AI agents are no longer just coding assistants—they’re becoming essential partners in knowledge work, handling information processing, complex problem-solving, and innovation. OpenAI and NVIDIA have reached a significant milestone with GPT-5.5, the latest frontier model powering Codex, NVIDIA’s agentic coding application. Deployed on NVIDIA GB200 NVL72 rack-scale systems, this breakthrough is already transforming workflows across the company. Over 10,000 NVIDIANs—spanning engineering, product, legal, marketing, finance, sales, HR, operations, and developer programs—are leveraging GPT-5.5-powered Codex to achieve what they describe as “mind-blowing” and “life-changing” results. Here are seven key insights from this deployment.

1. GPT-5.5 Powers Codex for Enterprise Knowledge Work

OpenAI’s GPT-5.5 is the engine behind Codex, NVIDIA’s agentic coding application. Unlike earlier models, GPT-5.5 excels at processing complex information, solving nuanced problems, and generating innovative ideas. This makes it ideal for knowledge work, where AI agents must understand context, reason across domains, and produce actionable outputs. By running on NVIDIA GB200 NVL72 infrastructure—designed for high-performance AI inference—Codex delivers responses with remarkable speed and accuracy. The combination allows NVIDIA teams to move beyond simple code generation and tackle intricate tasks like multi-file debugging, system design, and even legal document analysis. This shift marks a new era where AI agents become genuine collaborators in intellectual work.

2. Measurable Gains with GB200 NVL72 Economics

The GB200 NVL72 system is the backbone of this deployment, offering game-changing cost and performance benefits. Compared to prior-generation hardware, it achieves 35x lower cost per million tokens and 50x higher token output per second per megawatt. These economics make it feasible to run frontier models at enterprise scale without breaking budgets. For NVIDIA, this means GPT-5.5 can be used extensively across the organization, enabling rapid experimentation and iteration. The infrastructure’s efficiency also reduces energy consumption, aligning with sustainability goals. As a result, what was once reserved for high-priority projects is now accessible to thousands of employees, democratizing advanced AI capabilities.

3. Debugging Cycles Slashed from Days to Hours

One of the most dramatic improvements is in debugging. Previously, complex multi-file bugs could take days to isolate and fix. With GPT-5.5-powered Codex, these cycles are now closing in hours. The model can analyze entire code repositories, understand dependencies, and suggest precise fixes, often identifying root causes that humans might overlook. Engineers report that they can submit a bug report in natural language, receive a detailed analysis, and have a working patch tested within a single workday. This acceleration not only boosts productivity but also reduces frustration and allows developers to focus on higher-level design and innovation rather than tedious troubleshooting.

4. Overnight Experimentation Replaces Weeks of Work

Experimentation that used to require weeks is now completed overnight. GPT-5.5’s ability to reason about large, multi-file codebases lets teams test new features, refactor architecture, or validate hypotheses at unprecedented speed. For example, a product team can describe a desired feature in natural language, and Codex generates a complete end-to-end implementation, including tests and documentation, within hours. This capability is particularly valuable for iterative development, where quick feedback loops drive innovation. Engineers no longer need to wait days for results—they can iterate in real time, making decisions based on data rather than assumptions.

5. Enterprise-Grade Security with Zero-Data Retention

Security is paramount in any enterprise AI deployment. NVIDIA addressed this by implementing zero-data retention policies and providing dedicated cloud virtual machines (VMs) for each employee. These VMs act as sandboxes where Codex agents operate with full auditability. Agents access production systems via read-only permissions through command-line interfaces and Skills—the same toolkit used for automation workflows. Users control their agents from a familiar interface, ensuring ease of use without sacrificing security. This setup guarantees that sensitive company data never leaves a controlled environment, meeting strict compliance requirements while enabling powerful AI capabilities.

6. Read-Only Access with Full Audit Trails

To maintain integrity, Codex agents are configured with read-only access to production systems. They can analyze data, propose changes, and generate code, but they cannot modify critical systems without human approval. This approach minimizes risk while still allowing agents to provide meaningful insights. Additionally, every action is logged, creating a comprehensive audit trail for compliance and review. NVIDIA’s IT team rolled out cloud VMs to every employee, ensuring that each agent operates in an isolated environment. Users can manage their agents through Skills, which are pre-defined automation workflows, further standardizing and securing operations. This balance between capability and control is essential for enterprise adoption.

7. A Decade of Collaboration Behind the Breakthrough

The GPT-5.5 launch and Codex rollout are the culmination of over 10 years of partnership between NVIDIA and OpenAI. It began in 2016 when NVIDIA founder and CEO Jensen Huang personally delivered one of the first DGX-1 systems to OpenAI. Since then, the two companies have worked closely to optimize AI models for NVIDIA hardware, achieving breakthroughs in performance and efficiency. This collaboration extends beyond technology—it includes shared vision and mutual respect. As Huang told employees in a company-wide email urging Codex adoption: “Let’s jump to lightspeed. Welcome to the age of AI.” This long-term relationship sets the stage for even more advanced AI agents in the future.

In conclusion, NVIDIA’s deployment of GPT-5.5-powered Codex on GB200 NVL72 infrastructure is a landmark example of how frontier models can be harnessed securely and efficiently at enterprise scale. From slashing debugging times to enabling overnight experimentation, these seven insights highlight the transformative power of AI agents in knowledge work. As more organizations adopt similar setups, the line between human and machine collaboration will continue to blur, ushering in a new era of productivity and innovation.