OpenAI has released the Agents API in public beta, giving developers a way to build long-running AI agents with the same harness and infrastructure used by Codex. The API can manage context, tools, files, code execution, and parallel subagents while allowing each workload to run in an OpenAI sandbox, a developer’s own infrastructure, or a partner environment.

Source

OpenAI has released GPT-Live-1 through its API, allowing developers to build applications that listen and speak simultaneously instead of waiting for separate speech-recognition, reasoning, and speech-synthesis stages. The model delivers faster turn-taking, handles interruptions in real time, and costs $0.05 per minute for the voice layer.

Source

Microsoft has announced Viva Glint export APIs that let you pull employee survey response data out of Viva Glint through Microsoft Graph instead of downloading reports manually. Viva Glint is Microsoft's employee listening platform where organizations run engagement surveys and collect feedback. The APIs use OAuth 2.0 application permissions and require both Microsoft Entra app registration and configuration inside Glint. They are available only in a limited private preview as of September 2026. This article covers what the APIs do, what you need to set them up, and what limitations apply.

Source

.NET 11 Release Candidate 1 is now available with a go-live support license, making it suitable for production workloads before the final release. The first RC also works with Visual Studio 2026 Insiders and Visual Studio Code with the C# Dev Kit, giving teams a supported path to test upgrades and deployment pipelines.

Source

GitHub is previewing Project HydraFusion, also presented as Auto Model Selection in Visual Studio, to decide which AI model—or combination of models—should handle each coding request. Early testing reports up to 67% lower costs and improved quality, while giving platform teams a new model-governance problem: tracking which model wrote, reviewed, or changed code.

Source

Spotify has reduced Claude Code’s token consumption by about 90% by diverting routine file handling and boilerplate coding to cheaper models. Its Portal platform uses AiKA Modes and the Shunt plugin to keep Claude focused on editing, debugging, and complex reasoning.

Source

Microsoft announced Project Zenith on September 4, 2026, as a preconfigured Windows 11 experience for developer-class PCs. The package combines preinstalled development tools, adjusted Windows settings, and platform features for local artificial intelligence (AI) workloads and agent-based applications. Project Zenith ships first on AMD Ryzen AI Halo systems and requires at least 64 GB of unified memory (system RAM shared between the processor and graphics hardware) and at least 250 GB/s of memory bandwidth.

Source

OpenAI is testing a gated Codex for Windows dashboard that ranks the chats consuming the most five-hour, weekly, and credit allowances. The feature can also group subagent activity with the conversation that started it, giving users a clearer explanation for unexpectedly high usage.

Source

GitHub’s HydraFusion research preview lets Copilot assemble a workflow from multiple AI models instead of relying on one model for every coding request. GitHub reports up to 67% lower estimated costs and a 4.9-point quality improvement on Terminal-Bench 2.1, although users are billed for every model involved.

Source

Anthropic has already bundled Android Emulator support into Claude Desktop, including app control, screenshots, recordings, and setup automation—but the feature remains disabled behind a rollout switch. Reverse engineering of version 1.44121.4 shows the integration is substantially complete, even though Anthropic has not announced when users will receive it.

Source

Google’s Gemini 3.8 Flash is arriving in GitHub Copilot, giving paid users another model for coding in editors, terminals, IDEs, and cloud-based workflows. The rollout is gradual, and Business and Enterprise administrators can control whether users receive access.

Source

Cloudflare now supports running Cursor Cloud Agents inside Cloudflare Sandboxes, moving the agents’ terminal, filesystem, and browser operations into isolated environments controlled by the customer. Cursor still handles inference, planning, and orchestration, but enterprises gain tighter control over where source code, build caches, and secrets are used during autonomous coding tasks.

Source

GitHub Copilot can now submit approvals on pull requests that count toward branch-protection requirements, moving its code review role beyond comments and recommendations. The capability is in public preview, but it remains disabled until administrators explicitly authorize it.

Source

A repository’s own `.git/config` can still trigger attacker-controlled code through Claude Code, Codex, Cursor, goose, and other AI coding agents before trust prompts, authentication, model calls, or tool approvals occur. Manifold Security’s GitSpawn research found eight flaws across seven agents; fixes are available for several tools, but Hermes Agent, Qwen Code, Grok Build, and a second Claude Code path remained vulnerable during September testing.

Source

Microsoft is making the code-based Power BI Project (PBIP) format generally available while switching Power BI Report (PBIR) to the default report format in Power BI Desktop and the Power BI service. The change gives Power BI teams a more practical path to source control, CI/CD, and programmatic report development without removing support for existing PBIX files.

Source

Visual Studio Code Insiders 1.137.0 is now available for Linux through Microsoft’s VS Code PPA, giving administrators a new package to test after 1.136 Insiders previewed the next stable release. Microsoft published the build on September 2, 2026, as version `1.137.0-1788326293`.

Source