Microsoft's annual developer conference, Build 2026, officially commenced on June 2.
In contrast to previous years' focus on Copilot AI upgrades or cloud services, this year's event centered almost entirely around the core theme of "AI agents," spanning development tools, models, and local devices across all scenarios.
Microsoft CEO Satya Nadella stated at the conference, "We believe that enterprises should no longer merely consume cutting-edge models but should fully participate in the construction of the frontier model ecosystem."
Microsoft secured its place in the AI era through its early investment and deep partnership with OpenAI, which fueled a sustained rise in its stock price.
However, this binding relationship is now loosening.
On April 27, the two parties announced a revised agreement for the next phase of cooperation: OpenAI's IP licensing to Microsoft is no longer exclusive, allowing OpenAI to offer its full product suite to customers on any cloud platform, while Microsoft will no longer pay revenue shares to OpenAI.
As the "Microsoft-OpenAI alliance" shows clear signs of loosening, market concerns about Microsoft's lack of a proprietary foundational large model have grown.
With the boundaries of their cooperation being redrawn, Microsoft's dependence on OpenAI, coupled with the strength of its own in-house large models, has become a key metric for the market to evaluate its overall AI competitiveness.
Embracing an 'Agent-First' Philosophy
The shift is first occurring within the Windows operating system.
Pavan Davuluri, head of Microsoft's Windows and Devices division, explained that for the past four decades, Windows's core mission has been to run applications: users would open browsers, Office, and various enterprise software via mouse and keyboard to complete tasks.
But in the AI era, an increasing number of tasks will be handled by intelligent agents—users will no longer need to operate software directly; they can simply describe their goals to an AI agent, which will then execute them automatically.
From drafting emails and scheduling meetings to writing code and deploying applications, numerous workflows will be delegated to agents.
To this end, Microsoft is embedding agent capabilities directly into Windows Terminal, allowing developers to invoke AI for code writing, debugging, queries, and executing complex tasks without switching chat windows.
Additionally, Microsoft announced the integration of Linux container capabilities directly into the Windows system, enabling agents to more efficiently utilize relevant resources to complete tasks.
Addressing the Model Gap
Models are seen by the outside world as a weak link in Microsoft's ecosystem, making its first AI programming model, MAI-Code-1-Flash, a point of keen interest for developers.
According to Microsoft, this model is fully self-developed based on compliant, high-quality licensed data and is tailored to developers' daily workflows.
The model is now available in the personal edition of GitHub Copilot within VS Code.
Microsoft claims that in multiple benchmark tests, the model offers better cost-performance than Claude Haiku 4.5; real-world test data shows it can reduce token consumption by up to 60% when handling highly complex problems.
Hardware Innovations for AI Development
On the hardware front, one of the most anticipated products at the conference was the Surface RTX Spark Dev Box.
This mini development workstation is designed for developers, featuring NVIDIA's latest RTX Spark platform with a chip combining an Arm-architecture CPU and a Blackwell GPU, and equipped with 128GB of unified memory.
Microsoft stated that the Surface RTX Spark Dev Box can run large language models with up to 120 billion parameters locally.
Unlike traditional development computers, this device was designed from the ground up for AI development scenarios.
Microsoft pre-installs development tools on the device and has performed system-level optimizations.
In response to the vibrant OpenClaw ecosystem, Microsoft has officially announced its entry into this space.
At the conference, it launched the "Scout" agent built on OpenClaw.
Scout operates across cloud, desktop, and web, connecting Microsoft services like email, meeting applications, and cloud storage.
It can proactively filter emails and messages, automatically organizing items that require user decisions to help improve work efficiency.
Redefining the Computing Platform
Examining Microsoft's announcements at this conference—from the Surface RTX Spark Dev Box capable of running hundred-billion-parameter models locally, to the Windows AI development platform natively supporting Linux containers and Windows Terminal, the AI assistant Scout built on OpenClaw, Microsoft's first self-developed inference model MAI-Thinking-1, and several software and hardware products tailored for AI agent needs—all point towards the same objective: redesigning all products around the agent.
Regarding this, CEO Nadella believes this represents a true platform migration.
Microsoft is shifting from "building operating systems and devices for applications" to "developing products around agents."
In essence, Microsoft is betting on a new usage paradigm where users no longer open individual applications but instead invoke agents to perform tasks on their behalf.
Strategic Partnership with NVIDIA
Beyond its own products, NVIDIA's presence was also strongly felt at this year's Build conference.
During the event, Nadella held a transcontinental dialogue with NVIDIA founder and CEO Jensen Huang, who was attending Computex in Taipei.
The two leaders reviewed their long-term collaboration from DirectX and CUDA to large-scale Azure GPU clusters and clarified that their collaborative focus is now shifting to a "cloud-edge integrated" computing architecture.
This encompasses edge AI inference, the AI PC ecosystem, and liquid-cooled supercomputing data centers equipped with the Blackwell architecture, aiming to reduce computing costs and secure enterprise-level AI workloads.
Echoing the themes of the Build conference, "agents" were a hot topic at the concurrently held Computex in Taipei, with giants like NVIDIA, Intel, Qualcomm, and Arm collectively stating that the agent era has arrived.
The most significant announcement came from NVIDIA.
Jensen Huang announced the RTX Spark superchip, co-developed with Microsoft and MediaTek, with the goal of turning thin-and-light laptops and mini PCs directly into AI PCs capable of running agents locally.
Huang stated plainly that this is the result of NVIDIA and Microsoft's three-year collaboration to "reimagine the PC," with the chip jointly developed by the two companies and MediaTek.
Furthermore, NVIDIA unveiled the Vera CPU, touted as the "first CPU designed not for humans but for AI agents."
It claims to achieve approximately 1.8 times the task completion speed compared to x86, focusing on tasks like agent operations, reinforcement learning, and data orchestration.
While NVIDIA focuses on the AI computing platform, the series of software and hardware products launched by Microsoft at this Build conference precisely complement the software and ecosystem aspects of this blueprint.
In essence, Microsoft is building a complete software foundation for running agents.
It can be said that Microsoft is, to a certain extent, assuming the role of the "operating system provider" for the agent era: NVIDIA provides the AI computing platform, while Microsoft is responsible for constructing the software ecosystem for running agents.
Together, they are sketching a future vision of a computing platform redefined around intelligent agents.
Comments