Codex: ChatGPT's Cloud Agent Writes Code, Fixes Bugs, Automates Tasks

Codex is a cloud-based software engineering agent for ChatGPT users that automates tasks like coding, debugging, and testing, enhancing developer productivity and integrating across various platforms.

Codex: A Cloud-Based Software Engineering Agent

What is Codex?

• A cloud-based software engineering agent powered by codex-1.
• Can work on many tasks in parallel.
• Available to ChatGPT Pro, Team, and Enterprise users now, Plus users soon.

What can Codex do?

• Write features.
• Answer questions about your codebase.
• Fix bugs.
• Propose pull requests for review.

How does it work?

• Accessed through the ChatGPT sidebar.
• Tasks are processed independently in isolated cloud environments preloaded with your codebase.
• Can read and edit files, run commands (tests, linters, type checkers).
• Task completion takes 1-30 minutes.
• Provides verifiable evidence of actions through citations of terminal logs and test outputs.
• Can be guided by AGENTS.md files in your repository.

Codex CLI

• Lightweight open-source coding agent that runs in your terminal
• Smaller version of codex-1 for faster workflows
• Optimized for low-latency code Q&A and editing
• Easy developer account connection with ChatGPT account

Safety and Trust

• Trained to identify and refuse malicious software development requests.
• Operates within a secure, isolated container in the cloud, without internet access during task execution.

Early Use Cases

• Cisco: Exploring how Codex can help engineering teams bring ideas to life faster.
• Temporal: Accelerates feature development, debugs issues, writes/executes tests, and refactors code.
• Superhuman: Speeds up repetitive tasks (test coverage, integration failures), enables product managers to contribute lightweight code changes.
• Kodiak: Helps write debugging tools, improve test coverage, and refactor code for autonomous driving technology.

Tips from Early Testers

• Assign well-scoped tasks to multiple agents simultaneously.
• Experiment with different types of tasks and prompts.

Availability and Pricing

• Rolling out to ChatGPT Pro, Enterprise, and Team users globally, Plus and Edu coming soon.
• Generous access at no additional cost initially, then rate-limited access and flexible pricing.
• codex-mini-latest is priced at $1.50 per 1M input tokens and $6 per 1M output tokens (Responses API).

Limitations

• Lacks features like image inputs for frontend work.
• Cannot course-correct the agent mid-task.
• Delegating to a remote agent takes longer than interactive editing.

Future Vision

• Developers delegate repetitive tasks to agents, increasing productivity.
• Real-time pairing and task delegation will converge into a unified workflow.
• Deeper integrations across IDEs, ChatGPT Desktop, issue trackers, and CI systems