Microsoft Developer (@msdev) header
Microsoft Developer (@msdev) profile picture

Catch up on all the news from @GitHub Universe

Joined February 2013
We just released a free course: GitHub Copilot app for Beginners. It includes 7 chapters and a hands-on project that walks through core agent-development concepts.
10
74
4
482
25,898
The @github engineering team built a best-in-class, truly multi-model harness, which we've implemented across Microsoft products. Introducing Microsoft Copilot harness in Foundry Agent Service, a simpler way to build agents without rebuilding the execution layer. It handles orchestration, tools, context, and controls so you can focus on your agent’s skills and outcomes. Watch @jeffhollan's demo to see it in action.
18
44
14
430
1,330,839
You built it. Now put it in play. It’s not too late to submit your Inside the Game: Developer Hackathon project by Oct. 27 for the opportunity to have your work recognized and showcased. Finish your submission: msft.it/6014axj40
1
3
1
15
4,248
#GitHubUniverse kicks off next week, October 28–29. Grab your pass to join us in person or sign up to tune in virtually. ➡️msft.it/6013aappD
7
18
5,080
You have an idea. Now take it beyond the scoreline. ​ Build something that changes how football is understood, told, or experienced with AI.
4
4
27
4,892
Knowing when to use AI and when to question its output is part of learning to build. Watch @shanselman, software engineer Julia Gao, and CMU’s Dr. Michael Hilton discuss PRAISE and how to support the next generation of developers. youtube.com/watch?v=BglzGRnH…
1
3
5,701
Classification is a key use case for decision models. Check out how accurately and quickly Microsoft-Decision-1 can categorize a variety of queries compared to GPT-6 Sol.
1
3
20
5,000
Microsoft-Decision-1 achieved the highest accuracy in our 36-benchmark comparison, spanning nearly 150,000 questions across benchmarks kept blind from training.
2
2
30
5,587
The challenge starts now. Build an AI-powered solution that turns synthetic football data into match intelligence, narratives, recaps, or personalized fan experiences. Register by Oct. 20. Submit by Oct. 27: msft.it/6018aUKYq
3
3
34
6,337
Today, we announced the next step in the Project HydraFusion vision: local models and intelligent routing for GitHub Copilot on Windows. @cinnamon_msft shows what this can mean for your everyday coding workflows.
20
42
5
307
21,320
Grounding ties extracted values to source evidence. Confidence scores flag uncertainty. A developer-defined JSON schema gives your app structured output. Together, they help you validate results or route them for review before your app or agent acts.
1
14
3,683
In Azure Content Understanding, OCR and layout models preserve text, tables and reading order in Markdown before an LLM extracts fields. The perception layer reads the page. The LLM reasons over that representation.
1
1
15
1,949
Our tests showed LLMs outperforming specialized models on contracts and agreements, where meaning lives in the words. On tax forms and mortgage applications, where layout carries meaning, our existing approach still won. Match the approach to the document.
1
16
2,635
An agent reported 46.1 dB for F-03’s ORL at 1310 nm. That cell was empty. It borrowed the value from the 1550 nm column after extraction flattened the table. For your document pipeline, preserving structure is part of preserving meaning.
1
16
4,366