A friendly robot presents a Dataverse project-management data model in a modern office with analytics dashboards.
Microsoft's AI at Work roadmap has an item, ID 577815, that promises something Dataverse makers have wanted for a while. You describe a business system in plain English, and a coding agent builds it. The roadmap lists the item as "In development", with preview in June 2026 and general availability targeted for February 2027. Much of the plumbing already exists and can be tried today.

A friendly robot presents a Dataverse project-management data model in a modern office with analytics dashboards. What the roadmap item says​

Microsoft's own example is a recruiting system for a fictional company called Zava Construction. The prompt asks for five tables: Positions, Candidates, Interviewers, Interviews and Feedback. It also asks for lookups, a many-to-many relationship between Candidates and Positions, and a self-referential interview chain. The agent is told to put everything in a "Zava Recruiting" solution, load sample data and report which candidates are currently interviewing.

From that one prompt, the roadmap says the agent works autonomously through these steps:

  • It discovers the Dataverse environment and configures MCP.
  • It creates the solution using the Microsoft Power Platform CLI.
  • It builds the five tables, with choice columns, lookups and the many-to-many relationship.
  • It generates and runs a Python script to bulk-load realistic sample data.
  • It queries across the tables to answer the business question.

Microsoft's pitch is that you don't need skill names or tool flags. You also don't switch between documentation, the CLI and scripts. The stated business value is that you no longer build tables and solutions by hand in the maker UI.

The roadmap's general business-value text also mentions forms and views. The worked example doesn't list them. I wouldn't assume every run produces them.

The tooling behind it​

A Microsoft Power Platform blog post from May 2026 describes the Dataverse plugin for coding agents, then in public preview. It packages four tools: Dataverse MCP servers for ad-hoc discovery and natural-language queries, the Dataverse CLI (in preview) for data-plane actions, a Python SDK for batch and scripted operations, and the PAC CLI for admin tasks such as solution export. That matches the roadmap flow, where the CLI builds the solution and Python handles bulk loading.

The open-source repository adds detail on what each skill does:

  • dv-connect is one-time setup. It installs the Dataverse CLI, Python SDK and PAC CLI, authenticates against your environment, and registers the Dataverse MCP server with your agent.
  • dv-metadata authors and edits the data model: tables, columns, relationships, forms and views.
  • dv-solution manages the solution lifecycle: create, export, import, promote across environments and validate deployments.

The agent-support list has grown. A July 2026 Microsoft post says the plugin is now available for OpenAI and Codex, in addition to Claude, Cursor and GitHub Copilot. The roadmap item itself doesn't name supported agents, so treat that list as the plugin's current state, not a commitment for this feature.

Prerequisites are modest. You need a Dataverse environment, which can come from Power Apps, Dynamics 365 or Power Platform plans, or from the free Power Apps Developer Plan.

Why Work IQ is in the title​

Microsoft is folding this into Business Applications in Work IQ. Its documentation says the layer brings data, processes and operational context from business apps into Copilot and agents. It also says Dataverse intelligence is now part of it, and recommends Work IQ over the separate Dataverse intelligence service, which is going away.

Two Work IQ pieces matter to this story:

  • Business skills. These are reusable natural-language descriptions of processes, policies and domain knowledge. Microsoft says plainly that they are not executable code. They guide agent behavior. They are stored as solution-aware Dataverse objects, so they can move between environments.
  • Skill access. Owners can share skills with everyone in the environment, with selected security roles, or with specific users or groups. Role-based access requires both the chosen role and the Skill Sharing Role. Basic Users can create skills they own. Environment Makers can also share them and manage solution deployment.

What admins need to check​

"The agent configures MCP" doesn't mean the agent skips your governance. Microsoft's documentation says Power Platform admins must enable Business Applications in Work IQ. That includes allowing coding-agent access for MCP use.

Controls cascade across three levels:

  1. Tenant level (Microsoft 365 admin center). A data-sharing setting lets Dataverse and Microsoft 365 Copilot data flow both ways. You can set it to all users, no users, or specific groups. If it's off, nothing is shared, even for apps that opted in.
  2. Environment level (Power Platform admin center). The Work IQ setting turns on business skills, semantic modeling and Dataverse access through Copilot Chat, Research and Agent Builder.
  3. App level (Power Apps maker portal). This setting decides which model-driven apps share their tables, forms and views.

Default rollout timing is separate from this feature's availability. Microsoft's default-settings page says that from September 30, 2026, Work IQ is on by default for new eligible environments, and from November 2026 for existing eligible environments. Exceptions include nonproduction environments (trial, sandbox and developer), EU/EFTA environments during preview, and government-related environments. The service isn't currently available in DoD, GCC or GCC High. Don't read the rollout dates as a general-availability date for solution building.

Compliance teams should read the same page. It warns that PCI DSS, HITRUST and FedRAMP coverage of a source service doesn't automatically extend to Copilot experiences that retrieve its data. FedRAMP tenants should review boundary, residency and routing implications first.

MCP: tools and a billing change​

The Dataverse MCP server lives at an endpoint in the form https://{dataverseOrgName}.crm.dynamics.com/api/mcp. Its documented tool list includes create_table, update_table, create_record and read_query, plus delete_record and delete_table. Both delete tools run only after explicit user approval, which is a sensible brake for an agent.

Microsoft's MCP page also notes a change that affects cost. Since December 15, 2025, Dataverse MCP tools are charged when accessed by AI agents created outside Copilot Studio. Dynamics 365 Premium licenses and Microsoft 365 Copilot user licenses exempt Dynamics 365 data access. The roadmap item says nothing about licensing, so budget conversations should start from the MCP billing documentation.

Tool names have also changed. Describe_table, list_tables and fetch were replaced by describe, and the old data search tool became search_data. If your MCP client keeps allow or deny lists by tool name, review them.

Reviewing the generated schema​

The relationship choices in the example have real consequences. In Dataverse, a lookup column creates a 1:N relationship. A many-to-many relationship uses a hidden intersect table and lets rows on either side associate with many rows on the other. Relationship behavior also decides what happens on delete, assign, share and reparent. Cascade settings are the classic way to cause trouble.

Before promoting anything an agent builds, check these points:

  • Do the cascade and delete behaviors match your business rules?
  • Is the primary name column free of sensitive data? Microsoft warns that anyone who can see a record can see the primary name of a record it looks up, even without permission to the related row.
  • Is the sample data clearly labeled and kept out of production?
  • Are security roles defined? Generated tables don't imply access design.
  • Does the solution import cleanly into a test environment?

Microsoft's guidance on skills is to test with real scenarios and validate that the agent does what you intended.

Open questions​

The roadmap doesn't say which coding agents will support the whole flow, what it costs, which permissions each operation needs, or how rollback works. Two Microsoft dates also disagree. The roadmap lists general availability for February 2027. The archived release-plan page lists a public-preview date of June 1, 2026 and no general-availability date. That page now says it is archived and points readers to the roadmap. Roadmap dates are estimates and can change.

Bottom line​

For developers, this is the Dataverse CLI, MCP and Python plumbing wrapped in a prompt. The wrapper is already testable, even though the roadmap item is still "In development". For admins, the useful step is to decide now who gets Work IQ and coding-agent access, and in which environments, before the November default rollout does it for you. Treat generated schemas as drafts that need review.

 

References

  1. Microsoft Dataverse: Enable AI powered business workflows in Work IQ with Dataverse - Build Dataverse solutions with coding agents using Dataverse skills Microsoft 365 Roadmap 2026-10-22T23:45:41.230844Z
  2. GitHub - microsoft/Dataverse-skills: Microsoft Dataverse skills for AI coding agents. Wraps the Dataverse MCP server, Dataverse CLI, Python SDK, and PAC CLI behind specialist skills for building, querying, deploying, and administering Dataverse environments. · GitHub github.com
  3. Dataverse Is Your Agent Data Platform: Here’s What’s New - Microsoft Power Platform Blog microsoft.com