Jensen Huang, CEO of Nvidia, took the stage at Computex 2026 in Taipei to deliver a message that many in the software industry had been waiting to hear: agentic AI will not spell doom for traditional software companies. Instead, Huang argued, the rise of AI agents that can independently plan and execute complex tasks will make the underlying software platforms more critical than ever. His remarks come amid a growing narrative that advanced AI assistants could replace swathes of enterprise software, from CRM to productivity suites, by automating workflows and bypassing conventional user interfaces.

But Huang pushed back firmly. “Agentic AI doesn’t eliminate software,” he said during the keynote, which was live-streamed to millions of developers and IT professionals worldwide. “It expands the surface area of what software can do. The platform layer that orchestrates data, identity, and access becomes the control point—and that’s worth far more than any single application.”

The statement was a direct counter to fears that the likes of Microsoft, Salesforce, and Adobe might see their business models upended as AI copilots evolve from reactive assistants into proactive agents. In Huang’s view, the opposite is true: companies that own the operating system, the cloud infrastructure, and the productivity suite are best positioned to benefit from agentic AI because they control the choke points—authentication, file systems, APIs, and user context.

The Rise of Agentic AI and the Fear It Will Devour Software

Agentic AI refers to systems that go beyond simple query-and-response interactions. These agents set goals, make multi-step plans, use tools, and even collaborate with other agents to achieve outcomes. Microsoft’s Copilot, for example, is already moving toward agent-like capabilities with features like Copilot Studio that let users create autonomous bots. Third-party projects like AutoGPT and open-source frameworks such as LangChain have demonstrated the potential—and the disruption.

The fear is understandable. If an AI agent can directly interact with databases, send emails, schedule meetings, and update spreadsheets without the need for a human to open a CRM or ERP application, then the value of that application’s interface plummets. Some analysts have speculated that entire categories of enterprise software could be reduced to commoditized back-end services, with the AI agent becoming the new front door.

In 2025, a Gartner report predicted that by 2028, 30% of enterprise software interactions would occur through agentic AI rather than traditional UIs. That led to a flurry of think-pieces proclaiming the death of SaaS. Venture capital firm Andreessen Horowitz even published a widely circulated blog post titled “The SaaS Apocalypse?” Shortly after, Salesforce CEO Marc Benioff admitted that AI agents were a “seismic shift” but insisted his company’s platform would remain essential because agents need a trusted data foundation.

To understand Huang’s stance, it’s worth examining what agentic AI actually means in 2026. Unlike earlier chatbots, agentic systems can reason, use tools, and retain memory across sessions. Microsoft’s Copilot, embedded in Windows, has evolved into a true agent with the release of Windows 11 24H2’s Copilot+ PCs. On devices with NPUs capable of 40 TOPS or more, Copilot can operate locally for many tasks, reducing latency and keeping data private. Meanwhile, the cloud-based Copilot can tap into the full power of GPT-5 and offer advanced planning. This hybrid approach exemplifies how the platform—Windows—adds value by managing where and how AI computation happens.

Huang’s Counterargument: The Platform Layer Becomes the Crown Jewels

At Computex 2026, Huang provided a nuanced rebuttal that resonated with many enterprise IT leaders. He asserted that agentic AI will not commoditize software but will instead shift value from the application user interface to the platform layer that enables and governs those agents. This platform layer includes:

  • Identity and access management: AI agents must operate within strict permission boundaries. The software managing authentication and authorization becomes more valuable, not less.
  • Data management and integration: Agents need reliable, governed data sources. The platforms that store, organize, and serve that data (like Microsoft Graph, Salesforce Data Cloud, or Snowflake) become the essential backbone.
  • API orchestration and tooling: Rather than destroying APIs, agents depend on them. Platforms that expose robust, well-documented, and secure APIs will thrive. Nvidia itself provides full-stack platforms like CUDA and AI Enterprise.
  • The operating system: Windows, Linux, and macOS will be the runtime environments where many agents execute local tasks. With Windows Copilot Runtime and the integration of NPUs, Windows is becoming an AI-native platform.

Huang emphasized that these platform layers are not being eroded; they are being “embedded deeper into the stack.” He drew a parallel to the early internet era: “People thought the browser would replace the operating system, but what happened? The OS became even more central because it had to manage networking, security, and device drivers. Windows didn’t die—it evolved. The same thing will happen with AI.”

Microsoft’s Bet on the Agentic Platform

The strategic alignment with Huang’s vision is most visible at Microsoft. The Redmond giant has been aggressively building out what it calls an “agentic fabric” across its products. At Build 2026, Microsoft CEO Satya Nadella introduced Copilot extensibility frameworks that allow third-party agents to run directly within Windows, Office, and Teams. The company also announced new APIs that give agents access to user context via Microsoft Graph while maintaining enterprise-grade compliance.

Windows 11’s latest update, version 24H2, introduced AI-powered capabilities like Recall and advanced Copilot integration. But it’s the underlying architecture—the Copilot Runtime, backed by local NPUs found in Snapdragon X and Intel Lunar Lake chips—that positions Windows as a formidable agentic platform. By processing sensitive AI tasks on-device, Windows can offer low-latency, private agent capabilities that cloud-only solutions cannot match.

Nvidia’s role in this ecosystem is pivotal. While Nvidia is often seen as a hardware company, Huang reiterated that its software stack—CUDA, cuDNN, TensorRT, and now NIM inference microservices—comprises a platform layer for AI. During his Computex keynote, he announced partnerships with major ISVs to embed Nvidia AI capabilities into their products. He highlighted how Microsoft’s Azure AI services leverage Nvidia GPUs, but also how that relationship is symbiotic: Nvidia’s platform adds value to Windows and Azure, while Microsoft’s platforms provide the distribution and developer ecosystem that Nvidia needs.

The Hardware Angle: Nvidia’s New Chips

During his keynote, Huang unveiled Nvidia’s next-generation architecture, codenamed “Rubin,” expected in 2027. The new GPUs will feature dedicated hardware for agent orchestration, improving throughput for multi-agent systems by up to 10x, he claimed. Nvidia also announced a partnership with Microsoft to optimize Windows Copilot Runtime for RTX GPUs, enabling even more powerful on-device agents. For Windows users with RTX 5000 series or newer cards, this means advanced AI capabilities without consuming cloud credits.

Real-World Impact: How Agentic AI Is Reshaping Workflows

Huang’s optimism isn’t just theoretical. At Computex, he demonstrated several agentic workflows running on Nvidia-accelerated hardware. One demo involved an AI agent tasked with planning a company’s supply chain logistics. The agent used Microsoft Outlook to communicate with suppliers, Excel to analyze inventory, and a custom ERP connector to place orders—all without a human touching those applications. But Huang pointed out that the agent required deep integration with each platform’s APIs and authentication. “Every one of those applications became more valuable because they were part of an autonomous workflow,” he said.

Early adopters in the Windows ecosystem are already seeing this play out. A large financial services firm built an internal agent using Copilot Studio that integrates with Windows Hello for secure authentication, then accesses internal databases via Microsoft 365 connectors. The agent automates 70% of the firm’s monthly reporting tasks. IT managers there told windowsnews.ai that while the user interface of their legacy ERP is less frequently accessed, the value of the data and the APIs the ERP provides has actually increased. “We couldn’t have done this without the platform,” one senior architect said. “The agent needs that trusted source of truth.”

The Battle for the Agentic Platform Is Just Beginning

If Huang is right, the next frontier of competition won’t be over AI models or agents themselves—it will be over the platform layer that hosts, manages, and secures those agents. This explains why Microsoft, Google, Amazon, and Salesforce are racing to enhance their platforms with agent-oriented features. At Computex, Huang hinted that Nvidia’s strategy is to remain the “preferred compute substrate” for these platforms, but also to offer its own platform services like Nvidia AI Enterprise that can run agents anywhere.

For Windows users and IT pros, the implications are significant. The operating system will increasingly become the secure enclave for AI agents, managing local compute, identity, and data. Microsoft is already pushing the concept of “Windows AI PCs” with dedicated neural processing units. Intel’s latest Core Ultra processors and Qualcomm’s Snapdragon X Elite chips are explicitly designed with local AI inference in mind. Huang noted that Nvidia’s GPUs, often found in high-end workstations, will continue to be critical for training and fine-tuning, but the inferencing that agents rely on can be distributed across edge devices and cloud servers.

Addressing Skepticism: Security, Reliability, and the Human Factor

Not everyone is convinced that agentic AI will be a purely positive force for software companies. Some analysts at the conference raised concerns about security vulnerabilities and the “black box” nature of agent decisions. Huang acknowledged these challenges, stating that the industry must develop new governance frameworks. “We can’t just let agents roam free on corporate networks,” he said. “Platform-level guardrails are essential. That’s why identity, observability, and policy enforcement become premium features.”

Microsoft has already begun addressing this with its Purview compliance tools, which now include AI-specific policies. Windows administrators can use Group Policy to control which agents can execute, what APIs they can call, and even throttle their token consumption. These controls, Huang argued, are what will make enterprise platforms sticky. “Once you’ve set up the guardrails, you’re not going to rip them out. The platform that provides the best security and governance will own the enterprise.”

What This Means for Software Engineers and IT Professionals

Huang’s vision also affects the job market. He was keen to dispel the notion that AI agents will replace developers. Instead, he predicted a surge in demand for platform architects who can design and maintain the systems that agents interact with. “Every agent is a new API consumer,” he said. “We need people to build and manage those APIs. We need people to ensure the data is clean and trustworthy. The complexity doesn’t go away—it shifts.”

That shift is already visible in Windows-centric IT roles. System administrators who once managed Group Policy and patching are now being asked to configure AI agent policies and monitor agent performance. Microsoft’s latest certifications include “AI Platform Operations” tracks. Developers are learning how to build and secure tools that agents can use, which often requires deep knowledge of Windows APIs, Azure Active Directory, and other core platform services.

Conclusion: A Reassuring Vision, with Nuance

Jensen Huang’s Computex 2026 keynote served as a powerful counter-narrative to the doom-and-gloom predictions about agentic AI. By arguing that the platform layer—the operating system, the cloud infrastructure, the data management tools—will become the most valuable real estate in the AI era, he provided a reassuring thesis for software giants and the professionals who build on their platforms. For Windows enthusiasts, the message is clear: rather than eroding Windows’ relevance, agentic AI could make the operating system more indispensable than ever, provided it evolves to be the secure, intelligent agent hub.

But Huang’s view is not without its critics. Some argue that the platform layer itself could be commoditized by open-source standards, much as Linux challenged Windows Server in the early 2000s. Others point out that if agents truly become the primary interface, application vendors will face immense pressure to unbundle features and compete on API quality rather than user experience. That could squeeze margins, even if overall value grows.

Nevertheless, Huang’s confidence is backed by Nvidia’s staggering growth and its deepening partnership with Microsoft. As the lines between hardware and software blur, the combined power of Windows’ desktop dominance and Nvidia’s compute supremacy could define the agentic future. For now, the message from Taipei is one of evolution, not extinction. Software companies that invest in their platform capabilities may find that agentic AI does not kill them—it makes them stronger.