Key Takeaways
- An enterprise context layer is the shared, governed store of business knowledge, data, rules and permissions that every AI agent draws on.
- Microsoft IQ, announced on 17 November 2026, bundles Work IQ, Fabric IQ, Foundry IQ and Web IQ with more than 350 connectors into one such layer.
- The model is becoming a commodity; curated context is where lasting advantage, and much of the risk, now sits.
- Fix permissions and pick one process before you buy any context platform, and keep your curated context portable.
Your competitors can license the same AI model you use, on the same day, at the same price. What they cannot license is the way your organisation actually works: its approval rules, its customer history, its pricing exceptions and the judgement of your best people.
That gap is now a product category. On 17 November 2026, Microsoft launched Microsoft IQ and described it as an enterprise context management platform. Whether or not you run on Microsoft, the idea behind it will shape your AI roadmap for 2027. By the end of this guide you will have a working definition, a clear view of how the layer works, and the questions to settle before any vendor pitch.
What is an enterprise context layer?
An enterprise context layer is a governed, shared foundation that gives every AI agent the same view of your business: documents, records, relationships, rules and who may see what. Instead of each project wiring its own data feeds, agents inherit one curated source of context, so answers stay consistent and permissions are enforced once.
Think of it as the organisational memory that sits between your systems of record and your agents. The model supplies reasoning. The context layer supplies the facts, the vocabulary and the boundaries that make that reasoning relevant to your company.
Most enterprises already own the raw material: email and chat archives, a document management system, an ERP, a CRM and a data warehouse. What they lack is a single layer that connects these sources, resolves conflicts between them and decides which agent may read which piece. Without it, every AI pilot rebuilds the same plumbing, and two agents can give a client two different answers to the same question.
What is Microsoft IQ and what does it include?
Microsoft IQ is Microsoft's enterprise context platform, announced at Ignite on 17 November 2026. It combines four components, Work IQ, Fabric IQ, Foundry IQ and Web IQ, with more than 350 third-party connectors, so agents built in Copilot, Foundry or other assistants can draw on the same governed business context.
According to Microsoft's launch post, each component covers a different kind of context:
- Work IQ: how people work, including documents, emails, meetings, chats and past decisions. CRM and ERP records are now part of its grounding set.
- Fabric IQ: the state of the business, through governed data, business entities, relationships and rules.
- Foundry IQ: curated institutional knowledge such as policies and authoritative documents, now with knowledge graphs in preview.
- Web IQ: current external context from the web, with controls to include or exclude specific sites.
The newest idea is the business plugin: a curated package of context, skills and approved sources that many agents can inherit. Microsoft says these plugins can run in Microsoft Copilot, GitHub Copilot and Foundry through the Model Context Protocol, and in the ChatGPT and Claude marketplaces. In his Ignite keynote, Satya Nadella called Work IQ "the most important database for any organization".
Why is the context layer becoming the new battleground?
Frontier models now improve every few weeks and are available to everyone, so they no longer separate leaders from laggards. What does is how well agents understand a specific business. Vendors are racing to own that layer, because whoever holds your curated context holds the most valuable and least portable part of your AI stack.
Microsoft's own framing is blunt: every organisation has access to the same frontier models, so the advantage lies in everything around the model. That matches what boards are seeing. A Gartner survey released on 19 October 2026 found that only 13% of CIOs say their enterprise has realised significant or transformational value from AI so far. Generic assistants save time at the margin; agents that understand your contracts, customers and processes change how work is done.
There is a risk side too. A Gartner survey of 298 infrastructure and operations leaders, published on 16 November 2026, found that 95% of organisations had faced at least one agent-related incident in the past 12 months. One of Gartner's seven recommended governance pillars is data perimeters that prevent context poisoning and leaks. In other words, the layer that makes agents useful is also the layer that can make them dangerous.
How does an enterprise context layer work in practice?
It works in four steps. Connectors pull content from email, files, ERP, CRM and data platforms. The layer maps entities such as customers, products and contracts, and how they relate. Each request is trimmed to the user's permissions. Curated packages of context and rules are then shared across agents, so improvements made once are reused everywhere.
A logistics example
Picture a Hong Kong freight forwarder whose customer service agent receives the message: "Why is my shipment to Rotterdam late, and will I be charged storage?" Without a context layer, the agent sees only the FAQ. With one, it can read the booking record in the ERP, the latest carrier email, the client's contract terms on demurrage and the account manager's note from last week's call, all within the permissions of the staff member it acts for.
The answer it drafts is specific, consistent with what finance will invoice, and traceable to sources. When the operations team later improves the delay-handling rules, the same curated package updates the billing agent and the sales agent too. That reuse is the economic case: Microsoft describes curated context as an asset that compounds, because each improvement carries into the next workflow.
How is a context layer different from RAG, a data warehouse or fine-tuning?
RAG is a technique for fetching relevant passages at question time; a context layer is the governed foundation that RAG and other techniques draw on. A data warehouse holds structured numbers for reporting. Fine-tuning changes the model itself. The context layer sits between them, combining documents, records, relationships and permissions for agents.
| Approach | What it changes | Best for | Main limitation |
|---|---|---|---|
| Retrieval-augmented generation (RAG) | What the model reads per question | Answering from a defined document set | Each project often builds its own pipeline |
| Data warehouse or lakehouse | Where structured data is stored | Reporting and analytics | Little meaning or permission logic for agents |
| Fine-tuning | The model's behaviour or style | Narrow, repetitive tasks | Knowledge goes stale; costly to repeat |
| Enterprise context layer | What every agent knows and may see | Many agents sharing one view of the business | Needs curation, ownership and clean permissions |
The approaches are complementary. A mature stack typically keeps the warehouse, uses retrieval inside the context layer, and reserves fine-tuning for the few cases where behaviour, not knowledge, is the problem.
What does a context layer mean for a Hong Kong enterprise?
It raises three decisions: whose platform holds your curated context, whether your access permissions are clean enough for agents to inherit, and how personal data flowing into the layer meets PDPO duties. Many local firms run mixed Microsoft, Google and local ERP estates, so portability matters more than any single vendor's features.
Permissions become exposure. Microsoft stresses that Work IQ returns permission-trimmed content. That protects you only if the permissions are right. Years of shared drives and broad distribution lists mean agents may surface files that staff could technically open but never knew existed.
Personal data is in scope. Emails, chats and CRM notes are full of personal data. The PCPD's guidance on agentic AI keeps the organisation accountable for what its agents process, so a privacy impact assessment belongs before the connectors switch on, not after.
Mixed estates are normal. A property management group may keep tenant records in a local system, finance in an ERP and email in Microsoft 365. A context layer that cannot reach the local system leaves agents with half the picture. Ask which connectors exist for your actual systems, and what it costs to build the missing ones.
Pricing is usage-based. Microsoft lists Web IQ from US$1 per 1,000 calls and says Foundry IQ Serverless reaches general availability in December 2026 with usage-based pricing. Budget for context retrieval as a running cost that grows with agent use, not as a one-off licence.
What do leaders get wrong about context layers?
The most common mistake is assuming the platform creates the context. It only connects and governs it; someone still has to decide which sources are authoritative, retire outdated documents and own each business definition. Other errors include skipping the permissions clean-up and starting with the whole company instead of one process.
- "Connect everything first." More sources mean more contradictions. Start with the sources one process genuinely depends on.
- "IT will own it." Context is business knowledge. Each curated package needs a business owner who signs off what counts as the right answer.
- "Lock-in is a later problem." Your curated rules and definitions are intellectual property. Keep them documented in a form you can move, and prefer open interfaces such as the Model Context Protocol.
- "Better context makes agents safe." It makes them more capable, which raises the stakes. Pair the layer with identity, approval and logging controls.
Treat context as an asset you own, not a feature you buy
Microsoft IQ will not be the only platform of its kind; other suites are moving in the same direction. The strategic question is not which logo sits on the layer. It is whether your organisation has decided what its authoritative knowledge is, who owns it and who may use it. Leaders who answer that now will get value from whichever platform they choose. Those who skip it will automate their existing confusion at machine speed.
Start with one process where a better answer has a measurable payoff, clean the permissions around it, and curate the context for that process first.
Empowering AI. Centered on you.
Start with one AI employee that knows your business
A context layer pays off only when it is attached to real work. UD's AI Staff Solution deploys AI employees by function, with data kept in Hong Kong or your private cloud in line with the PDPO, zero setup fee and a UD team handling configuration. Book a free AI employee assessment and we'll walk you through every step, from choosing the first process to curating its context and measuring results.
Book a Free AI Employee Assessment
Frequently Asked Questions
Do we need Microsoft 365 to build an enterprise context layer?
No. Microsoft IQ is one implementation. The same idea can be built with other platforms, retrieval tools and open connectors, though a Microsoft-based estate will find Microsoft IQ easiest to adopt.
How much does Microsoft IQ cost?
Microsoft has not published one bundle price. It lists Web IQ from US$1 per 1,000 calls and usage-based pricing for Foundry IQ Serverless from December 2026; other parts depend on existing Copilot, Fabric or Azure licences.
Can agents see files that employees should not see?
Agents inherit the permissions of the user they act for. If those permissions are too broad, agents will surface overshared files, so review access before connecting sources.
Is a context layer compliant with the PDPO by default?
No platform makes you compliant by default. The organisation remains responsible for the personal data its agents process, so run a privacy impact assessment and limit which sources hold personal data.
Who should own the context layer internally?
Shared ownership works best: IT runs the platform and security, while each business unit owns the definitions, sources and rules for its own processes.