Last week, Microsoft held its annual Build developer conference. In its keynote, the company presented a full, bottom-up technical framework for its AI initiatives, from the underlying technology to the way you interact with it to security and compliance.
Build's keynote was heavily developer centric, so a lot of the topics that Microsoft discussed aren't necessarily new products or features that typical users will be able to try out right now. But its announcements give us an idea of Microsoft's vision for AI and how it may fit into your work.
And in many ways, the keynote was a stark contrast to what we saw at Google I/O a couple weeks ago. While Google's presentation was sprawling, Microsoft's was much more structured. Google focused on a broad range of audiences and use cases while Microsoft focused more narrowly on software development and enterprise IT. And for Microsoft, there was one big star of the show: AI agents.
Disclosure: Microsoft is a TechSoup partner and funder. We collaborate with Microsoft on a number of initiatives, including the Microsoft for Nonprofits program. This blog post was written and produced independently of those initiatives and was not subject to Microsoft's approval.
Microsoft's Big Focus: AI Agents
AI agents are a hot buzzword in tech circles right now: They're effectively automated AI assistants that can carry out tasks on your behalf. And Microsoft sees them as a large part of the future of work, a vision that permeated the entire keynote.
Examples of Microsoft's focus on agents include a new OpenClaw app for Windows — built around the increasingly popular open-source framework for "personal AI assistants" — as well as a GitHub app for software developers that makes it easier to manage coding agents all in one place.
Agents aren't new to Microsoft's tools — the company discussed them at length at last year's Build conference as well — but it is focusing on bringing a new generation of agentic AI that can do more on its own, without any human intervention.
Meet Scout, Your AI Assistant
These more capable agents are also coming to Microsoft 365 Copilot by way of a new feature known as Copilot Autopilots. According to Microsoft CEO Sattya Nadella, these are "autonomous, long-running agents with full enterprise compliance that run in your [Microsoft 365] tenant." In a blog post, Microsoft says:
"Autopilots stay active in the background, understand how work gets done across your apps and systems, and take action without needing to be prompted each time." Because they operate with their own identity, they can carry out tasks within the permissions and policies you and your organization set. This creates a more durable way to keep work in motion so it continues even when your attention is elsewhere."
The company' first Autopilot is Scout, a "personal agent" that can carry out tasks for you at work that's built atop OpenClaw. According to Microsoft, it can work with your Microsoft 365 data: Outlook email and calendars, files in OneDrive and Sharepoint, chats in Teams, and so on. For example, you can set up Scout to act as a digital administrative assistant of sorts and have it manage your calendar on your behalf.
The utility is clear if you're into these sorts of personal AI agents, and organizations can get an early look if they meet the preview program's requirements. A dose of caution may be in order in light of an internal document that 404 Media obtained that indicated that the company wants to get people hooked on Scout. For his part, Nadella pushed back against that design goal in an internal memo, The Information reports.
Agent-Infused Hardware: An Idea with Legs or a False Start?
At Build, Microsoft announced Project Solara, a somewhat nebulous initiative to produce agent-centric hardware devices. According to Microsoft, its goal is to develop a range of devices that bring AI agents to new places, to the "nooks and crannies where conventional computers either do not exist or are not optimal," Microsoft's Steven Bathiche said in the keynote.
The idea here is that your agents can go with you wherever you are, using whatever device is best for a given situation. It would all use a common underlying software platform, so you could create one set of agents that you can use anywhere. In the keynote, Microsoft presented two example devices, an advanced key-card device and a desktop "hub," not too dissimilar in appearance from devices like the Echo Hub from Amazon. To get a sense of what Microsoft has in mind, I'd recommend checking out this video clip from the keynote.
To be clear, these aren't devices that you can run out and buy today. Instead, it's an underlying software platform that hardware manufacturers can use to build their own devices.
Microsoft isn't the first company to try and bring AI to new kinds of devices. For example, a startup named Humane released its AI Pin in 2024, and Apple's former Chief Design Officer, Jony Ive, partnered with ChatGPT to develop AI-centric products. Ive's initiative has yet to bear fruit, and Humane closed up shop in February 2025 after the AI Pin met scathing reviews.
It's also an open question whether organizations — particularly nonprofits — will have the tech budget to spend on such devices.
AI Agent Security
Microsoft spent a significant amount of time discussing measures that it's taking to help keep AI and agents in check. A lot of it was technical and geared more toward developers, but one announcement in particular is worth looking at in more detail. And that's Microsoft Execution Containers, or MXC for short.
MXC is a new "sandbox" technology that's designed to contain and constrain AI agents. In AI tools that use it, MXC will allow you to set permissions for what an AI agent can and can't do on your PC. For example, in a demo of the OpenClaw app for WIndows, the presenters showed how an IT admin can prevent AI agents from carrying out certain tasks. In the case of this demo, they prevented it from deleting files from the Desktop.
MXCs are enforced at the policy level, so IT admins can determine the appropriate level of agent access across the organization. MXC technology is still in the preview stage of development, so it'll be a while until you see it employed in the real world.
If you'd like to get into the nitty-gritty of MXCs, check out this post on the Windows Developer Blog. And if you'd like a wrap-up on all the security-related announcements at Build, head over to the Microsoft Security website.
More Efficient Models a Priority
At points throughout the keynote, Microsoft acknowledged the need to make AI models more efficient. It placed a heavy focus on "token efficiency," or getting more done with fewer chunks of data to be processed by its AI models.
Microsoft also claimed that its own AI models are up to 10 times more cost-effective than comparable models from other vendors.
Nadella also spoke briefly about Microsoft's efforts around improving data center efficiency — a hot-button issue in many communities. For example, one data center that Microsoft discussed takes a "fill-once" approach to its cooling systems. Rather than requiring constant water intake, it's designed to draw water from local water systems only once. Nadella says it requires about as much water as a restaurant would need in one year.
There's a lot of work to do here in terms of minimizing AI's impact on communities and the environment. But it's encouraging to see Microsoft acknowledge the issue here.
More Powerful Hardware Hints at a Hybrid AI Future
Microsoft announced the new Surface RTX Spark Dev Box, a developer-oriented PC built around Nvidia's new RTX Spark system-on-a-chip. Nadella also touted Nvidia's announcement of the DGX Station for Windows, a computer that Nadella characterized as a "desktop data center." According to Nadella, it's capable of running models akin to OpenAI GPT versions 2.5 or 3 — not quite as sophisticated as today's AI models, but the sort of thing that required full-fledged data centers to run only a few years ago.
As hardware becomes more powerful, it seems reasonable to believe that in the future, you'll be able to process many more AI functions on your own PC, leaving data centers to run tasks that require more serious number-crunching. In fact, this is the architecture that Apple is already using in its Apple Intelligence AI technology.
What's Next: Quantum Computing, AI-Aided Research
To close out the keynote, Microsoft announced its latest step forward in quantum computing, a chip called Majorana 2. According to Microsoft, Majorana 2 is far more reliable than its predecessor, Majorana 1, and the company claims that it's "on a path to achieve a scalable quantum computer that is commercially valuable by 2029."
Quantum computing employs quantum mechanics to process calculations, and it could be far faster at certain tasks than traditional computers. It's a technology that researchers have been working toward for many years, but it's not quite ready for prime time for a variety of reasons. That said, Microsoft says that "such a machine could tackle intractable problems in global health, food supply, sustainability, energy production and more."
Microsoft also released a new AI-powered research tool, Microsoft Discovery, to the general public. It's a platform that uses agents to assist organizations with research and development. You can, for example, ask it to pull together the latest peer-reviewed research on ocean microplastics, or have it run an analysis of data from an experiment. If you're curious, you can learn more at the Microsoft Discovery website.
It's easy to be skeptical of AI's role in research and the sciences. I'm far from an expert in that field and my gut tells me that a dose of caution — a "trust but verify" approach — is still in order. But AI is already contributing significantly to this space. In 2024, Harvard researchers devised an AI model that could help detect and diagnose cancer. And just last month, a ChatGPT chatbot solved a problem that mathematicians had been trying to solve for 80 years.
AI for Humans — Not the Other Way Around
One theme that I noticed throughout the keynote was the way Microsoft Microsoft framed AI as a supplement to — not a replacement for — human thought and insight.
Yes, the keynote covered AI-based "vibe coding" and included a brief mention of new AI models that can generate or modify content. But for the most part, Microsoft positioned AI as an assistant, as a way to bring data together from various sources within an organization and apply learnings from it, as another tool — albeit a powerful one — in your toolbox.
"There are really two stories which people can tell about this moment," Nadella said at the end of the keynote. "One is that technology concentrates power, reduces human agency, and leaves the society to absorb the consequences. The other is that we use this next wave to unlock opportunity for developers, scientists, enterprises, and every community. And our job is to make the second story true."
Here's hoping that these words shape a shared vision for the future.
If you'd like to learn more about all of Microsoft's announcements at Build, check out this playlist of videos from the keynote.
[Thumbnail photo: Shutterstock]
