Automotive

Vehicles increasingly differentiate through software. Microsoft's reference architecture for software-defined vehicles names the strains: customer expectations met mostly through software, integration complexity the industry has struggled with for more than a decade, and tools and processes that were not built for this shift, which push up integration costs and delay start of production.

Embedded code for driver assistance and automated driving must also follow MISRA C and C++ coding guidelines, which run to hundreds of pages and which few engineers know in full. Woven by Toyota reports that a single driver-assistance recognition module needed about 60,000 errors corrected, work its engineers had been doing by hand after static analysis.

At the same time, carmakers run large global office workforces and datacenter estates, so cost and security pressure sits beside the engineering challenge.¹²

Microsoft groups its automotive offer under Microsoft for Mobility, with scenarios for digital engineering, software-defined vehicles, connected fleets, and driver assistance and autonomous vehicle development. For software-defined vehicles it offers a reference architecture for a cloud-native development toolchain and contributes open source in-vehicle components through the Eclipse Foundation SDV Working Group. The software-defined vehicle scenario lists GitHub, Azure Deployment Environments, Microsoft Dev Box, Azure compute and Azure Arc among its building blocks.

Two published cases show different entry points. Woven by Toyota built a multi-agent proof of concept on Azure OpenAI Service, Azure App Service, Azure Cosmos DB and GitHub Enterprise, orchestrated with AutoGen: a Coder agent proposes fixes, a Reviewer agent critiques them and an Evaluator agent explains its reasoning with a certainty score, while engineers accept or reject every change. On in-house code, Woven reports the proof of concept corrected 81.5 percent of MISRA compliance errors automatically. Stellantis took a workforce route: all employees have Copilot Chat, an initial 20,000 Microsoft 365 Copilot licenses go to selected roles, and infrastructure is moving to Azure with a targeted 60 percent datacenter footprint reduction by 2029, a target rather than a result.

Not every program needs this. Static analysis plus experienced reviewers remains the baseline for MISRA work on small codebases, and a supplier with one product line may get more from its existing ALM and CI tools than from a new cloud toolchain. AutoGen and Semantic Kernel now have a direct successor in Microsoft Agent Framework, so new agent projects should start there.¹²³⁴⁵