Microsoft unveiled new specialized artificial intelligence models and enterprise tools at its Ignite conference on Tuesday, seeking to curb soaring AI compute costs by combining cloud processing with local device workloads. The announcements, delivered at the Moscone Center in San Francisco, emphasize targeted software development models and agentic automation designed to run securely across business environments.
At the core of the development updates is the debut of MAI Code 2 and MAI Code 2 Flash in GitHub Copilot, Microsoft confirmed on the GitHub Blog. Billed under usage-based pricing, the next-generation Microsoft AI models are designed for agentic software engineering tasks, extending beyond basic code completion to help software engineers execute complex development operations.
Hybrid Intelligence and Local Processing
To manage expense and efficiency, Microsoft is promoting what it calls “Hybrid Intelligence” across Windows, which automatically routes computational tasks between local hardware and cloud systems. The company stated on its Windows Blog that running tasks directly on hardware—such as developer PCs powered by Nvidia RTX Spark or AMD Ryzen AI Max processors—allows businesses to stretch token budgets and minimize reliance on expensive cloud data centers.
GitHub Copilot Fusion will leverage this hybrid framework by routing specific coding operations to local models, including MAI Code 1.1 Flash, on supported machines. Corporate partners such as Adobe, Dell, HP, Insight, Nvidia, and PwC have joined Microsoft’s Hybrid Intelligence pilot to test local model integration in development workflows.
“Adobe’s flagship Windows apps like Premiere and After Effects already use local AI to deliver creative value to our users,” said Ely Greenfield, chief technology officer and executive vice president of Adobe’s Creative Products, according to the Windows Blog. “Like many leading organizations, we’re looking to optimize growing AI spend, both ours and our customers’, without giving up access to frontier capabilities.”
On Copilot+ PCs, Microsoft is preparing local actions that allow Copilot to organize files or run read-only diagnostics on PC hardware without sending queries to the cloud. The system is designed to address routine technical issues directly on the device, reducing internal IT support tickets and cutting data transfer overhead.
Controlling AI Agents and Custom Models
Beyond code creation, Microsoft introduced updates to Microsoft Foundry aimed at measuring and reducing expenses. Foundry now includes an agent optimizer in general availability to test candidate software modifications against baseline evaluations for latency, safety, and operational costs, according to the company’s Book of News. Microsoft also previewed Managed Training and Programmed Training, allowing enterprise engineers to fine-tune open-weight models using custom business data alongside supervised fine-tuning and reinforcement learning.
The corporate software push comes as organizations navigate the security risks of agentic AI. To isolate autonomous systems, Microsoft highlighted Microsoft Execution Containers, known as MXC, which runs on Windows 11 to govern what files and networks agents can touch. Third-party developers such as Anthropic, OpenAI, Meta, Box, Perplexity, and Replit are building on or deploying MXC containers, Microsoft noted.
Additionally, the company introduced Windows 365 for Agents, enabling autonomous software to operate inside dedicated Cloud PCs to interact with legacy software and internal tools lacking application programming interfaces. Microsoft reported that its internal operations team cut order processing time by 60 percent using the Cloud PC agent setup.
What Comes Next
The Ignite conference is scheduled to run through Nov. 20, with Microsoft Chief Executive Officer Satya Nadella and commercial head Judson Althoff opening the event. As thousands of corporate attendees evaluate the new offerings, Microsoft faces the challenge of proving that its hybrid agent architecture and proprietary MAI models can deliver practical savings over competing cloud services.



