The race to dominate the enterprise AI market is intensifying as tech companies develop increasingly capable automated assistants. OpenAI has launched significant updates to its developer tools and coding platforms, aiming to give businesses more control and security when building and deploying artificial intelligence. With enhancements to the OpenAI Agents SDK and a major revamp of its automated tool Codex, the company is directly challenging rivals like Anthropic in the competitive landscape of agentic AI.
As agentic AI—systems that can perform complex, multi-step tasks autonomously—becomes the tech industry’s latest success story, companies are rushing to provide the necessary infrastructure. OpenAI’s recent moves demonstrate a strategic push to make its tools more multifaceted and secure for corporate workflows. The updates address two major areas: giving AI more direct control over desktop environments through Codex, and providing developers with safer, more robust building blocks through the OpenAI Agents SDK.
Expanding Desktop Control with Codex
OpenAI has significantly beefed up Codex to compete with Anthropic’s Claude Code, which has recently gained traction as a preferred tool for many businesses. The most notable addition to Codex is its new ability to operate in the background on a user’s computer. The tool can now open desktop applications and perform operations using a cursor that clicks and types.
This functionality allows Codex to deploy multiple agents working in parallel on a Mac without interrupting the user’s ongoing work in other applications. According to OpenAI, this allows the agent to serve as a coding buddy that handles auxiliary tasks while the user focuses on top-line projects. Potential use cases include iterating on frontend changes, testing applications, or operating within apps that lack an application programming interface (API). Observers of the AI coding sector may notice similarities to Anthropic’s recent announcement that Claude and Cowork could remotely control a Mac and a desktop on a user’s behalf while they are away from the keyboard.
Beyond background desktop control, Codex now features an in-app browser. This allows users to issue commands that the agent will execute on specific web applications, which OpenAI notes is particularly useful for frontend and game development. The company plans to eventually expand this capability so Codex can command the browser entirely, moving beyond web applications on localhost.
To further integrate into daily corporate workflows, Codex has introduced several other features. A new “memory” function, currently in preview, enables the tool to recall previous work sessions and generate context based on a user’s specific work habits. An added image-generation capability allows users to create product concepts, slide visuals, mockups, and placeholder images.
Additionally, OpenAI announced 111 plug-in integrations with applications such as CodeRabbit, GitLab Issues, Slack, and Google Calendar. These integrations enable Codex to perform minor clerical tasks, like reviewing Slack channels and calendar events to generate a daily to-do list. To provide more flexibility, a new pay-as-you-go pricing option has been introduced for ChatGPT enterprise and business customers.
Securing AI with the OpenAI Agents SDK
While Codex handles direct user assistance, OpenAI has also updated its software development toolkit to help businesses create their own automated helpers running on OpenAI models. The latest version of the OpenAI Agents SDK introduces crucial security and operational features, most notably a new sandboxing ability.
Running autonomous agents unsupervised carries inherent risks due to their sometimes unpredictable behavior. The sandbox integration addresses this by allowing agents to operate within controlled computer environments. They can work in a siloed capacity within a specific workspace, accessing files and code only for designated operations, which protects the overall integrity of the broader system.
Karan Sharma, a member of OpenAI’s product team, explained the focus of the update. “This launch, at its core, is about taking our existing Agents SDK and making it so it’s compatible with all of these sandbox providers,” Sharma stated.
Alongside sandboxing, the updated toolkit provides developers with an in-distribution harness for frontier models. In agent development, a harness comprises the components surrounding the underlying model. This new harness allows agents to interact with files and approved tools within a workspace. Sharma noted that pairing the sandbox with the new harness capabilities is designed to let users build “long-horizon agents” using their existing infrastructure. These long-horizon tasks involve more complex, multi-step operations.
The new harness and sandbox capabilities are launching first in Python, with standard API pricing for all customers. OpenAI plans to expand the toolkit over time, noting that TypeScript support is planned for a later release. The company is also working to bring features like code mode and subagents to both Python and TypeScript environments.
The Enterprise AI Rivalry
These updates underscore OpenAI’s aggressive pivot toward enterprise capabilities amid fierce competition. Once viewed as the undisputed leader in the AI industry, OpenAI is now engaged in a tight contest with Anthropic over who can deliver the most powerful and convenient AI tools for businesses.
This enterprise focus coincides with a strategic retreat from certain consumer-facing products, such as the social video app Sora 2. It also arrives as the company navigates various controversies, including lawsuits regarding the alleged mental health impacts of ChatGPT on some users. By strengthening both Codex and the OpenAI Agents SDK, the company aims to solidify its infrastructure and prove that its automated tools can securely and efficiently handle the complex demands of modern businesses.
