Copilot workflow updates and voice AI evaluation #79
Today's Letter
GitHub Copilot for Visual Studio, June update
- GitHub published the June update for Copilot in Visual Studio on July 14, 2026, focused on usage visibility, MCP server trust, and C++ modernization workflows.
- The Copilot Usage window now shows real-time usage under the usage-based billing model, with alerts for approaching limits, hitting limits, and overage activation.
- Visual Studio now validates MCP server configuration and asset fingerprints against a trusted baseline at startup, and prompts for review if changes are detected.
- GitHub Copilot modernization agent for C++ is now generally available for MSVC upgrade scenarios, with Automated and Guided modes exposed through Solution Explorer and Copilot Chat.
- Next edit suggestions can now propose follow-up edits anywhere in the active file through extended range suggestions, instead of staying near the cursor.
- Pull requests can be added to Copilot Chat from the Git Repository window, which brings in the PR description, changed files, and comments as context.
- Visual Studio also adds in-IDE pull request review for GitHub and Azure DevOps, including browsing, commenting, approving, and completing PRs without leaving the IDE.
- GitHub said the update is available across all Copilot plans, including Free, Student, Pro, Pro+, Max, Business, and Enterprise.
Source: github.blog
Hugging Face, Real World VoiceEQ benchmark introduced
- Hugging Face published Real World VoiceEQ on July 15, 2026, a benchmark aimed at measuring the human quality of voice AI interaction.
- The benchmark evaluates more than 40 proprietary and open-source voice models across 15+ dimensions and more than 60 metrics covering ASR, TTS, speech-to-speech, and speech understanding.
- It focuses on acoustic signals that transcripts miss, including tone, emotion, speaker identity, hesitation, pacing, emphasis, and background context.
- The dataset was built from more than 1 million human ratings across demographics, speaking styles, and acoustic environments, including 785,000 TTS ratings and 48,000 STS ratings.
- The post says current voice progress is increasingly specialized, with no single system ranking in the top five across all eight TTS capability groups.
- It also argues that voice models are improving faster at speaking than listening, with some systems still relying mainly on transcript content instead of paralinguistic audio cues.
- All evaluations were run on HumeAI's Kairos platform, which Hume says can also support custom evaluation, failure analysis, preference-data collection, and RLHF workflows for production voice systems.
Source: huggingface.co
Jocoletter curates AI, software, and product trends for developers and builders.
#GitHub #HuggingFace