Machine-drafted, human-approved AI industry news. Every story reviewed before it hits the wire.
Microsoft has added Managed Identity support to the Logic Apps Standard VS Code extension, allowing developers to use the same authentication method locally and after deployment. The feature requires specific extension and bundle versions and a local setting. Managed connectors do not yet generate dynamic values in the designer, while service provider connectors do.
Microsoft has made the GitHub Copilot harness in Copilot Studio generally available, enabling long-horizon automation with frontier reasoning models. The new harness uses usage-based billing, and new authoring tools are being introduced. Cost predictability for complex agents remains an open question.
Azure Functions Agents now emits OpenTelemetry traces from the runtime itself, so every agent run shows up in Application Insights without adding instrumentation code. Two settings enable the export, and a Model Router example shows how the trace makes otherwise invisible model-selection decisions visible. The behavior applies to any agent on any model.
Microsoft has released a public preview of Markdown for Agents for Azure App Service. It converts HTML to Markdown when a client sends the `Accept: text/markdown` header, reducing median payload size by 97 percent in internal tests. The feature is Windows-only for now and requires manual REST API or CLI configuration.
Universal Music Group, Sony Music, Warner Music Group, and independent labels have proposed barring AI-generated recordings from official charts unless they are "substantially human made," a term left undefined. The proposal also requires lawful AI use and disclosure, but lacks operational definitions, verification mechanisms, and any commitment from chart operators to adopt it. IFPI's endorsement gives it international weight, yet it remains a recommendation with unresolved enforcement questions.
Microsoft has made Foundry Local generally available, a cross-platform runtime for running language models on end-user devices. It also extended its AI Red Teaming Agent with a local scan workflow, now in public preview, to evaluate on-device models for safety and security. The pairing addresses latency, connectivity, and data locality concerns of cloud inference while enabling systematic red teaming.
GitHub's Copilot Business and Enterprise plans use a pooled credit model with user-level hard caps and group overage budgets. Overage charges accrue unless administrators explicitly enable a stop-usage toggle. Features like chat and agents draw from credits, while code completions remain unaffected.
Microsoft has added cost management and pricing tools to the Azure Resource Manager MCP server, enabling AI agents to query cost data, retrieve pricing, and manage budgets. The update includes two toolsets for Cost Management and Pricing, accessible through Visual Studio Code and GitHub Copilot with opt-in configuration. Limitations include dependence on Azure's backend data accuracy and update latencies.
Anthropic released version 5 of the Model Context Protocol, replacing its stateful session model with a stateless request/response architecture. The spec adds OAuth 2.0 authorization and a versioned extensions framework for features like interactive UIs and long-running tasks. Adoption data shows over 400 million monthly SDK downloads and 950 MCP servers listed in Claude’s directory.
Moonshot AI's Kimi K3, a 2.8 trillion-parameter open-weight model with a 1 million-token context window, is available for deployment on Microsoft Foundry through Fireworks AI. The model's large context window eliminates the need for chunking when handling long documents or codebases, though attention reliability across the full window remains an active area of research. Pricing is $3.30 per million input tokens and $16.50 per million output tokens, reflecting the model's size and inference cost.
