This week in Data & AI: 12 October 2026 – 18 October 2026
Agents moved closer to production this week, with Routines, Tool Search and agent-to-agent calls reaching GA in Microsoft Foundry and Copilot Studio agents gaining their own accounts and memory. On the data side, Databricks shipped multi-statement transactions and a GA ADBC driver for Power BI, while Fabric admins get a redesigned Capacity Metrics app in preview.
Microsoft Fabric
Capacity Metrics app gets a redesign, with throttling and per-item history
A new version of the Fabric Capacity Metrics app adds a health page that shows how much time each capacity spends throttled, an hourly utilisation heat map, a table of the items consuming the most CU and a 14-day history for individual items. It sits alongside the existing GA app, so admins can trial it without losing their current reports. For anyone running shared capacities, it should make it much quicker to find the workload behind a throttling incident.
Source: A new look and new capabilities in the Capacity Metrics App (Preview)
Power BI
Power BI October update: a native Gantt chart and changes to surge protection
The October release adds a built-in Gantt chart and "show selected first" options for list and button slicers (both in preview), and the Report Authoring skill for Fabric agents can now add and format custom visuals. Semantic models now report CU usage continuously during long-running background operations, so surge protection can step in earlier rather than relying on late throttling. Teams with long refreshes should keep an eye on rejection and delay patterns after the change.
Source: Power BI October 2026 Feature Summary
Microsoft Foundry
Foundry agent building blocks reach GA: Routines, Tool Search and agent-to-agent
Routines (starting agents from schedules, timers or GitHub and Teams events), Tool Search (loading tool definitions only when needed) and agent-to-agent delegation are now generally available in Microsoft Foundry, and Toolboxes for prompt agents enter public preview alongside the existing GA hosted-agent Toolboxes. Voice agents, resilient long-running hosted agents and trace-based Insights arrive in public preview. Together these cover much of what is needed to run scheduled, multi-agent data workflows in production.
Source: What's new in Microsoft Foundry | September 2026GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna generally available in Foundry
Foundry now offers a three-tier GPT-6 line-up: Astra as the most capable model, Sol as a cheaper option close to Astra in quality, and Luna for high-volume work. Having clear cost tiers in one family makes it easier to match models to tasks such as extraction, classification and agent reasoning. Model Router session affinity, which keeps related requests on the same model, is in preview.
Source: What's new in Microsoft Foundry | September 2026
Copilot & Agents
Copilot Studio agents get their own accounts, process memory and triggers
Copilot Studio agents can now work in Microsoft 365 under their own credentials and carry preferences and process context from one cycle of a business process to the next. Triggers are now available for the GitHub Copilot harness, with these features rolling out through the Frontier programme. Admin controls for agent accounts are planned for the coming months, so governance teams should plan how agent identities will be approved and audited.
Source: Delegate shared business processes to agents that are more persistent and proactive with Microsoft Copilot Studio
Databricks
ADBC driver becomes the default Databricks connector for Power BI
The Arrow Database Connectivity (ADBC) driver for Power BI is now generally available and is used by default for new connections in Power BI Desktop and the Power BI service. Teams reporting on Databricks from Power BI should test the driver on their heavier models and refresh patterns before relying on it in production.
Source: Azure Databricks platform release notes: October 2026Agentic SQL code converter reaches GA, including T-SQL
The agentic code converter in Azure Databricks is now generally available for SQL. It translates code from dialects such as T-SQL, Snowflake and BigQuery into Databricks SQL, then validates and fixes the output before writing it to a folder. It is worth a look for SQL Server or Synapse migrations, although converted code still needs review and testing.
Source: Azure Databricks platform release notes: October 2026Multi-statement transactions now GA for Delta Lake and Iceberg tables
Azure Databricks can now group several SQL statements into one atomic transaction on managed Delta Lake and Apache Iceberg tables, using either BEGIN ATOMIC … END or BEGIN TRANSACTION … COMMIT. Catalog commits must be enabled. This removes a long-standing gap for teams migrating warehouse logic that relies on all-or-nothing updates across tables.
Source: Azure Databricks platform release notes: October 2026
Claude
Fast mode for Claude Opus 5.5 leaves research preview and reaches Microsoft Foundry
Fast mode for Claude Opus 5.5 no longer needs an access request or beta header, and comes with default rate limits in the Start, Build and Scale tiers. It is also now available through Claude in Microsoft Foundry for Azure-hosted deployments. Fast mode for older Opus models remains in research preview.
Source: Claude Platform release notesOpenTelemetry export for Claude API and Managed Agents, in beta
Anthropic has launched OpenTelemetry export in beta, so Messages API requests and Claude Managed Agents sessions can be traced in an existing observability platform. The Claude SDK emits spans in seven languages, and API-side spans are enabled by adding a destination in the Claude Console. This lets teams monitor agent latency, cost and failures alongside the rest of their data platform.
Source: Claude Platform release notes