Microsoft Fabric IQ and Power BI: Why Semantic Models Are Becoming the Brain of AI Agents

When the Reports Start Making Decisions

This article is part of the EPC Group Microsoft Build 2026 series. For the full strategic read on Project Solara, the Copilot Super App tease, MAI, Scout, MDASH, and RTX Spark — see the pillar: Project Solara, the Death of Apps, and the One Copilot That Wasn't.

There's a familiar moment in every Power BI engagement I've run over the years: the point where the client stops looking at the dashboard as a reporting tool and starts trusting it enough to base actual decisions on it. It takes longer than people expect. First you fix the data pipeline. Then you reconcile the definitions — what exactly does "revenue" mean in this model, for this business unit, on this date? Then you certify the semantic model, document the measures, and establish the governance process that keeps it accurate as the business changes.

That moment — when a business leader looks at a number and trusts it — is the product of enormous invisible work. Data engineers, BI developers, governance teams, and IT operations all contribute to it. Most of them never get named in the board presentation.

I've been thinking about that invisible work a lot since Build 2026, because Microsoft just announced something that changes what Power BI semantic models are for. They're not just reporting infrastructure anymore. With Fabric IQ now generally available, semantic models have become the reasoning substrate for AI agents — the knowledge layer that determines whether an autonomous agent understands your business or just pattern-matches on your data.

The invisible work you've been doing for years? It just became the most strategically important thing in your AI architecture.

---

Fabric IQ: What "Generally Available" Actually Means

Microsoft announced at Build 2026 that Fabric IQ is generally available. It is the intelligence layer within the Microsoft IQ umbrella that, as Microsoft describes it, "models how the business operates." It is built on two foundations: Power BI semantic models — the structured, governed representations of trusted business metrics — and Ontologies, which define business entities and their relationships so that agents can reason in the language of the business rather than the language of a database schema.

Power BI semantic models are the "brain" in Microsoft's framing — and the metaphor is apt. A brain that has been trained on accurate, well-governed, precisely defined knowledge reasons well. A brain trained on contradictory, stale, or ambiguously labeled data reasons plausibly but incorrectly. The difference between those two outcomes, when you're talking about an autonomous agent making business decisions, is not a UX problem. It's an operational risk.

Ontologies take this further. A semantic model tells an agent what "Q3 Revenue" is. An ontology tells the agent what "Revenue" means in relation to "Customer," "Product," "Region," and "Contract Type" — how those entities connect, how they should be traversed when answering a business question, and what the authoritative source of truth is for each relationship. When Ontologies become GA "in coming months," enterprise agents will be able to navigate your business knowledge graph rather than just querying individual tables.

---

The Agent 365 Integration: Fabric as a First-Party MCP Tool

The integration between Fabric IQ and Agent 365 is one of the most architecturally significant connections in the Build 2026 platform stack. Agent 365 reached GA on May 1, 2026 — it is the enterprise control plane for observing, governing, and securing agents, delivering Defender, Entra, Intune, and Purview protections. What Build 2026 added is the Fabric IQ integration: Fabric IQ is now integrated with Agent 365 as a first-party MCP tool in preview, meaning agents managed through Agent 365 can natively query your semantic models as a governed, permissioned tool invocation.

Think about what that means operationally. An agent in your finance workflow, running under its own Entra identity, governed by Agent 365's security controls, can call Fabric IQ as a tool to retrieve a certified measure from your Power BI semantic model — with the same governance chain that applies to every other agent action. The data the agent reasons on is the same data your CFO is looking at in the board report. Not a copy. Not an export. The live, governed, certified semantic model.

This is what "grounded AI" looks like when the architecture is done correctly.

Fabric IQ is also being extended into Microsoft 365 Copilot (through Cowork and Copilot Chat) for Frontier customers with Copilot licenses. It's accessible via GitHub Copilot CLI through Agent Skills for Fabric — the recently released open-source capability. And Ontologies are accessible from Microsoft Foundry as knowledge sources in preview, which means agents built on Foundry's multi-model platform can use your business ontology as a knowledge graph rather than reasoning from scratch on every query.

---

Operations Agents: GA and Running in Production

Operations agents in Fabric are now generally available. These are the agents that reason over shared live context, make policy-based decisions, and take action in the moment — on live data, not batch exports. The phrase "in the moment" is doing significant work in that description: these aren't scheduled batch jobs that run a report overnight and trigger a workflow. They monitor live signals, evaluate conditions, and act on current state.

Fabric Real-Time Intelligence provides those live signals. Your streaming data, your event hubs, your change feeds — operations agents can reason over this stream and act on it under the governance of Fabric's security and policy layer. For operational workflows where the value of a decision degrades rapidly with time — inventory management, fraud detection, supply chain exception handling — this is the architecture that makes agents genuinely useful rather than analytically interesting.

Graph in Fabric is GA. Graph traversal native to the Fabric platform means agents can navigate entity relationships directly in the data layer rather than requiring external graph processing. Planning in Fabric reaches GA later this month.

---

OneLake: The Single Source of Agent Truth

OneLake — Microsoft's unified AI-ready data lake that consolidates the multi-cloud estate into a single logical store — is the foundation underneath all of this. Operations agents, semantic models, Fabric Data Warehouse, Real-Time Intelligence streams: they all surface through OneLake as a unified data reality.

The recently GA OneLake catalog in Microsoft Foundry means agents built on the Foundry platform can discover, catalog, and consume data from OneLake directly. Shortcuts to SharePoint and OneDrive are GA. Shortcuts from Fabric Data Warehouses are in preview. The Database Hub in Fabric — in private preview — allows mirroring databases into OneLake, extending the unified data surface to external sources.

For enterprise architects, OneLake's strategic value is that it removes the "which copy of the data is right?" question from the agent development workflow. There is one lake. The agents query it. The governance, lineage, and certification that applies to the data in that lake applies uniformly to every agent that accesses it. When agents reason from a single source of truth, their outputs are coherent. When they reason from six copies of the same data with different transformation histories, they aren't.

---

Fabric Data Warehouse Gets GPU Acceleration

A detail from Build 2026 that deserves more attention than it received: the Fabric Data Warehouse is the first fully managed data warehouse to offer GPU acceleration. Query acceleration in early access preview will arrive in the next few weeks — no query rewrites required. It uses NVIDIA accelerated computing and custom CUDA kernels to accelerate query execution on the managed warehouse infrastructure.

This matters for agentic workloads specifically. Agents that reason over large datasets — fraud analysis, supply chain optimization, financial consolidation — generate query patterns that are fundamentally different from the human-driven BI queries warehouses were designed for. Human analysts run a query, read the result, formulate a follow-up, and run another. Agents can run hundreds of queries in a minute as they build up a reasoning chain. GPU-accelerated query execution at the warehouse layer is the architectural preparation for that query pattern at scale.

You don't need to redesign your data warehouse schema or rewrite your queries to get this. The acceleration is managed, applied transparently, and governed by the same Fabric infrastructure you already operate. That's a meaningful operational advantage over environments that require schema-level optimization to achieve equivalent query performance.

---

The Rayfin and Replit Connection

Two adjacent Fabric announcements from Build 2026 are worth noting for their architectural implications. Rayfin is a new open-source SDK and CLI that lets developers describe what they want to build and receive an enterprise-grade application backend — database, authentication, and more — in application code, deployed directly to Fabric, with data landing in OneLake. It's a significant reduction in the time from "I need to build an agent application" to "I have a governed data layer for it."

The Replit partnership takes a similar approach from the application development direction: build enterprise applications in Replit, and the data and services stay in the customer's Fabric tenant. This is an important boundary condition for enterprise adoption of AI development tools — the developer experience can be cloud-native and fluid, while the data governance remains inside the organizational boundary.

---

The New [Azure](/services/azure-cloud-services) Database Ecosystem

Supporting the Fabric architecture is a set of new and updated database capabilities from Build 2026 that give agents persistent, intelligent storage options beyond the warehouse and the lake.

Azure HorizonDB enters public preview: fully managed, PostgreSQL-compatible, storage up to 128 TB, compute up to 3,072 vCores, sub-millisecond multi-zone commit latency, vector search, integrated AI model management, and direct connectivity to Foundry and Fabric. For agents that need relational semantics with vector capabilities at enterprise scale, HorizonDB is the purpose-built answer.

Azure Cosmos DB adds an agent memory toolkit that standardizes persistent memory using Cosmos DB, Azure Durable Functions, and Foundry models. This is the managed solution for one of the most persistently awkward problems in stateful agent architectures: where does the agent store what it learned during the last session? The Cosmos DB Linux Emulator is now GA, enabling local build and test on Linux, macOS, and Windows.

Azure Database for PostgreSQL integrates Defender for Cloud in preview, with Oracle and PostgreSQL discovery and assessment tooling for organizations evaluating migration paths.

These database capabilities aren't tangential to the Fabric IQ story. They're the complementary storage and memory tier for agents that use Fabric IQ as their reasoning substrate. The semantic model answers "what does our business say is true?" The agent memory toolkit answers "what did this agent learn from the last 10 conversations with this user?" Together they give agents a knowledge architecture that has both institutional intelligence and operational continuity.

---

Messy Data, Messy Agents: The Governance Argument

I want to make the governance argument explicitly, because I've watched it get glossed over in the Build coverage and I think it's the most important strategic point in this entire stack.

Power BI semantic models are only trustworthy AI reasoning substrates if they are trustworthy. This sounds obvious. It is, apparently, not obvious enough, because the majority of enterprise Power BI environments I've audited contain a mixture of certified and uncertified datasets, measures with identical names that calculate differently across departments, date dimensions that don't align consistently, and column-level documentation that was last updated when the BI team had more people than it does now.

An agent querying a certified, well-documented semantic model with clear business definitions will reason accurately. An agent querying a model where "Margin" means three different things depending on which report tab you're on will reason confidently and incorrectly — and it will do so at machine speed, at scale, in responses delivered to business users who trust it.

The same principle applies to OneLake data quality. Operations agents reasoning over live data that contains duplicates, inconsistent foreign keys, or stale records will make policy-based decisions on flawed inputs. Real-Time Intelligence streams that haven't been validated against the semantic layer will produce agent actions that contradict what the business model says should happen.

AI doesn't improve your data. It accelerates whatever your data already is. If your data is clean, governed, and trusted, AI agents will produce clean, governed, trustworthy outputs. If your data is a migration backlog with a dashboard painted on top of it, your agents will discover that reality faster than any audit ever did.

This is the work EPC Group does. Not theoretically — practically. We've been running Power BI and Fabric modernization engagements for clients across industries for years. Semantic model certification, measure governance, OneLake migration, data warehouse modernization. We've built the Fabric architectures that are now positioned to host these operations agents. And we've seen what happens when you skip that governance work and let AI run on uncertified models.

If your organization is looking at Fabric IQ GA and thinking "we should build agents on top of our Power BI environment," the first question to ask is: which of your semantic models would you bet your Q3 business decisions on? Start there. Certify those. Build your agent architecture on top of the ones you'd sign your name to. Then extend outward.

For a full view of how Fabric IQ connects to the broader Build 2026 agentic architecture — including Work IQ, Foundry, MAI models, Agent 365, MDASH, and Project Solara — read our full Build 2026 enterprise breakdown at epcgroup.net.

To assess your Power BI and Fabric environment's readiness to serve as the reasoning substrate for enterprise agents, contact EPC Group for a Fabric & Semantic Model Readiness Assessment.

---

FAQ

Q: What is Fabric IQ and how does it differ from Power BI?
A: Fabric IQ is the intelligence layer in the Microsoft IQ umbrella that makes the business's semantic models and ontologies available to AI agents as a governed knowledge layer. Power BI is the BI tool; the semantic models built in Power BI are the structured data representations. Fabric IQ is the infrastructure that exposes those models to agents through the Agent 365 MCP integration and other pathways. Power BI is now one component of the broader agentic data architecture.

Q: What are Fabric Ontologies and when are they GA?
A: Ontologies in Fabric define business entities and their relationships — they allow agents to understand how "Customer," "Product," "Revenue," and "Contract" relate to each other in the language of your business, not just as tables. Ontologies are accessible from Microsoft Foundry as knowledge sources in preview. GA is expected "in coming months" per the Build 2026 announcements.

Q: Are operations agents in Fabric ready for production use?
A: Yes. Operations agents in Fabric are now generally available — they can reason over shared live context, make policy-based decisions, and take action on current data. Production readiness on your side depends on the quality and governance of the data they're acting on.

Q: What is GPU acceleration in Fabric Data Warehouse and when does it arrive?
A: Fabric Data Warehouse is the first fully managed data warehouse to offer GPU acceleration, using NVIDIA accelerated computing and custom CUDA kernels. Query acceleration early access preview is coming in the next few weeks after Build 2026, with no query rewrites required.

Q: What should we do before building agents on top of our Power BI semantic models?
A: Audit which semantic models are certified and which aren't. Validate that your measure definitions are consistent across the organization. Ensure your date dimensions and key columns are clean and documented. Identify the models your business would trust enough to make decisions from — those are your starting point for agent grounding. Models that are in any way disputed or inconsistently defined should be remediated before agents use them as a reasoning substrate.

---

Sources

• Azure Blog — Microsoft Fabric & Databases: https://azure.microsoft.com/en-us/blog/microsoft-build-2026-building-agentic-apps-with-microsoft-fabric-and-microsoft-databases/
• Microsoft 365 Dev Blog — Work IQ: https://devblogs.microsoft.com/microsoft365dev/work-iq-production-ready-intelligence-for-every-agent/
• The Verge — Biggest Announcements: https://www.theverge.com/tech/941738/microsoft-build-2026-biggest-announcements
• Mashable — Agent-First Future: https://mashable.com/tech/microsoft-build-2026-agent-first-future
• PCMag — Build 2026 Recap: https://www.pcmag.com/news/microsoft-build-2026-recap-didnt-catch-the-live-stream-heres-everything
• AI Magazine — Microsoft Foundry/Trojan Horse: https://aimagazine.com/news/is-microsoft-ai-the-ultimate-enterprise-trojan-horse
• CNET — Build 2026 Live Coverage: https://www.cnet.com/news-live/microsoft-build-2026-news-ai-copilot/
• Let's Data Science — Windows Re-Centered: https://letsdatascience.com/news/microsoft-re-centers-windows-at-build-2026-18fadf11

---

Contact EPC Group:
contact@epcgroup.net · 888-381-9725 · www.epcgroup.net

---

---

---

EPC Group Can Help

Microsoft Build 2026 raised the ceiling on what agentic AI can do across the Microsoft estate — and the floor on what your tenant has to be to deploy it safely. EPC Group has been doing this work for 29 years across Fortune 500 and federal organizations, with six Microsoft Solutions Partner designations and a perfect 100 NPS on G2.

If any of the following sound like your next 90 days, that is exactly the work we do:

• Microsoft 365 Copilot Readiness Assessment — the 30-day tenant audit, Purview labeling, and adoption motion that turns an under-5% Copilot conversion into a measured rollout.
• Governed AI on Microsoft Framework — one named architecture covering Entra, Purview, M365, SharePoint, Fabric, Power BI, Copilot, Defender, and Agent 365. Seven layers, five maturity stages, one accountable owner.
• Virtual Chief AI Officer (vCAIO) — fractional senior leadership for organizations that need the operating model in 2026 but cannot hire a full-time AI executive yet.
• AI Consulting Services, AI Governance, Microsoft Copilot Consulting, Power BI Consulting, Microsoft Fabric Consulting, and Azure Consulting.

Email contact@epcgroup.net, call 888-381-9725, or request a consultation. Senior architects only — no offshore handoff, no junior account managers.

LinkedIn Post

Here's something that didn't get nearly enough attention in the Build 2026 coverage.

Fabric IQ is now generally available. And it means Power BI semantic models are no longer just BI infrastructure. They are the reasoning brain for AI agents.

Let me explain why that changes everything about how you should think about your Power BI investment — and your data governance backlog.

WHAT FABRIC IQ ACTUALLY DOES

Fabric IQ is the layer of the Microsoft IQ umbrella that models how your business operates. It's built on two things: Power BI semantic models (your structured, governed representations of trusted business metrics) and Ontologies (which define business entities and relationships so agents reason in the language of your business, not the language of a database schema).

Fabric IQ is integrated with Agent 365 as a first-party MCP tool — in preview now. That means agents running under their own Entra identities, governed by Defender and Purview, can query your certified semantic models as a tool invocation. The data the agent reasons on is the same data your CFO sees in the board report.

That's grounded AI. That's what correct looks like.

WHAT'S NOW GA

Operations agents in Fabric — GA. They reason over shared live context, make policy-based decisions, and take action in the moment. Not batch. Live.

Graph in Fabric — GA. Planning in Fabric — GA this month. Fabric Data Warehouse GPU acceleration — early access preview in weeks. First fully managed data warehouse with GPU acceleration. No query rewrites required.

OneLake catalog in Foundry — GA. One lake. One unified data surface for your agents to query against.

THE PART THAT KEEPS ME UP AT NIGHT

Here's the governance argument that nobody in the conference coverage made.

Power BI semantic models are trustworthy AI reasoning substrates only if they are trustworthy.

I've audited enough enterprise Power BI environments to know that "trustworthy" is not the default state. Certified and uncertified datasets mixed together. Measures with the same name calculating differently across departments. Date dimensions that don't align. Documentation last updated when the team was twice its current size.

An agent querying a clean, certified, well-documented semantic model will reason accurately. An agent querying a model where "Margin" means three different things depending on which report tab you're on will reason confidently and incorrectly — at machine speed, at scale, delivered to business users who trust it.

AI doesn't improve your data. It accelerates whatever your data already is. If your semantic models are clean and certified, your agents will produce clean, certified insights. If your models are a migration backlog with a dashboard painted on top — your agents will find every inconsistency and act on it.

THE EPC GROUP TAKE

We've been running Power BI and Fabric modernization engagements for years. Semantic model certification, measure governance, OneLake migration, data warehouse modernization. We've built the Fabric architectures that are now positioned to host these operations agents. And we know exactly what needs to be in order before those agents are trusted with business-critical decisions.

If Fabric IQ GA is on your roadmap — and it should be — start with a Fabric and Semantic Model Readiness Assessment. Identify which semantic models you'd bet your Q3 decisions on. Certify those. Build your agent architecture on top of the ones you can sign your name to. Then extend from there.

The invisible work your BI team has been doing for years — the data governance, the measure documentation, the certification process — just became the most strategically important work in your AI architecture.

Time to make sure it's done right.

What's the state of your Power BI semantic model certification program right now? And does your team know it's about to become AI infrastructure?

#MicrosoftBuild #FabricIQ #PowerBI #DataGovernance #AgenticAI #MicrosoftFabric #EPCGroup

---

X Post

Fabric IQ is GA. Power BI semantic models are now the reasoning brain for AI agents — operations agents run on live Fabric data, Fabric Data Warehouse gets GPU acceleration, and Ontologies are coming to connect the business knowledge graph. Messy data = messy agents. epcgroup.net/microsoft-fabric-iq-power-bi-ai-agents/ #FabricIQ #MicrosoftBuild