OpenAI takes aim at Anthropic with beefed-up Codex that gives it more power over your desktop

OpenAI has rolled out a significant upgrade to Codex, its AI-powered coding agent, expanding the tool's capabilities and giving it deeper access to desktop environments. The move is widely seen as a direct challenge to Anthropic and its competing developer-focused AI tools.

The Details

Codex has been rebuilt with a broader set of agentic abilities, allowing it to interact more directly with a user's local machine rather than operating strictly within sandboxed environments. This shift toward greater desktop-level control puts it in closer competition with Anthropic's Claude, which has made similar computer-use functionality a centerpiece of its own product strategy.

Key changes in the updated Codex include:

• Expanded ability to read, write, and execute code across the local desktop environment
• Improved multi-step task handling, allowing the agent to work through complex workflows with less human intervention
• Tighter integration with OpenAI's broader model ecosystem

The upgrade reflects a clear pattern: the leading AI labs are racing to build agents that don't just answer questions but actively complete tasks on behalf of users. That distinction is becoming a major competitive battleground.

Why This Matters for Service Providers

For MSPs and IT service providers, this is worth tracking closely. AI agents that can take action on desktops represent both an opportunity and a risk vector. As these tools become more capable, clients will increasingly ask whether they're safe to deploy, and service providers who can answer that question confidently will have a real advantage.

There is also a procurement angle here. The competition between OpenAI and Anthropic is intensifying, which tends to drive down pricing and improve feature parity across platforms. Service providers who are evaluating AI tooling for internal use or resale should expect the landscape to look meaningfully different within the next 12 months.

At the same time, more powerful agentic tools raise legitimate security questions. Clients running managed environments will need guidance on access controls, permissions, and audit trails as these capabilities expand.

Looking Ahead

Watch for Anthropic to respond with further updates to Claude's computer-use features, and pay attention to how enterprise security vendors position themselves around agentic AI as a new attack surface. If your team hasn't started building an AI governance framework for client environments, now is a practical time to start.

For the full story, read the original article on TechCrunch AI.