Weekly ML Roundup: Fabric agentic analytics, IQ, and graphs

by TechHub

This week in ML, Microsoft Fabric moved closer to an agent-ready analytics platform, with new ways to ship backends into Fabric, ground agents in governed context, and model relationships directly on OneLake. Rayfin positions Fabric as a default deployment target for data-powered apps, while Fabric IQ (now GA) and its ontology support aim to standardize how agents request context with permissions and auditability built in. Graph in Fabric (GA) adds GQL-based relationship querying, and the Fabric Operations agent plus Fabric Skills show how Microsoft wants teams to monitor, automate, and code against Fabric with guardrails instead of one-off scripts.

This Week's Overview

Microsoft Fabric becomes an “agentic analytics” platform

Rayfin: open-source backends that deploy into Fabric

Microsoft introduced Rayfin, an open-source SDK and CLI that treats your backend as code and deploys it directly into Microsoft Fabric. The pitch is straightforward: apps inherit Fabric governance and identity, and the data they generate can land in OneLake from day one for analytics and AI workloads, building directly on last week's Fabric thread around tightening governance and operational clarity across OneLake-backed data products.

For developers, this shifts “app + data platform” from an integration project to a default deployment target. Demos highlighted Fabric SSO authentication, type-safe schemas and migrations, and an agent-generated full-stack starter that can be iterated on with GitHub Copilot while keeping the backend Fabric-native.

Fabric IQ and Ontologies: a governed context layer for agents (now GA)

Fabric IQ reached general availability as a shared “context layer” that ties together semantic models, ontologies, Real-Time Intelligence, and (now) graph capabilities, extending last week's emphasis on Fabric governance controls (auditing, export restrictions, and admin visibility) into an explicit grounding layer for agent access. This week’s messaging focused on making grounding explicit: the ontology becomes the durable contract between business concepts and the data products agents can read, write, and reason over.

Practically, the direction is toward standard interfaces for agents to request context with guardrails. Fabric IQ updates include Ontology support for MCP (Model Context Protocol) and integration points with Foundry IQ and Copilot Studio through Agent 365, so teams can connect agents to governed data without rebuilding retrieval and permissions per app.

Graph in Fabric (GA): OneLake-backed graphs with GQL

Graph in Fabric is now generally available, adding relationship-first modeling on top of OneLake with GQL support (ISO/IEC 39075), a natural follow-on to last week's Fabric focus on governed, OneLake-first operations now that more workloads can model and query relationships without leaving the platform. That matters for agentic apps because multi-hop reasoning and impact analysis often map better to graph traversals than repeated joins across normalized tables.

Microsoft positioned Graph as inheriting Fabric governance by default, and tied it directly to Fabric IQ Ontology. Example scenarios include security and operations analysis (such as Microsoft Sentinel Graph) where agents need to trace relationships across entities and explain the path they took.

Fabric Operations agent (GA): LLM-assisted monitoring with governed actioning

The Fabric Operations agent is now generally available, using LLM-generated monitoring rules plus Real-Time Intelligence signals to detect issues, investigate, and (with authorization) trigger remediation actions across Fabric and connected systems, building on last week's thread about making monitoring and triage more repeatable and auditable. New pieces called out this week include Teams notifications, natural-language Q&A, and tracing/auditability hooks designed for enterprise governance.

From an engineering standpoint, the key question is change control: the agent can propose or execute actions, but identity and policy still need to be anchored in Microsoft Entra ID and governed through Agent 365. If you run Fabric in production, the practical takeaway is to treat this like any automation system: start in “recommend” mode, validate rule generation against known incidents, then ratchet up permissions.

Fabric Skills: teaching coding agents the right way to call Fabric

Microsoft published Fabric Skills, an MIT-licensed library of “skill bundles” that help GitHub Copilot and other coding tools (Claude Code, Cursor, Windsurf, and CLI-based workflows) use Fabric APIs and CLIs correctly, reinforcing last week's governance-and-ops storyline by turning “approved ways to automate Fabric” into shareable, versioned artifacts. The details here are important: it covers workload-specific authentication patterns and token audiences, which are common failure points when an AI tool tries to script “just call the API” steps.

For teams experimenting with agentic coding internally, this is a useful pattern: codify approved API usage as repo-loaded configuration so assistants follow your guardrails by default. It also creates a contribution path where org-specific best practices can be upstreamed instead of living only in internal wikis.