Microsoft’s Build 2026 conference opened today in Seattle and online, and the announcements signal a significant shift in how Microsoft positions its AI development tooling for professional engineering teams. The two most consequential changes for organisations running Azure or GitHub are the replacement of the underlying model in GitHub Copilot and a substantial expansion of model choice in Azure AI Foundry.
Project Polaris: Microsoft builds its own coding model
The headline announcement is Project Polaris, a proprietary mixture-of-experts AI model developed in-house by Microsoft Research and the GitHub Copilot team. From August 2026, Polaris will replace GPT-4 Turbo as the default model powering GitHub Copilot across all paid plans.
Microsoft has not released benchmark details publicly, but internal evaluations cited during the keynote claim Polaris outperforms GPT-4 Turbo on multi-file refactoring tasks, repository-level code understanding, and test generation accuracy. The mixture-of-experts architecture is designed to reduce inference latency for the short, iterative completions that dominate real-world Copilot use, while routing more complex multi-step tasks to larger model components automatically.
For engineering teams, the change is largely transparent – Copilot’s interface and IDE integrations remain identical. The meaningful implication is that Microsoft is no longer dependent on OpenAI model timelines for its flagship developer product. GitHub Enterprise organisations should review their Copilot rollout plans, as the August transition may warrant a brief internal evaluation period before broad deployment.
Azure AI Foundry expands the European model catalogue
Azure AI Foundry received several updates directly relevant to European development teams:
Mistral AI models are now available in Foundry. Mistral is a French company, and its models are subject to French and EU data governance frameworks. For European organisations with strict GDPR requirements or sector-specific data localisation rules (DORA, NIS2, healthcare regulations), the ability to use a Mistral model through an Azure-managed endpoint with EU data residency is a meaningful compliance advantage over consuming the model directly from Mistral’s API or from a US-hosted provider.
Native multimodal support is now built into Azure AI Foundry workflows, allowing developers to chain text, image, audio, and video inputs without custom orchestration code. Previously, multimodal pipelines required combining separate models and managing the handoff manually.
A visual RAG designer has been added to Foundry Studio, enabling teams to build retrieval-augmented generation pipelines with a drag-and-drop interface connected to Azure AI Search, Cosmos DB, and external data sources. This significantly lowers the barrier to deploying document-grounded AI applications – a common pattern in legal, financial, and compliance workflows.
Cohere and Stability AI have also been added to the Foundry model catalogue, broadening the options available for embedding, reranking, and image generation tasks.
Windows as an agent platform
A broader theme running through Build 2026 is Microsoft’s positioning of Windows as an “agent platform.” A unified AI SDK bundles ONNX Runtime and DirectML 2.0, enabling local model inference on Windows devices without cloud round-trips. The practical target is enterprise scenarios where latency or data sensitivity makes cloud inference unsuitable.
For security-conscious European organisations, local inference means sensitive documents or customer data can be processed by an AI model without leaving the device – a configuration that simplifies GDPR Article 25 data minimisation requirements and reduces exposure in regulated sectors.
Microsoft Defender now covers AWS RDS
A separate announcement that reached general availability on June 1 is worth noting alongside Build: Microsoft Defender for Open-Source Relational Databases now supports Amazon RDS instances. Organisations running databases on both Azure and AWS can manage their database security posture from a single Defender dashboard, simplifying the unified threat detection and audit trail requirements under NIS2 and DORA.
What European development teams should take from Build 2026
Three things stand out as immediately actionable:
Evaluate the Copilot model transition. The August switch to Project Polaris will be automatic for most plans. Teams with high Copilot adoption should identify power users, run a brief parallel evaluation period before the transition, and document any productivity differences. If Polaris underperforms for your specific codebase type, GitHub Enterprise plans retain the option to override the default model.
Reassess your AI Foundry model choices. If your organisation has been avoiding Azure AI Foundry due to limited model selection or GDPR concerns about US-origin models, the addition of Mistral with EU data residency changes that calculus. French and German public sector organisations in particular have been waiting for a Mistral-on-Azure option.
Review your AI agent governance before adding Foundry pipelines. The visual RAG designer and multimodal orchestration capabilities in Foundry make it easier than ever to build AI systems that process sensitive data. That ease of use needs to be matched with governance: data classification, access controls, audit logging, and a review of whether the resulting system qualifies as a high-risk AI system under the EU AI Act’s August 2, 2026 compliance deadline.
If you want help evaluating whether Azure AI Foundry, GitHub Copilot, or a combination of open and commercial models is the right foundation for your engineering team’s AI tooling – or if you need guidance on aligning your AI deployments with GDPR and the EU AI Act – contact Excello Digital. We help European engineering teams make pragmatic architecture decisions that hold up under both technical and regulatory scrutiny.