Our team spent 10 years teaching machines to read documents. Here’s what we learned when LLMs entered the picture. LLMs improved language understanding, but reliable document extraction still depends on the system around the model. Those lessons shaped Azure Content Understanding: preserve document structure, trace extracted values to source evidence, and flag uncertainty so you can validate results before your app or agent acts on them. Read the team’s lessons from a decade of document intelligence, and what they mean for your document pipeline. 📰 https://lnkd.in/gtM4MM_z
Microsoft Developer
Software Development
Redmond, WA 709,948 followers
Welcome to a community built for passionate developers.
About us
Dive into the biggest news from Microsoft Build and learn how to start building with it today 👇
- Website
-
https://blogs.microsoft.com/blog/2026/06/02/microsoft-build-2026-be-yourself-at-work/
External link for Microsoft Developer
- Industry
- Software Development
- Company size
- 10,001+ employees
- Headquarters
- Redmond, WA
Updates
-
WSL containers is now generally available. To try it out, simply run wsl --update in your terminal or download the latest release from GitHub. Full Announcement: https://lnkd.in/gGCQVaBC Latest Release: https://lnkd.in/gQn57vmq
-
Microsoft Developer reposted this
Great agents need multiple ingredients: a strong model, a flexible harness, and critically, the right knowledge about your business processes, priorities, technical details, etc. Knowledge in the real world is messy: PDFs with long tables or embedded scans, audio conversations, images with charts or schematics, etc. Content extraction to get agents the knowledge they need is a notoriously hard problem, and one we've been working on for a long time. I'm excited that the team has started a blog series that dives deep into this problem space. First up: why LLMs alone are not enough, and the roles of Azure Document Intelligence and Azure Content Understanding: https://lnkd.in/ghQgemww More to come over the next few weeks.
-
Go beyond the scoreline with AI. Registration is open for Inside the Game: Developer Hackathon, bringing Microsoft, the Premier League, and developers worldwide together to explore what's possible at the intersection of football, data, and AI. Build an AI-powered solution. Sharpen your skills. Bring a new perspective to the game. Register by October 20: https://lnkd.in/gaMZDyeP
-
If you’re running agents in production, “deny” needs to mean the action actually stops. Today, controls can behave differently from one agent framework to another. That makes it harder to reuse policy checks and approvals, and harder to prove they were enforced. We built Agent Hooks to give control builders and framework builders a shared contract. It defines where controls run, what context they receive, and how a framework must handle their verdicts. A conformance kit tests whether the framework follows through. The spec, five SDKs, and runnable demos that need no API keys are open source: https://msft.it/6046a9m3a
-
-
Today, we’re introducing run-assert-eval. With a single prompt in Visual Studio Code, the skill identifies risks specific to your agent, measures how often they occur, generates runtime policy based on those findings, and reruns the same eval to see whether the policy worked. See how we tested it: https://lnkd.in/gE_kreVv Get started with the ASSERT repo: https://lnkd.in/gyitaiEC
-
-
Microsoft Developer reposted this
What’s next in AI and cloud is taking shape now. The Microsoft Ignite Session Catalog is live. Explore the catalog, build your agenda and get a first look at what’s ahead. Go deeper with the experts shaping the technology, and explore practical guidance for turning innovation into business impact. See what’s waiting at #MSIgnite: https://msft.it/6048agYcn
-
Microsoft Developer reposted this
Want to build agentic applications without implementing the agent loop yourself? 💡 Join our beginner-friendly livestream series to explore the GitHub Copilot SDK and learn how to embed Copilot’s agent runtime into your own application or service. You’ll learn how to: • Create multi-turn conversations • Stream responses in real time • Expose application functions as tools • Connect external capabilities through MCP servers • Control agent behavior with hooks and permission handlers • Work with clients, sessions, prompts, and session state We’ll also clarify when to use the Copilot SDK versus the interactive Copilot CLI: use the CLI for developer tasks in the terminal, and the SDK when you want to bring Copilot’s agentic capabilities into your own software. No prior GitHub Copilot SDK experience is required. Plus, special sessions are available for .NET and Java developers. Register for English, Chinese, Portuguese, Spanish, Japanese, or Korean 👇 https://gh.io/letslearn
-