OpenAI Updated Codex: Agent Now Works With Computer, Remembers Context and Plans

OpenAI has significantly expanded Codex's capabilities, effectively elevating it from the 'code assistant' category to a class of full-fledged work agents. Following the April 16, 2026 update, the service can not only write and verify code but also work with applications on a computer, remember user context, use a built-in browser, and continue long tasks without constant human involvement. For OpenAI, this is an important move in direct competition with Anthropic, whose Claude Code many consider a more mature tool for agentic development.

The company states that Codex is already used by over 3 million developers per week, and now OpenAI is trying to establish it not only as a tool for programmers but as a universal assistant for product, research, and operational teams. The logic is clear: if an agent can understand code, documents, comments, spreadsheets, and external services, it becomes useful far beyond the IDE.

The main change is background work with desktop applications. Codex can now see interfaces, click with the cursor, type, and open programs, performing actions the way a user would. At launch, OpenAI is rolling out this mode on macOS and separately emphasizes that multiple agents can work in parallel without interfering with what a user is doing in other windows. The practical benefit is straightforward: agents can be assigned routine tasks like frontend testing, re-running test scenarios, navigating internal tools, or actions in applications where there is no API at all.

The application also introduced its own browser. Through it, users can leave instructions directly on the page, while Codex executes actions in specific web interfaces. This is particularly useful for frontend development, interface testing, and rapid design iteration. Additionally, OpenAI added image generation so that in one workflow, users can create not only code but also mockups, diagrams, slides, and visual sketches.

Another notable addition is integrations: Codex now has over a hundred connections to third-party tools, including services for code review, task management, CI/CD, and enterprise work. The update also affected the developer's standard workflow. Codex now can parse comments in GitHub reviews, work with multiple terminal tabs, connect to remote devboxes via SSH, and open not only code but also PDFs, spreadsheets, presentations, and documents in the sidebar.

A separate summary panel appeared where you can see the agent's plan, sources used, and collected artifacts. This is an important detail: OpenAI is trying to transform Codex from a chat for one-off queries into a unified workspace where the agent writes code, verifies results, prepares edits, and displays the intermediate state of work.

Another layer of updates relates to long-running tasks. Codex gained a memory mode in which it remembers user preferences, past fixes, and already-discovered context. Moreover, the agent can reuse old threads, self-plan work continuation, and 'wake up' later to return to tasks after hours, days, or even weeks. In practice, this enables scenarios like daily summaries from Slack, Notion, Gmail, or Google Docs, as well as automatic tracking of pull requests and work discussions.

In parallel, on April 2, 2026, OpenAI switched Codex pricing to a more flexible consumption-based payment model, making the tool more convenient for teams with irregular usage. The point of the update is that Codex increasingly resembles not a typical code generator but an operating shell for handling digital tasks. If OpenAI can maintain reliability, transparency of agent actions, and reasonable pricing, Codex could become stronger precisely where the company previously lagged behind Anthropic: in everyday practical usefulness. But now the competition is no longer just for programmers, but for the position of the primary interface for team collaboration with AI.