A glowing “Governed Context Layer” connects Work, Fabric, Foundry, and Web IQ tools to multiple digital devices.
Microsoft used Ignite 2026 to launch Microsoft IQ, and the pitch is that the AI model is no longer the differentiator. Every company can rent the same frontier models. Microsoft IQ brings together Work IQ, Fabric IQ, Foundry IQ and Web IQ, plus more than 350 third-party connectors, to create an organization's enterprise context. Microsoft calls it an enterprise context management platform. In plainer terms, it is a governed layer between your company's data and whichever agent you point at it.

The announcement came from Yitzhak Kesselman, the corporate vice president of Microsoft IQ. It is a vendor announcement. The performance, savings and "advantage that compounds" claims are Microsoft's own, and the post gives no independent benchmarks.

A glowing “Governed Context Layer” connects Work, Fabric, Foundry, and Web IQ tools to multiple digital devices. The big idea: business plugins​

The packaging mechanism is the business plugin. Business plugins let teams curate and govern the context, skills and authoritative sources that every agent inherits. Microsoft's Azure Ignite overview gives an example. A plugin can tell an agent to use the components and orders semantic model in Fabric IQ to find affected orders, the supplier contracts in Foundry IQ to check late-delivery terms, and Work IQ to find who owns the decision.

The reuse promise covers many destinations. Users can build business plugins wherever they run agents, including Microsoft Copilot, GitHub Copilot, Microsoft Foundry through Model Context Protocol (MCP), and the ChatGPT and Claude marketplaces.

Availability of the plugins themselves is narrower than the headline suggests. Microsoft's Ignite materials say the new business plugins are available as part of the Frontier program. Treat them as early access rather than a broad rollout.

The four IQs​

  • Work IQ covers how people work: documents, conversations, collaboration and past decisions.
  • Fabric IQ covers the state of the business: governed data, entities, relationships and rules.
  • Foundry IQ covers curated institutional knowledge, such as policies and authoritative documents.
  • Web IQ covers current web context, including news, images and videos.

What's generally available and what's still preview​

Microsoft mixes generally available and preview items throughout the post. If you plan a pilot, this is the part to read closely.

AreaGenerally availablePreview or later
Work IQBusiness Applications (CRM and ERP records), Retrieve, real-time meeting context, long-running tasks, agent-to-agent support, granular MCP policies, OpenAI and Anthropic pluginsMemory, Word/Excel/PowerPoint creation and editing, Listen event subscriptions, agent policy governance with human approval
Fabric IQPower BI semantic models in Microsoft CopilotFabric IQ in Excel's Edit with Copilot, ChatGPT and Claude connectors, ontologies in GitHub Copilot, Fabric apps, IQ sharing for ontologies. Fabric data agents in Copilot are "coming soon."
Foundry IQSharePoint (indexed) knowledge source with permissions supportKnowledge graphs, Microsoft Retriever 1. Foundry IQ Serverless is due for general availability in December 2026.
Web IQAvailable to every Azure customer, no waitlistCommerce is in private preview

There are discrepancies to watch. The Azure post lists Business Applications in Work IQ as generally available. The Work IQ team's own Tech Community post is more cautious. It says Business Applications in Work IQ brings ERP (preview) and CRM from Dynamics 365, and Power Platform data, together with Microsoft 365 context. That suggests the ERP piece may still be preview even where CRM is generally available. Check the specific component against current documentation before you commit.

The Anthropic plugin is another case. The Azure post lists OpenAI and Anthropic plugins as generally available. The Work IQ post says the Work IQ plugin is now in the ChatGPT plugin directory for ChatGPT Business and Enterprise customers, while the Anthropic plugin is coming soon. The two posts do not obviously agree on the Anthropic plugin, so confirm its status before planning around it.

Governance: where IT admins should look​

The post leans on trust and control. Several items matter to an enterprise IT team:

  • Permission trimming. Retrieve is described as returning grounded, permission-trimmed Microsoft 365 content and federated connector data in a single call. The Work IQ team adds that Work IQ respects your organization's Microsoft 365 controls and only retrieves information you have permission to access.
  • Granular MCP policies. Access can be scoped by path, audience and trusted app, rather than through a tenant-wide switch.
  • Human approval. The public-preview agent policy governance lets agents pause for approval before consequential actions.
  • Web IQ controls. Scopes let organizations include or exclude specific sites and domains. Bring Your Own License combines licensed publisher content with web results.

Governance that travels with a plugin into ChatGPT or Claude is a hard problem. Microsoft says permissions and governance carry across supported assistants. The material reviewed does not show how that works in practice, so test it in your own tenant. Also decide in advance whether connecting a Microsoft-sourced plugin to a third-party assistant is sanctioned by your policy.

Web IQ: the pricing detail​

Web IQ is the one component with a concrete price. Microsoft says it is open to every Azure customer, with no waitlist, at $1 per 1,000 calls. Auto and Auto Advanced modes unify Web, Images, Videos, News, Places and Weather retrieval. Places and Weather are new. Microsoft cites 207.5 million real-world entities. The Commerce preview covers more than 40 billion product listings and offers.

That price is a starting point. The announcement does not explain how calls are counted or what the higher modes cost. The "less development effort and lower cost" claim is Microsoft's.

Foundry IQ Serverless is also pay-as-you-go, with no capacity to provision. It has free monthly allowances. No prices or allowance amounts were published in the post.

Continuous improvement in Foundry​

The announcement also covers agent improvement. Teams define success criteria and compare quality, latency and cost per successful outcome. Foundry and Agent Optimizer will continually improve systems and agents with quality, speed, cost and outcomes as the benchmark. Agent Optimizer generates candidate configurations and ranks them, and teams approve changes before deployment.

When configuration tweaks stop helping, the same criteria can drive post-training. Managed Training is the automated option. Programmed Training is in public preview. It lets AI engineers write custom training loops, including supervised fine-tuning and reinforcement learning, for supported open-weight models, with Foundry managing the compute. The post does not name the supported models or give training prices.

The customer example: AEMO​

Microsoft's showcase customer is the Australian Energy Market Operator. Its NEM-GEM capability was co-engineered with Microsoft in under 12 weeks. Microsoft says it helps operators filter thousands of signals and decide faster, with humans staying in control. That is a single vendor-supplied case with no measured outcomes.

Analysis: who should care​

Strengths.

  • Reusable context avoids re-explaining the business to every new agent.
  • Permission-aware retrieval is the right default for enterprise data.
  • Support for non-Microsoft assistants reflects real multi-vendor environments.

Open questions.

  • Lock-in versus openness. Microsoft calls the platform open and model-diverse. But the context layer sits on Microsoft's Fabric, Foundry and Microsoft 365 data, so the center of gravity is still Microsoft.
  • Preview sprawl. Many headline features are preview or private preview, and the "coming soon" items have no dates.
  • Metering. Cost depends on call counts, serverless usage and training compute that the post does not price.
  • Security validation. No independent testing of cross-assistant permission behavior has been published.

A practical way to start​

Microsoft's own advice is modest: start with one meaningful business process where a better decision would deliver measurable value. Based on the announcement, a sensible pilot looks like this:

  1. Pick one process where data already lives in Microsoft 365, Dynamics 365 or Fabric.
  2. Confirm which Work IQ, Fabric IQ and Foundry IQ pieces you need are generally available. Confirm whether you can get into the Frontier program for business plugins.
  3. Review MCP policies and permission trimming with your identity and data governance teams.
  4. Define success criteria for quality, latency and cost before you build, so Agent Optimizer has something to measure.
  5. Decide whether third-party assistants such as ChatGPT or Claude are in scope.

Microsoft IQ is a credible attempt to turn scattered enterprise data into a reusable asset for agents. The proof will come from tenant-level testing, published pricing and independent security validation, none of which the launch materials provide yet.

 

References

  1. Introducing the new Microsoft IQ: Putting your organization's intelligence to work in every agent - Microsoft Azure Microsoft Azure 2026-11-17T18:05:00+00:00
  2. Microsoft Ignite 2026: What's new in Azure, the platform purpose-built for AI azure.microsoft.com
  3. What's new in Work IQ at Ignite 2026: agents that get work done techcommunity.microsoft.com