- Type
- AI coding agent and model family
- Developer
- OpenAI
- First released
- 2021 (code model); 2025 (agentic tool)
- Key features
- Natural-language coding, autonomous execution, terminal and IDE integration, sandboxed cloud
- Related
- Claude Code, Cursor, GitHub Copilot, Devin
- Type
- AI coding agent and model family
- Developer
- OpenAI
- First released
- 2021 (code model); 2025 (agentic tool)
- Key features
- Natural-language coding, autonomous execution, terminal and IDE integration, sandboxed cloud
- Related
- Claude Code, Cursor, GitHub Copilot, Devin
OpenAI Codex is an artificial intelligence system developed by OpenAI for software development. The name covers two distinct generations of technology: the original Codex model released in 2021, which translated natural language into code and powered the first version of GitHub Copilot, and the agentic Codex introduced in 2025, which operates as an autonomous coding agent that plans, writes, runs, and iterates on code within a developer's environment. The 2025 Codex is available inside ChatGPT, as a cloud-based agent, and as Codex CLI, an open-source command-line tool.[1][2]
History
The original Codex was a language model based on GPT-3, fine-tuned on a large corpus of public source code to generate code from natural-language prompts. OpenAI announced it in August 2021, and it powered GitHub Copilot, which launched as a technical preview the same year. The model family included the code-davinci-002 and code-cushman-001 variants, and the technology demonstrated strong performance on programming benchmark tasks such as HumanEval.[2]
In May 2025, OpenAI reintroduced "Codex" as an agentic coding tool. Unlike the earlier model, the new Codex is a software agent: given a task, it can search through a codebase, write and edit multiple files, execute commands, run tests, and fix its own errors, with the user reviewing and approving changes. It was launched within ChatGPT and as a cloud-based agent with a containerised sandbox, followed by the release of Codex CLI as an open-source project under the Apache 2.0 licence, with the source code published on GitHub. Codex integrates with popular development tools, including VS Code and the terminal, and can be invoked from CI/CD pipelines.[1][3]
Key Concepts
Agentic coding: Codex operates in a loop of planning, editing, executing, and observing results. It can run commands in a sandboxed environment, read error output, and revise its approach without requiring the user to write intermediate code.
Human-in-the-loop approval: In interactive use, Codex proposes changes that the user can accept, reject, or modify before they are applied, giving developers control over the final state of the codebase.
Parallel task execution: The cloud version of Codex can spawn multiple parallel tasks, allowing it to work across several files or modules simultaneously while maintaining an overall task list.
Open-source CLI: The Codex CLI is a lightweight, transparent interface that runs the agent locally and connects to OpenAI's models, making the tool auditable and scriptable. It supports configuration files and non-interactive usage for automated workflows.[3]
Evaluation: Codex and similar agents are benchmarked on software-engineering tasks, including real-world issue resolution datasets derived from GitHub, which measure whether an agent can produce patches that pass the original project's tests.[1]
Applications and Impact
Codex is used across the software development lifecycle: generating new features from requirements, refactoring legacy code, writing unit tests, debugging failing builds, and automating repetitive maintenance tasks. Enterprises use it to accelerate onboarding of new engineers and to maintain large codebases with fewer manual edits. For independent developers, the CLI provides a low-cost entry point to agentic coding, and its open-source nature has allowed third parties to build on it. Codex competes directly with other coding agents such as Claude Code, Cursor, GitHub Copilot agent mode, and Devin.[1][3]
>See Also
References
OpenAI Codex has become a widely used tool among Malaysian software developers, startups, and technology training programmes. Its availability as an open-source CLI reduces cost barriers, allowing Malaysian developers to experiment with agentic workflows without committing to a paid plan, which matters for the local SME and freelancer community.[3]
Talent development: MDEC and other agencies promoting digital skills have incorporated AI coding tools into upskilling programmes, and Malaysian universities increasingly teach agent-assisted software engineering alongside traditional programming. Tools such as Codex are seen as a way to raise developer productivity in Malaysia's expanding tech sector.
Data and PDPA considerations: When Codex cloud agents process a company's source code, that code is transmitted to and processed on OpenAI's infrastructure outside Malaysia. Malaysian organisations working with proprietary or client-sensitive code must evaluate this against the cross-border data transfer provisions of the Personal Data Protection Act (PDPA) 2010, and may prefer the locally-running Codex CLI with self-managed repositories for sensitive projects.
Local adoption: Malaysian product teams use Codex for rapid prototyping, test generation, and maintenance of legacy systems, particularly in fintech, e-commerce, and government-linked digital services where development speed is a competitive factor. The National AI Office's push for sovereign AI capability has also prompted discussion of how dependency on foreign coding agents should be balanced with local skills development.[4]