Microsoft IQ bets on reusable context for enterprise AI agents | Magica
Microsoft IQ adds reusable business plugins, with savings still to prove
Editorial Team
••📖8 min read
Microsoft’s Ignite upgrades package enterprise knowledge, skills and tools into reusable business plugins. The expansion builds on IQ’s June release and competes with existing graph retrieval services; total savings and portability remain unproven.
Microsoft IQ’s new business plugins aim to let multiple agents inherit the same governed business knowledge, skills and tools.
The platform builds on earlier IQ releases; graph retrieval also has established competitors.
Several upgrades remain in preview, and the announcements do not establish the total savings from reuse.
Microsoft is expanding Microsoft IQ, its enterprise context management platform, with business plugins designed to carry a company’s knowledge and operating rules from one AI agent to the next. Announced at its Ignite conference on November 17, the plugins target what Microsoft describes as a source of repeated work: teaching successive agents where information lives, what it means and which tools they may use.
The software and cloud company already supplies the Microsoft 365 workplace applications, Fabric data platform and Foundry environment for building AI applications that feed this system. In its announcement, Microsoft says IQ combines four context services and more than 350 third-party connectors. New business plugins package context, skills and authoritative sources for reuse across agents, teams and platforms.
The commercial argument extends beyond cheaper model calls to the work of connecting agents to a business. Reusing that foundation could reduce integration and maintenance work. It would also put Microsoft’s context platform between an organization’s information and the assistants using it. The announcements establish the proposed mechanism, rather than measured savings across successive deployments.
AWS already offers graph retrieval. Microsoft’s knowledge graphs are in preview, with Foundry IQ Serverless general availability scheduled for December 2026. Graphic: Magica. Source: AWS.
What an agent would inherit
The four IQ services supply different kinds of information. Work IQ covers documents, conversations, collaboration and past decisions. Fabric IQ supplies governed business data, entities, relationships and rules. Foundry IQ connects agents to institutional knowledge in unstructured content, such as policies and authoritative documents. Web IQ adds outside information from the web.
Binance Wallet says subscriptions for its Anthropic IPO xStocks campaign reached the 500 million USDC limit. Final allocations, token delivery and applicable refunds remain pending; the cap does not establish proceeds received by the AI company.
Editorial Team
Business plugins are the proposed unit of reuse. Microsoft says they can work with Microsoft Copilot, GitHub Copilot, Foundry through Model Context Protocol, or MCP, and third-party assistants through the ChatGPT and Claude marketplaces. MCP provides a route for connecting agents to tools and context; the business plugin supplies the organization-specific material those connections need.
A connector makes information reachable; the plugin is intended to package the business definitions, trusted references and instructions needed to use it. Microsoft’s event announcement describes the platform as capturing, governing and reusing knowledge across the business.
Some changes deepen that foundation. Work IQ’s generally available business-application support adds customer relationship management and enterprise resource planning records to agents’ grounding material. Its Retrieve capability, also generally available, returns permission-filtered Microsoft 365 content and federated connector data in one call. Fabric IQ connectors for ChatGPT and Claude, which expose governed Power BI semantic models, remain in preview, Microsoft says in the platform announcement.
This is an expansion of an existing strategy. At Build in June, Microsoft had already described IQ as generally available, bringing together Work IQ, Fabric IQ and Foundry IQ, with organizational context accessible across GitHub Copilot, Foundry and Copilot Studio. Reuse across agents was therefore part of the earlier pitch, too. Web IQ was also announced at Build, initially for select Azure customers. Ignite broadens access and adds plugins and component upgrades; it does not mark the first appearance of the IQ platform.
Microsoft’s illustrative Foundry IQ interface shows a connected knowledge base and cited answer. Displayed metrics are company-reported demo figures, not measured workflow savings. Source: Microsoft. Source: Microsoft Azure.
Retrieval has a cost—and a permissions boundary
Foundry IQ’s upgrades show how Microsoft proposes to reduce repeated work inside an agent. Its new knowledge graphs use an organization’s ontology—a definition of its entities and relationships—to map connections across documents during indexing, retaining the source of each connection. Customers can also bring an existing graph and query it in natural language or the graph query language GQL. Topics, a complementary capability, organize content by theme for broader questions without requiring an ontology. Both capabilities are in public preview, according to the technical announcement.
Microsoft Retriever 1, also in public preview, is a specialized model for planning searches without deploying a frontier model for that retrieval plan. Microsoft says it runs on an existing index; its intended saving is less expensive search planning. The broader retrieval system is designed to reduce the number of searches an agent needs. The announcement does not provide a controlled comparison establishing how much an entire workflow saves.
A customer testimonial in Microsoft’s technical post describes Rockwell Automation’s plant troubleshooting assistant drawing on manuals, historical downtime records and expert insights, with each step linked to its source. The account illustrates the intended use when technicians cannot reach an expert, but supplies no comparative cost or accuracy results. Microsoft’s platform announcement also describes an operational intelligence system co-engineered with Australian grid operator AEMO in less than 12 weeks, with humans retaining control. That development time does not establish how much work a later agent could reuse.
Nor is graph-based retrieval itself new. Amazon Web Services, Amazon’s cloud business, made GraphRAG generally available in its Bedrock Knowledge Bases service on March 7, 2025. Its announcement describes automatically generating embeddings and an entity graph in Neptune Analytics, then combining vector search with graph traversal. AWS says the feature requires no additional setup or charges beyond its underlying services, and is available where both Bedrock Knowledge Bases and Neptune Analytics are supported. That makes graph retrieval an existing competitive option, although the two announcements provide no like-for-like cost or quality comparison. Microsoft’s broader case rests on integration with its workplace, business-data and agent tools.
Shared context also needs boundaries. Microsoft says Foundry IQ’s generally available source-permission controls enforce supported, configured SharePoint and Azure storage permissions under the caller’s identity, including access through MCP. Reusing a knowledge base therefore need not mean giving every agent identical access.
The finer controls have different release states. OneLake item-level permissions and Purview sensitivity-label enforcement are in public preview. Work IQ’s granular MCP policies are generally available, while its memory and agent policy governance with human approval are in public preview. Deployment decisions must therefore be made capability by capability. The technical announcement also schedules the public preview of a managed Foundry IQ resource in every new Foundry project for December 2026, distinct from Serverless general availability.
A shared foundation can also become a dependency
Microsoft’s incentive is visible in the architecture. The platform brings together systems it already operates, while accepting outside models and assistants. In the keynote, Microsoft described making the system underneath Copilot available as a platform for customers’ organizational intelligence. It also said effective optimization involved the product, agent tools and models together.
That approach gives Microsoft a role even when an organization changes the model or assistant doing the work. Cross-platform access may widen customer choice, but it does not by itself establish how easily the curated context, permissions and evaluations could move to another provider. That portability question follows directly from making shared context a lasting business asset.
There is also a reliability tradeoff: more reuse makes the shared service more consequential. An independent analysis places Microsoft’s pitch alongside similar efforts by other enterprise software vendors to put AI across application silos. It also notes that the keynote opened by addressing the previous week’s Microsoft 365 outage. That recent failure makes service resilience a concrete consideration for buyers relying on Microsoft for a shared foundation, alongside the promised reduction in duplicated work.
The next test is cost per successful outcome
Microsoft’s case for lower costs comes as its AI leadership says enterprises are reporting rapidly rising token spending, according to the conference report. That is an account of customer feedback from Microsoft, rather than a measured industry spending series. Its other cost proposals include specialized models and routing each task to an appropriate model; shared context is only one part of the strategy.
Foundry IQ Serverless is scheduled for general availability in December 2026, with usage-based pricing that scales to zero when idle and requires no capacity commitment. Web IQ is available to every Azure customer, Microsoft says in the platform announcement, with pricing starting at $1 per 1,000 calls. That is one component’s starting price, not the cost of running an IQ-based business workflow. The launch material supplies neither a complete umbrella price nor a like-for-like total-cost benchmark for the plugin approach.
Microsoft’s continuous-improvement tools offer a way to test the claim. Teams define success, compare quality, latency and cost per successful outcome, and approve configurations before deployment. Agent Optimizer ranks proposed improvements; post-training is a further option when configuration changes stop delivering gains. Managed Training automates customization; Programmed Training, in public preview, lets engineers write supervised fine-tuning or reinforcement-learning loops for supported open-weight models while Foundry manages the compute.
The decisive evidence will be whether a second and third agent can reuse a business plugin with less integration and maintenance work, while preserving answer quality and access controls. That comparison needs to include retrieval, indexing and governance costs. A useful pilot would compare that reuse with the organization’s existing approach and account for ongoing updates to the shared knowledge. Those results would determine whether IQ’s common foundation saves more work than it adds to operate.
Tencent says US restrictions on remote computing will likely slow its next AI model while leaving cloud revenue largely protected. Southeast Asian operators face licensing and financing uncertainty, though US cloud access and domestic demand offer partial alternatives.
Microsoft is rolling out Agent 365 Business Analytics through its Frontier early access program, pairing AI spending with estimates of assisted work. The dashboard expands its existing analytics as agent workloads add usage charges, but its dollar-value estimates do not establish financial returns.
Adobe’s Firefly update saves recurring visual subjects and creative preferences, extending an existing push into repeatable production. Memory has narrower reach than the assistant, while campaign demonstrations leave reliability and savings unmeasured.
The European Central Bank warns that debt and interlinked financing could carry an AI investment reversal into credit markets and the European economy. Resilient earnings, limited direct bank lending and stronger bank buffers temper its comparison with past bubbles.
Reported introductions helped Alice Huang secure Alibaba as a chip-rental customer and obtain financing. The relationship raises questions about Nvidia's overseas sales network, but does not itself establish illegal exports.
KPMG says AI agents pursued some issues more deeply in a parallel UK audit, though humans still had to correct their work. Client fee savings remain unconfirmed, even as KPMG International has negotiated lower fees from its own auditor by citing AI.
Payhawk announced a $280 million all-equity Series C led by Growth Equity at Goldman Sachs Alternatives, valuing it at $2 billion. The funding backs an expansion from managing business spending into broader AI finance operations, where competitors already combine automation, payments and controls.
A 16-country survey of 15- to 22-year-olds finds a median of 33% have used AI as a companion or friend. Users have higher adjusted odds of loneliness, but the research cannot establish whether AI helps, harms or replaces human connection.
South Korean President Lee Jae Myung has ordered a review of rewards of at least 100 million won for reporting technology-assisted exam cheating and legislation requiring technology companies to supply evidence. Ahead of the November 19 college entrance exam, the proposals would add informant incentives and post-exam investigations to classroom checks, but implementation rules remain unsettled.
Microsoft is pitching Copilot Workspaces as a shared environment for people and AI agents. Team collaboration is already in early access, but personal Autopilot integration is still coming soon, with final agent pricing unannounced.
Google is rolling out file references and sharing for Gemini skills as adult personal accounts lose the ability to use Gems. Automatic conversion preserves supported material, but file limits and missing tools can interrupt existing workflows.
Muddy Waters disclosed a short position in Nvidia while questioning a possible connection involving senior adviser Ajay K. Puri. Its own account leaves the key identity question unanswered, separating its new claims from an existing server-diversion prosecution.
Nvidia reported $115.8 billion in quarterly revenue in its first earnings release since co-founder Jensen Huang resigned. Strong demand gives interim CEO Colette Kress a financial cushion, while a campaign for Huang’s return collides with unresolved investigations and the board’s reported opposition.
OpenAI launched GPT-6.2 Astra and Sol for professional work, cutting Astra’s standard input and output token prices by 20%. The models offer cheaper paths through selected tasks, but research evaluations, reporting failures and unfinished independent safeguard testing complicate the case for broader autonomy.
Microsoft’s first live speech-to-speech model offers OpenAI API compatibility and planned enterprise integrations. Its published evaluations show a low-cost option, with several rivals completing more agentic tasks.
Canberra-based Trellis Data is offering its 220-billion-parameter Agentaus 1 model to business and government. Its pitch goes beyond keeping data in Australia, but performance, deployment costs and a planned capital raising remain key tests.
Perplexity is adding direct collaboration between frontier models to Computer, letting a lead agent coordinate up to 16 supporting agents. Company-reported benchmarks suggest mixed teams can outperform repeated copies of one model, but leave customer costs and production reliability unresolved.
Marsh reported 105 client cyber incidents in the last two weeks of October, against 48 in the final two weeks of May. The broker links the increase partly to AI, while warning that insurance panel access may not guarantee an available response team.