Construct and scale AI brokers with Microsoft Foundry

Brokers are reshaping software program. Not mere assistants, they’re changing into dynamic collaborators layered inside our on a regular basis instruments.

One yr in the past, at Microsoft Ignite, we got down to redefine enterprise intelligence with Foundry. Our conviction was clear: software program would evolve past inflexible workflows, changing into methods that purpose, adapt, and act with objective. We envisioned builders shifting from prescriptive logic to shaping clever habits.

At this time, that transformation is accelerating throughout industries and organizations of each dimension. The shift is tangible: brokers are now not simply assistants, they’re dynamic collaborators, seamlessly built-in into the instruments we use each day. For builders, brokers are reshaping software program, and we’re delivering a platform that empowers each developer and each enterprise to embrace this second with confidence and management.

Microsoft Foundry helps builders all over the place flip imaginative and prescient into actuality with a modular, interoperable, and safe agent stack. From code to cloud, at this time demonstrates our concentrate on empowering builders with a robust, easy—and trusted—path to manufacturing AI apps and brokers. Right here is the TL;DR:

1. Foundry Fashions added new fashions from Anthropic, Cohere, NVIDIA, and extra. Mannequin router is now typically obtainable in Microsoft Foundry and in public preview in Foundry Agent Service. 
2. Foundry IQ, now in public preview, reimagines retrieval-augmented era (RAG) as a dynamic reasoning course of, simplifying orchestration and enhancing response high quality. 
3. Foundry Agent Service now affords Hosted Brokers, multi-agent workflows, built-in reminiscence, and the flexibility to deploy brokers on to Microsoft 365 and Agent 365 in public preview. 
4. Foundry Instruments, empowers builders to create brokers with safe, real-time entry to enterprise methods, enterprise logic, and multimodal capabilities.
5. Foundry Management Airplane, now in public preview, centralizes id, coverage, observability, and safety indicators and capabilities for AI builders in a single portal. GitHub Superior Safety and Microsoft Defender integration, now in public preview, helps enhance collaboration between safety and improvement groups throughout the complete app lifecycle.
6. Foundry Native, now in personal preview on Android, the world’s most generally used cell platform.
7. Managed Occasion on Azure App Service, now in public preview, helps organizations transfer their net purposes to the cloud with only a few configuration adjustments. 

All of it begins with builders, and GitHub is the world’s largest developer group, now serving over 180 million builders. AI-powered instruments and brokers in GitHub are serving to builders transfer quicker, construct more and more revolutionary apps, and modernize legacy methods extra effectively. Greater than 500 million pull requests have been merged utilizing AI coding brokers this yr, and with AgentHQ, coding brokers like Codex, Claude Code, and Jules can be obtainable quickly immediately in GitHub and Visible Studio Code so builders can go from concept to implementation quicker. GitHub Copilot, the world’s the preferred AI pair programmer, now serves over 26 million customers, serving to organizations like Pantone, Ahold Delhaize USA, and Commerzbank streamline processes and save time.

During the last yr, builders have moved from experimentation to manufacturing. They want instruments that allow them design, check, monitor, and enhance clever methods with the identical confidence they’ve in conventional software program. That’s why we constructed a brand new era of AI-powered instruments: GitHub Agent HQ for unified agent administration, Customized Brokers to encode area experience, and “convey your individual fashions” to empower groups to adapt and innovate. With Copilot Metrics, groups evolve with information, not guesswork.

We’re dedicated to giving each developer the instruments to design, check, and enhance clever methods, to allow them to flip concepts into impression, quicker than ever. Managed Occasion on Azure App Service, now in public preview, lets organizations transfer current .NET purposes to the cloud with only some configuration adjustments.

Enter Microsoft Foundry: The AI app and agent manufacturing unit

Enterprises want a constant basis to construct intelligence at scale. With Microsoft Foundry, we’re unifying fashions, instruments, and data into one open system, empowering organizations to run high-performing agent fleets and clever workflows throughout their enterprise.

At this time, groups can select from over 11,000 frontier fashions in Foundry, together with optimized options for scale and specialised fashions for scientific and industrial breakthroughs. I’m proud to announce Rosetta Fold 3, a next-generation biomolecular construction prediction mannequin developed with the Institute for Protein Design and Microsoft’s AI for Good Lab. Fashions like these allow researchers and enterprises to sort out the world’s hardest issues with state-of-the-art expertise.

Right here is our high Ignite information for Foundry:

1. Use the best mannequin for each activity with Foundry Fashions

Innovation thrives on adaptability and selection. With greater than 11,000 fashions, Microsoft Foundry affords the broadest mannequin choice on any cloud. Foundry Fashions empowers builders to benchmark, examine, and dynamically route fashions to optimize efficiency for each activity.

At this time’s bulletins embody:

• Beginning at this time, Anthropic’s Claude Sonnet 4.5, Opus 4.1, and Haiku 4.5 fashions can be found in Foundry, advancing our mission to provide prospects alternative throughout the trade’s main frontier fashions, and making Azure the one cloud providing each OpenAI and Anthropic fashions. Additionally this week, Cohere’s main fashions be a part of Foundry’s first-party mannequin lineup, offering final mannequin alternative and adaptability.
• Mannequin router (typically obtainable) permits AI apps and brokers to dynamically choose the best-fit mannequin for every immediate—balancing price, efficiency, and high quality. Plus, in mannequin router in Foundry Agent Service (public preview), permits builders to construct extra adaptable and environment friendly brokers; notably useful for multi-agent methods. mannequin router in Foundry Agent Service (public preview), permits builders to construct extra adaptable and environment friendly brokers; notably useful for multi-agent methods.
• A brand new Developer Tier (public preview) makes mannequin fine-tuning extra accessible by leveraging idle GPU capability.

2. Empower brokers with data utilizing Foundry IQ

The extra context an agent has, the extra grounded, productive, and dependable it’s more likely to be. Foundry IQ, now obtainable in public preview, reimagines retrieval-augmented era (RAG) as a dynamic reasoning course of somewhat than a one-time lookup. Powered by Azure AI Search, it centralizes RAG workflows right into a single grounding API, simplifying orchestration and enhancing response high quality whereas respecting person permissions and information classifications.

Key options embody:

• Simplified cross-source grounding with no upfront indexing.
• Multi-source choice, iterative retrieval, and reflection to dynamically enhance the standard of agent interactions.
• Foundry Agent Service integration to complement agent context in a single, observable runtime.

Foundry already powers greater than 3 billion search queries per day. By combining Foundry IQ with Microsoft Cloth IQ and Work IQ from Microsoft 365 Copilot, Microsoft gives an unparalleled context layer for brokers, serving to them join customers with the best data on the proper time to make knowledgeable choices.

3. Construct context-aware, action-oriented brokers with Foundry Agent Service

To be drive multipliers, brokers want entry to the identical instruments and data because the individuals they assist. Foundry Agent Service empowers builders to create refined single and multi-agent methods, connecting fashions, data, and instruments right into a single, observable runtime.

At this time’s bulletins embody:

• Hosted Brokers (public preview) allow builders to run brokers constructed with Microsoft frameworks or third-party frameworks in a completely managed surroundings, to allow them to concentrate on agent logic somewhat than operational overhead.
• Multi-agent workflows (public preview) coordinate specialised brokers to execute multi-step enterprise processes utilizing both a visible designer or a code-first API. Workflows allow long-running, stateful collaboration with restoration and debugging built-in.
• Reminiscence (public preview) permits brokers to securely retain context throughout periods, lowering exterior data-store complexity and enabling extra customized interactions out-of-the-box.
• Microsoft 365 and Agent 365 integration (public preview) permits builders to immediately deploy brokers from Foundry to Microsoft productiveness apps, making it simpler to achieve customers immediately throughout the M365 ecosystem whereas leveraging Agent 365 for safe orchestration, governance, and enterprise-grade deployment.

4. Allow brokers to take motion utilizing Foundry Instruments

The fitting instruments can remodel brokers from easy responders into clever problem-solvers. With Foundry Instruments, now in public preview, builders can present brokers with safe, real-time entry to enterprise methods, enterprise logic, and multimodal capabilities to ship enterprise worth.

Now, builders can:

• Discover, join, and handle public or personal MCP instruments for brokers from a single, safe interface.
• Allow brokers to behave on real-time enterprise information and occasions utilizing greater than 1,400 connectors with enterprise methods similar to SAP, Salesforce, and UiPath.
• Enrich workflows with out-of-the-box instruments similar to transcription, translation, and doc processing.
• Expose any API or perform as an MCP software by way of API Administration, reusing current enterprise logic to speed up time-to-value.

5. Advancing safety and belief with Foundry Management Airplane 

Scaling intelligence requires belief. As organizations depend on brokers and AI powered methods for extra of their workflows, groups want clearer visibility, stronger guardrails, and quicker methods to determine and deal with danger. This yr we’re increasing safety and governance with two key bulletins: Foundry Management Airplane, now in public preview in Microsoft Foundry, and a brand new integration between Microsoft Defender for Cloud and GitHub Superior Safety, additionally in public preview. Collectively they provide builders and safety groups a extra linked strategy to monitor habits, information entry, and preserve AI methods protected throughout the complete lifecycle.

Foundry Management Airplane brings id, controls, observability, and safety collectively in a single place so groups can construct, function, and govern brokers with confidence. Key capabilities embody: 

• Controls that apply unified guardrails throughout inputs, outputs, and gear interactions to maintain brokers centered, correct, and inside outlined boundaries. 
• Observability with in-built evaluations, OTel based mostly tracing, steady pink teaming, and dashboards that floor insights on high quality, efficiency, security, and value. 
• Safety anchored in Entra Agent ID, Defender, and Purview to offer sturdy id, coverage pushed entry, built-in information safety, and real-time danger detection throughout the agent lifecycle. 
• Fleet vast operations that unify well being, price, efficiency, danger, and coverage protection for each agent, irrespective of the place it was constructed or runs, with alerts that floor points the second they seem empowering builders to take motion. 

Defender for Cloud + GitHub Superior Safety integration 

Builders and safety groups usually work in separate instruments and lack shared indicators to prioritize dangers. The brand new Defender for Cloud and GitHub Superior Safety integration closes this hole. Builders obtain AI instructed fixes immediately inside GitHub, whereas safety groups observe progress in Defender for Cloud in actual time. This provides either side a quicker, extra linked strategy to determine points, remediate them, and preserve AI methods safe all through the app lifecycle. 

6. Foundry Native Involves Android: Powering Cloud to Edge  

Six months in the past, we launched Foundry Native on Home windows and Mac. In that brief time, it’s reached 560 million units, making it one of many fastest-growing runtimes in enterprise historical past. Main organizations like NimbleEdge, Morgan Stanley, Dell, and Items are already utilizing Native to convey intelligence immediately into the environments the place work occurs, from monetary companies to healthcare and edge computing.

At this time, we’re taking the subsequent step. Foundry Native is now in personal preview on Android, the world’s most generally used cell platform. This implies brokers can run natively on billions of telephones, unlocking real-time inference, privacy-aware computation, and resilience, even the place connectivity is unpredictable. 

We’re additionally saying a brand new partnership with PhonePe, considered one of India’s fastest-growing platforms. Collectively, we’ll convey agentic experiences into on a regular basis client purposes, displaying how Native can remodel not simply enterprise workflows, however day by day life at large scale. 

7. Modernize your net apps for the period of AI in weeks, not months 

We see prospects constructing web new AI purposes and integrating AI into current purposes. Each require a contemporary basis. Managed Occasion on Azure App Service, obtainable in public preview, lets organizations transfer their .NET net purposes to the cloud with only a few configuration adjustments, saving the effort and time of rewriting code. The result’s quicker migrations with decrease overhead, and entry to cloud-native scalability, built-in safety and AI capabilities in Microsoft Foundry.

Be taught extra and get began with Foundry

We hope you be a part of us at Microsoft Ignite 2025, in-person or nearly, to see these new capabilities in motion and find out how they’ll assist your greatest ambitions for your enterprise.