During Microsoft’s annual Build conference on Tuesday, the tech giant unveiled a variety of new and enhanced AI projects, showcasing everything from a super app to proprietary reasoning models, a cybersecurity solution, and AI agents reminiscent of OpenClaw. This collection of announcements sends a clear signal: Microsoft is ready to assert itself as a formidable competitor in the AI landscape, and it’s finally stepping into the spotlight.
Historically, Microsoft’s AI efforts heavily relied on its initial, exclusive partnership with OpenAI. However, that collaboration has shifted over time, leading to a practical separation as of late April, although Microsoft continues to serve as OpenAI’s primary cloud provider—for the moment. This year's Build conference felt like a newly single individual rejuvenated and ready to impress. “Developer conferences are particularly exciting during times of significant change,” Microsoft CEO Satya Nadella noted while addressing attendees, emphasizing the new opportunities that arise in such periods.
In a candid conversation with The Verge, AI lead Mustafa Suleyman didn't mince words about Microsoft’s ambitions. “Our objective is to establish ourselves among the top four AI labs globally,” Suleyman stressed. “Currently, we’re not in that league with the likes of Google DeepMind, OpenAI, and Anthropic, but that’s the goal I’m here to achieve. I want to build highly advanced multimodal models from the ground up, without relying on others.”
One of the notable announcements from Build was the introduction of MAI-Thinking-1, Microsoft’s inaugural reasoning model, alongside six supplementary models geared towards image processing, voice recognition, transcription, and coding. This medium-sized model is aimed primarily at enterprise clients and is specifically designed for complex tasks involving math and coding. Although Microsoft faces a significant gap—OpenAI launched reasoning models as early as fall 2024—Suleyman highlighted MAI-Thinking-1’s competitive pricing, claiming it performs better on numerous benchmarks than OpenAI’s models, especially critical in the current economic climate affecting AI investments.
Suleyman emphasized that MAI-Thinking-1 was developed entirely independently, without using insights from OpenAI’s models. He pointed out that “the pivotal moment” for Microsoft was the renegotiation of its contract with OpenAI, allowing the company to pursue large-scale model training exclusively using its own intellectual property and data.
Nadella also spotlighted the recently introduced AI cybersecurity tool, MDASH, which incorporates 100 AI agents to identify exploitable vulnerabilities more effectively than any individual model. This announcement seemed to challenge Anthropic’s recently introduced Claude Mythos Preview, as well as OpenAI’s competing solutions, as all major players look to secure their foothold in both governmental and enterprise sectors.
When it comes to AI agents, Microsoft is adopting a multi-faceted strategy. The popular open-source platform OpenClaw has paved a potential direction for these agents, and following OpenAI’s swift acquisition of its creator, Microsoft now finds itself racing to keep pace. During Build, Nadella affirmed the company's intent to integrate OpenClaw’s capabilities within Windows. Developers were actively engaged in discussions about how they were leveraging OpenClaw.
Steinberger, the developer behind OpenClaw, made an unexpected appearance, eliciting enthusiasm from the crowd as he discussed the platform’s improvements in security and user trust. “Users kept asking if they could use it in their workplaces,” Steinberger mentioned, noting that OpenClaw is now compatible with various company settings and can even be operated as a plug-in.
In parallel, Microsoft is rolling out its own Copilot “super app,” which will incorporate OpenClaw-like agents. This super app concept is a significant focus for OpenAI, with Greg Brockman at the helm of a development initiative tying together ChatGPT, the Codex coding platform, and the Atlas web browser. Microsoft’s version similarly seeks to unify various existing Copilot AI assistants, assigning its agents—termed “Autopilots”—the role of supportive user interfaces. Cassidy Williams, GitHub’s senior director of developer advocacy, described Copilot as “your home base for development and operations on your computer,” exemplifying how multiple agents can collectively execute tasks like application development.
The Autopilot agents are explicitly designed for business clientele, with Nadella noting their autonomous capabilities compliant with enterprise standards. The initial Autopilot offering, named “Scout,” will serve as an “always-on personal agent,” but clients can customize their own versions. These agents will manage tasks such as email sorting, group chat participation via Teams, calendar checks, and daily briefings. Throughout the event, Microsoft staff continuously underscored Copilot’s security features aimed at reassuring enterprise customers wary of potential AI mishaps.
Suleyman reiterated Microsoft’s vision of a “humanist superintelligence”—an AI framework prioritizing human welfare—a strategic rebranding amid growing public apprehension about AI.
Additionally, Nvidia CEO Jensen Huang, who is closely associated with OpenAI, joined via video to showcase how Nvidia’s RTX Spark chip is enhancing Microsoft’s AI initiatives. “Imagine texting my PC while traveling, and it activates necessary tools—it exemplifies the transition from personal computers to personal AI,” Huang remarked.
While years spent developing its relationship with OpenAI have positioned Microsoft uniquely in the AI race, the competitive landscape remains intense. As OpenAI and others pivot towards enterprise revenue generation, Microsoft’s advantages are clear: an extensive client base, a reputation for reliability, and substantial funding. Much like Google, Microsoft has ample resources, allowing it to explore ambitious projects with minimized risk.
Suleyman expressed, “Amidst a search for startup valuations or impending IPOs, we can afford to operate with a sense of humility and focus on long-term goals.” He added, “We have the capacity to acquire models from companies like Anthropic as needed, and Azure’s vast selection of 11,000 models offers users flexible options, granting us the time to execute our initiatives effectively.”
However, uncertainty looms over several aspects. Although Microsoft highlighted numerous benchmark successes and advancements regarding its new models, translating that into real-world application remains a significant challenge. The concept of AI super apps is still largely theoretical, and the company is entering an already crowded AI agent market, pushing forth a product yet to be effectively demonstrated. Consequently, while aspirations are grand, the potential for these initiatives to falter is still a pressing reality.
