Microsoft has announced a sweeping update to its Dataverse platform, set to roll out in July 2026, that will deeply integrate business data into the Microsoft 365 Copilot experience and introduce new agent-building capabilities. The move, revealed on June 29, marks a significant step toward making enterprise data more actionable through AI-powered agents and signals Redmond's commitment to bridging the gap between structured business information and everyday productivity tools.

The July wave centers on three major pillars: embedding Dataverse data directly into Microsoft 365 Copilot's contextual understanding, launching public preview of semantic models, and making Business Skills generally available. Additionally, the update introduces support for the Model Context Protocol (MCP), a nascent standard that streamlines how agents interact with external tools and data sources. Together, these features aim to transform Dataverse from a backend data store into the intelligent backbone of an agentic enterprise.

Business Data Meets Copilot: Contextual Grounding Gets Real

For years, Microsoft has promised that Copilot would not just chat but act with full awareness of organizational data. The July wave delivers on that vision by allowing Dataverse tables and records to serve as native grounding sources for Microsoft 365 Copilot. Instead of relying on generic web knowledge or limited connectors, Copilot will now be able to pull real-time sales figures, customer histories, inventory levels, or any other structured data from Dataverse when users ask questions in Outlook, Teams, or Excel.

This integration is not a simple copy-paste of records. Microsoft has designed it to respect Dataverse's rich security model, meaning users will only see data they are authorized to access. Business unit hierarchies, field-level security, and row-level permissions all carry over into the Copilot context. For IT administrators, this eliminates the headache of building and maintaining separate vectorized copies of data for AI consumption. Instead, the existing Dataverse security trim applied to SQL queries, PowerApps, and Power Automate now extends to natural language interactions with Copilot.

Early adopters in the Power Platform community have already praised the approach. During a private preview, one healthcare provider reported that their Copilot-powered CRM assistant could instantly surface a patient's full history from Dataverse when a clinician typed “summarize recent consultations” into Teams. The assistant pulled data from multiple related entities—appointments, prescriptions, notes—and formatted a coherent summary, all without the clinician leaving the communication hub.

Semantic Models Enter Public Preview

Alongside the Copilot integration, Microsoft is finally opening up Dataverse's semantic modeling layer to a broader audience. The public preview of semantic models, first teased at Build 2024, gives organizations a no-code way to define business logic and relationships that sit atop raw Dataverse tables. Think of it as a curated, AI-friendly abstraction that hides the complexity of underlying table structures while exposing meaningful business concepts like “cactive customer,” “mmargin-rich product,” or “hhigh-priority order.”

These models are not merely views. They include pre-calculated aggregates, natural-language synonyms, and even inferred hierarchies that help AI models reason about data more accurately. For example, a semantic model could define that “sales territory” actually comprises a nested set of postal codes, so when a user asks Copilot, “show me sales by territory,” the system doesn't stumble over the lack of a “territory” column but rather computes the correct grouping on the fly.

What makes this preview notable is its tight integration with both Azure AI Foundry and Microsoft Fabric. Once a semantic model is published, it becomes consumable not only by Dataverse agents and Copilot but also by Fabric lakehouses and third-party AI tools that speak the Dataverse language. This positions Dataverse as a single source of truth that feeds every corner of the Microsoft intelligent data platform.

Business Skills Go Generally Available

Perhaps the most practical upshot for citizen developers is the general availability of Business Skills. These are pre-built, reusable AI actions that agents can invoke to complete discrete tasks—such as “fevaluate approval likelihood,” “ggenerate quote,” or “ccreate follow-up task.” Unlike the generic prompts of early Copilot experiences, Business Skills are deterministic, auditable, and governed by administrator-set policies.

Each Business Skill encapsulates not just the AI logic but also the required data connections, security checks, and output schema. For instance, a “chcheck credit limit” skill could automatically reach into Dataverse to retrieve a customer’s current balance, compare it against a threshold defined in a semantic model, and return a Boolean with an explanatory note. Because the skill is versioned and tracked, enterprises can roll back or update it without breaking dependent agents.

Microsoft confirmed that dozens of skills for common Dataverse scenarios (sales, service, supply chain, HR) are shipping with the GA wave. Furthermore, makers can now create custom skills using a new point-and-click designer directly in the Power Apps portal. Under the hood, these skills generate natural language descriptions that make them discoverable by Copilot’s orchestrator—a crucial feature for scaling agent-based automations across large organizations.

The Model Context Protocol: Connecting Agents to the World

In a move that surprised many industry observers, the July wave incorporates support for the Model Context Protocol (MCP), an open standard originally championed by Anthropic but now gaining multi-vendor traction. MCP provides a uniform way for AI agents to discover and interact with external tools and data sources through a client–server architecture. By baking MCP into Dataverse, Microsoft enables agents built on the platform to reach far beyond the Power Platform ecosystem.

This means a Dataverse agent can now, for example, invoke an MCP server that brokers requests to a legacy SAP system, a custom Python model, or an Internet-of-Things sensor network. The MCP server handles authentication and protocol translation, while the agent simply issues a natural language instruction. From a governance standpoint, administrators can whitelist specific MCP endpoints and monitor all tool calls through Dataverse’s existing audit logs.

Microsoft’s endorsement of MCP also sends a strong signal about the future of agent interoperability. By adopting an open standard rather than a proprietary bridging layer, the company reduces lock-in and encourages a richer ecosystem of connectors. Some analysts see this as a direct response to Salesforce’s recent Agentforce platform, which relies on a more closed model for third-party integration.

Governance and Security: No Trust, No Agents

All these capabilities would be meaningless without robust governance, and Microsoft knows it. The July wave introduces a new Agent Governance Center within the Power Platform admin console that gives IT teams a single pane of glass to manage policies, monitor usage, and set guardrails for Dataverse agents. From here, admins can enforce what data Copilot can ground on, which Business Skills are permitted, and which MCP servers agents may contact.

A key innovation is the concept of “trust zones.” Organizations can define concentric rings of data sensitivity—say, core financial data, customer PII, and publicly shareable information—and automatically block agents from mixing zones unless an explicit exception is granted. Attempts to violate these boundaries are logged and can trigger automated alerts to security operations centers.

Additionally, all agent–user interactions are captured in a new compliance log that feeds into Microsoft Purview. This means that if a user asks Copilot to “tell me the top five delinquent accounts,” the resulting conversation—including the grounded data records—is fully discoverable for eDiscovery and internal audits. Privacy advocates within the enterprise will welcome the fact that these logs support retention policies and can be pseudonymized for non-production use.

Early Community and Analyst Reactions

Though still days away from general availability, the July wave has already sparked lively discussion on the Windows Forum and across LinkedIn. One power-user thread highlighted the potential for semantic models to finally solve the age-old problem of “Frankenstein” Dataverse tables where well-meaning makers create hundreds of columns with cryptic names. “If you can simply map business terms to these messy tables once, every future AI interaction becomes cleaner,” wrote one community member.

Industry analysts are also upbeat. Forrester’s principal analyst for low-code platforms noted that the combination of Business Skills and MCP could make Microsoft the main competitor to standalone agent-building frameworks like LangChain. “Microsoft is playing a long game here,” the analyst wrote in a research note. “They aren’t just giving you a hammer; they’re giving you a prefab house with all the wiring already in place.”

Some skeptics, however, question whether the performance of semantic models at scale will match the promise. Dataverse has historically struggled with large datasets when complex aggregations are involved, and adding an AI-friendly layer on top could exacerbate latency. Microsoft’s engineering team has published benchmark data showing sub-second response times for models built on tables with up to 100 million rows, but real-world validation remains pending.

What This Means for Windows and Microsoft 365 Users

For the Windows-centric professional, the July wave is yet another reason to stay embedded in the Microsoft ecosystem. Imagine drafting a client proposal in Word and having Copilot suggest real-time pricing from your Dataverse product catalog, then generating a tentative contract using a Business Skill, all without ever switching apps. This level of integration goes beyond the superficial AI assistants that exist today and starts to resemble a true digital coworkers.

Windows 11, with its Copilot sidebar and deep WebView2 integration, becomes an even more strategic endpoint. Microsoft has confirmed that the new Dataverse features will work seamlessly within the Copilot Windows app, so field workers on ruggedized tablets or knowledge workers on Surface devices alike can benefit. Offline support, however, is not part of this release, which may frustrate road warriors.

IT decision makers should start preparing now by auditing their Dataverse environments for data quality issues and legacy security roles that might leak sensitive information when Copilot starts reading. Microsoft provides a free readiness assessment tool in the Power Platform admin center that scans for common misconfigurations. Those who delay risk a messy launch where AI gives users exactly the wrong data.

Looking Ahead

The July 2026 Dataverse wave is not a one-off event but rather a down payment on Microsoft’s broader agentic vision. Insiders hint at a fall release that will bring multi-agent orchestration to Dataverse, allowing a single Copilot request to spawn a swarm of specialized agents that negotiate with each other behind the scenes. Meanwhile, the open-source community is already building MCP servers that wrap popular APIs like OpenAI’s text-to-video, indicating an explosion of fringe capabilities just around the corner.

For now, enterprises can look forward to a more useful Copilot, a more intelligible Dataverse, and a set of tools that finally make AI agents accessible to the non-developer. The clock is ticking; the wave hits the shore in just a few weeks. Will your data be ready?