Microsoft Build 2026: See everything Microsoft announced at the event.

A Microsoft Build 2026 Microsoft has placed AI agents at the center of its strategy. During the keynote, Satya Nadella, the company's CEO, introduced a new family of MAI models (Microsoft AI), a platform for agents across multiple devices, Microsoft Scout for work, new security layers in Windows, and hardware with NVIDIA RTX Spark to run AI locally.

Below, we've compiled the main new features presented at the conference, explaining what's changing for users, businesses, and developers. As some of the announcements are still in preview, availability and pricing in Brazil have not yet been confirmed by Microsoft.

The MAI family: Microsoft's first proprietary AI models.

Microsoft has introduced the new family. MAY, a line of AI models created to reduce dependence on external solutions and offer proprietary options within the Microsoft ecosystem. The main announcement was the MAI-Thinking-1, described by the company as its first reasoning model, with 35 billion parameters assets and context window of 128 thousand tokens.

In practice, a reasoning model is geared towards tasks that require multiple steps, such as writing code, analyzing lengthy documents, planning workflows, and solving complex problems. According to the coverage of MashableMicrosoft also highlighted the lower cost per token compared to similar models, an important point for companies that want to scale AI without blowing their budget.

• MAI-Thinking-1: reasoning, complex instructions, long context, and code generation;
• MAI-Image-2.5: Image generation, with Flash variant;
• MAI-Transcribe-1.5: Transcript in multiple languages;
• MAI-Voice-2: voice, also with Flash variant;
• MAI-Code-1: According to Mashable, this model focuses on programming and is available in tools such as Copilot and VS Code.

The models will be integrated into Microsoft Foundry and to products like PowerPoint, OneDrive, and Copilot. For the end user, this means that creation, summarizing, voice, and scheduling features can become faster, cheaper, and integrated into Microsoft applications.

Related official video: Microsoft has released a video dedicated to the new AI models.

Project Solara: a platform for agents across multiple devices.

Another strong point of the keynote was the Project Solara, described by Mashable as an Android-based platform to run multiple AI agents in a secure environment. The idea is to go beyond the traditional PC: agents could accompany the user across different devices, maintaining context and performing tasks without constantly relying on a Windows window.

Satya Nadella Microsoft showcased device concepts, including a wearable accessory with a Qualcomm chip and a desktop device to track workflows. The proposal still sounds experimental, but it reveals Microsoft's direction: transforming agents into a continuous computing layer, not just chatbots within an application.

For companies, the value would lie in enabling agents to help think, plan, and execute tasks in controlled environments. For consumers, the promise is of AI being more present in everyday life, although details regarding availability, price, and the final form of the products are still lacking.

Related official video: Microsoft's conversation with Christian AmonCEO of QualcommThis helps to contextualize the partnership behind the new devices.

Microsoft Scout and Autopilots: always-on agents for the job.

Microsoft also presented the AutopilotsA new category of always-on agents, with their own identity and the ability to act on behalf of the user within the permissions defined by the organization. The first example is... Microsoft Scout, integrated with Microsoft 365.

Different from one chatbot who is waiting for a question, the Scout It was designed to operate in the background. It can observe information from Teams, Outlook, OneDrive, SharePoint, calendar, emails e contacts To anticipate tasks: prepare for meetings, suggest schedules across time zones, identify pending deliverables, reserve blocks in the agenda, and address stalled decisions before they become bottlenecks.

Microsoft claims that the Scout use the Work IQ, the intelligence layer of Microsoft 365 Copilot, and open technology of OpenClawAvailability begins in a limited way: the official blog informs about experimental access through the program. Frontier, with requirements such as enrollment in Frontier, policies via Intune, attestation opt-in, and a GitHub Copilot license for experience installation.

Related official video: Microsoft discussed Autopilots during the keynote; the overview video of the models also appears connected to Scout and the new agents.

Windows, OpenClaw, and containers: security for AI agents.

Microsoft wants to position Windows as an "agent-native" environment. To that end, it has presented the idea of... Microsoft Execution ContainersIn preview, these areas function as isolated zones for agents to execute code, access files, and interact with networks with less risk to the main system.

This layer is important because autonomous agents can execute multi-step flows and take actions on behalf of the user. If this execution happens in uncontrolled sessions, it increases the risk of data leaks, unauthorized access, or unauthorized actions. With operating system-enforced containers, IT administrators can define limits and policies with more predictability.

In the case of ScoutMicrosoft also promises identity governed by Entra, protected credentials, specific permissions, human approval for sensitive actions, and enforcement of Microsoft Purview policies, such as sensitivity labels and data loss prevention.

Related official video: The Microsoft Developer channel published a specific video about OpenClaw and Windows at Build 2026.

Surface RTX Spark: local AI computer with NVIDIA

The partnership with NVIDIA appeared in the advertisement of Surface RTX Spark Dev Box, called by Mashable Surface Ultra in its coverage. The product was presented as a machine for developers who need to run AI workloads locally, without constantly relying on the cloud.

According to Mashable, the hardware uses the new chip. NVIDIA RTX Spark, offers up to 1 petaflop of AI performance and can reach 128 GB of unified memoryThe promise is to run models with up to 120 billion parameters on the device itself, something relevant for teams working with prototypes, agents, automation, and private testing.

For the Brazilian audience, the key point is understanding the concept: PCs with local AI keep some of the processing within the machine, which can reduce latency, improve privacy, and allow work even when the connection to the cloud is limited. There is still no price or estimated arrival date in Brazil.

Related official video: The Microsoft Surface channel published the presentation of the Surface RTX Spark Dev Box.

Microsoft Discovery and scientific research with AI.

The supplementary coverage of Mashable also mentions Microsoft DiscoveryMicrosoft's scientific research platform, which has become generally available, aims to use AI to accelerate steps such as hypothesis formulation, data analysis, simulations, and collaboration among researchers.

This type of tool targets laboratories, universities, and R&D teams working with materials, pharmaceuticals, energy, chemistry, and other areas where scientific discovery involves extensive testing. Instead of replacing scientists, AI acts as a support layer to prioritize promising paths and reduce trial-and-error cycles.

Although the announcement is more technical, it helps explain why Microsoft insists so much on agents and infrastructure: the company wants to sell AI not just as a text assistant, but as a platform for work, code, science, and business automation.

Related official video: We did not find a specific official Microsoft video for Microsoft Discovery in the keynote excerpt; therefore, we are keeping the overall keynote video as a reference.

Majorana 2: A new step towards quantum computing.

Nadella also mentioned the Majorana 2, a quantum chip that, according to coverage by Mashable based on BloombergThis is part of Microsoft's goal to develop a usable quantum computer by 2029. Quantum computing uses principles of quantum physics to solve certain types of problems that would be very difficult for conventional computers.

It's still a long-term bet. Unlike the announcements of the MAI, Scout, and Surface RTX Spark models, which already have previews or products in development, quantum computing remains surrounded by technical uncertainties. Even so, the mention shows that Microsoft wants to connect AI, cloud computing, specialized hardware, and advanced research in a single strategy.

Related official video: We didn't find a specific official video about Majorana 2 in the Microsoft Build 2026 video list; the topic appears as a mention within the keynote.

What does this mean for users and businesses?

A Build 2026 This reinforces a shift: Microsoft wants AI to move beyond being just a feature within Copilot and become an operational layer for products, devices, and infrastructure. For users, this could bring more proactive assistants to Microsoft 365. For businesses, the promise is to automate tasks with greater governance. For developers, the focus is on proprietary models, Foundry, local execution, and secure environments for agents.

The key is to separate promises from actual availability. Several announcements are still in preview, are experimental, or have no defined price or global launch date. Even so, the direction is clear: Microsoft is preparing Windows, Microsoft 365, Surface, Foundry, and partnerships with NVIDIA and Qualcomm for a world where AI agents perform end-to-end tasks.

See also:

Microsoft announces Copilot Actions and Copilots Agents for AI-powered task automation

NVIDIA RTX Spark: New CPU promises to transform Windows PCs into local AI machines.

Sources: Mashable, Mashable — MAI templates, Mashable — Microsoft Scout, Mashable — Surface RTX Spark, Microsoft 365 Blog, Microsoft on YouTube, Microsoft Developer on YouTube e Microsoft Surface on YouTube.

reviewed by Luis Antonio Costa in 03 / 06 / 26