs and limited retention
• SSH devbox support is alpha—not production-ready
• Pricing: Requires ChatGPT subscription (Plus, Team, or Enterprise)

The Bottom Line

OpenAI's Codex expansion is a significant step toward autonomous development workflows. The ability to operate your computer, remember context across sessions, and work across 90+ integrated tools makes it more than just a code completion tool—it's becoming a genuine development partner.

Best for:

• Frontend developers who want AI assistance with visual iteration
• Teams managing complex workflows across JIRA, CI/CD, and GitHub
• Developers comfortable with AI having substantial autonomy

Think twice if:

• You're on Windows or Linux (computer use unavailable)
• You prefer human-in-the-loop for all code changes
• You're concerned about AI agents running unattended

What's Next?

OpenAI hints at "much more to come soon," suggesting this isn't the ceiling. We'd like to see:

• Windows and Linux support for computer use
• More transparency around what Codex sees and stores
• Better controls for automation boundaries
• Pricing tiers that reflect the expanded capabilities

For now, if you're already in the OpenAI ecosystem, the updated Codex is worth exploring. If you're evaluating AI development tools, this raises the bar significantly.