On June 2, 2026, Microsoft used the Build keynote stage to announce something the industry has needed for eighteen months but that no vendor had cleanly packaged: a single intelligence layer that gives AI agents permissioned, real-time access to the entire enterprise data estate. Microsoft IQ went generally available across GitHub Copilot, Microsoft Foundry, and Copilot Studio that morning, with the Work IQ APIs following on June 16 (Microsoft 365 Blog, Crypto Briefing).

The headline buyers care about is not the architecture. It is the diagnosis. After three years of pilot enthusiasm, the single most consistent reason enterprise AI agents fail in production is the same problem that has plagued every previous generation of enterprise software: the data they need lives in silos they cannot reach. Ninety-five percent of IT leaders cite integration as the primary AI adoption barrier, and Gartner reports that 63% of organizations do not have — or are unsure they have — the data management practices required for AI to work (Joget). Microsoft IQ is Microsoft's answer to that diagnosis. Whether it is the right one is a $600 terabyte question.

What Changed

Microsoft IQ is a four-engine context platform that sits between AI models and enterprise systems, normalizing data, applying permissions, and delivering agent-ready signals on demand. The four engines and what they actually do:

• Work IQ processes Microsoft 365 signals — email, calendar, meetings, chats, files, people, and collaboration patterns — and builds a continuously updated semantic understanding of how work flows through the organization. Microsoft says Fortune 500 customers carry an average data footprint exceeding 600 TB inside Work IQ (Microsoft 365 Blog).
• Fabric IQ connects agents to structured business data through OneLake, Microsoft's unified data lake, so an AI agent answering a question about quarterly revenue can query actual analytics infrastructure instead of guessing (Petri).
• Foundry IQ handles knowledge retrieval across enterprise sources — documents, knowledge bases, internal repositories — with governance and policy compliance baked in.
• Web IQ is a model-agnostic web passage retrieval API delivering grounding data at sub-165ms P95 latency, roughly 2.5x faster than Microsoft's prior generation of web grounding tools (Microsoft AI).

Layered on top sits the Foundry Agent Service, which lets developers deploy hosted, long-running stateful agents that maintain context across sessions and execute multi-step tasks over extended periods — a category sometimes called "always-on" agents that Microsoft also introduced with its Scout autopilot last week (VentureBeat).

The Work IQ APIs are the developer surface. Public preview is on GitHub today, with five domains hitting general availability on June 16:

1. Chat — programmatic access to Microsoft 365 Copilot, returning answers with citations and agent attribution.
2. Context — agent-ready contextual aggregates in formats optimized for prompt construction.
3. Tools — simplified verb-based operations against Microsoft 365 entities (send email, schedule meeting, upload document).
4. Workspaces — secure intermediate-state storage inside tenant boundaries so agents can hold reasoning state without leaking data.
5. A consumption-based billing layer that runs on Copilot Credits, with a new cost management dashboard in the Microsoft 365 admin center for spend caps and per-user limits.

Charles Lamanna, Microsoft's EVP of Copilot, Agents, and Platform, framed the strategic intent: "Work IQ is a new intelligence layer for Microsoft 365, designed to understand how work gets done." The technical reality is more pointed: Work IQ is the first commercially shipped attempt to expose Microsoft 365's accumulated organizational knowledge as a permissioned API rather than locking it inside individual Copilot user sessions.

Permissions and provenance are not bolt-ons. IQ respects existing Microsoft Entra identity boundaries and inherits Microsoft Purview sensitivity labels, so a sales agent cannot see HR severance documents and a contractor cannot read board materials. Every action is auditable and discoverable, which matters because the same week IQ launched, Noma published research showing that 65% of recent AI agent breaches stemmed from access-control failures at the integration layer (related analysis).

Why This Matters

For CIOs and CTOs, IQ is a structural change to how enterprise AI projects are scoped. For the last two years, the dominant integration pattern for agents has been bespoke: every new use case required custom plumbing into SharePoint, custom connectors into Salesforce, custom retrieval against Confluence, custom permission mapping in code. That work is the reason 78% of enterprises run agent pilots but fewer than 15% scale them to production (HackerNoon). Custom integration does not scale; it accumulates. Each new agent inherits the integration debt of the last three.

IQ collapses that pattern into one platform call. An HR self-service agent that previously needed three sprints of integration work — Workday connector, M365 calendar access, Purview policy mapping — now consumes a single Work IQ Context API and inherits all three out of the box. The architectural payoff is large, and it shows up in two places enterprise leaders care about: time-to-pilot and cost-to-scale.

For CFOs, the cost story is more nuanced. IQ rides on consumption-based pricing in Copilot Credits, which means the line item that used to live in IT integration budgets — labor — moves to a usage meter Microsoft controls. That is a recurring decision every enterprise will revisit quarterly: do you pay platform rent to Microsoft, or do you pay engineering salaries to maintain custom integrations? The new cost management dashboard is Microsoft's acknowledgment that finance leaders need governance on the usage side before they will sign off on production deployments. CFOs who have lived through the past 90 days of GitHub Copilot's usage-based billing transition will already be familiar with the failure mode: developers turn on agentic features, monthly bills spike 200-300%, and the next budget meeting becomes a fight (related coverage).

For COOs and business unit leaders, IQ changes which use cases are actually viable. The Microsoft IQ product page lists four canonical scenarios that map directly to operating-model pain points:

• HR self-service agents that reduce help-desk volume by reading policies, Workday entitlements, and the employee's own calendar.
• Sales enablement agents that pull prospect research, CRM history, engagement scoring, and recent email threads into a single brief.
• Customer support triage agents that route tickets to the right engineering team based on issue type, customer tier, and historical pattern.
• Product development agents that combine consumer behavior data, internal roadmaps, and competitive signals.

What used to take a six-month integration engagement with a systems integrator now becomes a Copilot Studio configuration. That is not theoretical — Accenture's $743,000-seat Microsoft Copilot deployment from April was the proof point for what this looks like at scale (related case study). What IQ adds is the data layer those agents needed to actually be useful.

The risk worth naming explicitly is concentration. If every new enterprise AI use case routes through Microsoft IQ, the share of total IT spend running through a single vendor crosses thresholds that most CIOs already consider unacceptable. Berkshire's June 2 disclosure that Apple now holds $10 billion of Alphabet stock — and the cloud-concentration anxiety it surfaced — applies in miniature to every enterprise weighing whether to standardize on IQ (related coverage). The Forrester forecast that 30% of enterprise vendors will ship their own MCP servers in 2026 is the hedge: if you can swap context engines the way you swap LLMs, you preserve optionality.

Market Context

Microsoft is not first to the enterprise context-layer race. It is, however, the first to ship a unified version with the distribution to make it the default for the Fortune 1000. The competitive landscape as of June 2026:

• Glean raised at a $7.2 billion valuation in early 2026 and remains the strongest standalone enterprise search and work assistant, with 200+ enterprise connectors and the cleanest knowledge-graph story. Glean wins where companies have already invested in cross-SaaS search.
• Iconiq's Whirl AI (prior coverage) raised a Series B specifically to compete in the enterprise-context-layer category, with strong adoption in financial services where data residency and on-prem deployment matter.
• Salesforce Data Cloud + Agentforce delivers a similar bundle inside the Salesforce ecosystem and now extends cross-platform to Google Cloud (related coverage). For organizations whose system of record is Salesforce, Agentforce is the lower-friction path.
• ServiceNow's Project Arc and AI Control Tower (related coverage) competes on the orchestration and governance layer, but its data integration footprint is narrower than Microsoft IQ.
• SAP's Joule Studio 2.0, rolling out in June 2026, packages 224 AI agents and 51 assistants for finance, supply chain, HR, and procurement workloads — a parallel play that ties agents to SAP's business-process spine (SAP News).
• Model Context Protocol (MCP) has now hit 97 million package downloads and is becoming the de facto integration standard for agents (related coverage). Microsoft IQ supports MCP, which preserves the option to point IQ-enabled agents at non-Microsoft data sources.

The analyst view is bifurcated. Gartner's August 2025 forecast that 40% of enterprise applications will integrate task-specific AI agents by the end of 2026 still tracks. But Gartner also predicts that 50% of AI agent deployment failures by 2030 will trace back to insufficient runtime governance and multisystem interoperability — the exact problem space IQ is targeting. Forrester's 2026 prediction that 30% of enterprise vendors will ship MCP servers reinforces the trend: the context layer is becoming a category, not a feature.

What makes Microsoft IQ structurally different is distribution. Microsoft 365 reaches roughly 400 million paid commercial seats, GitHub reaches 150 million developers, and Copilot Studio is already inside the IT estate of most Fortune 1000 customers. IQ doesn't need to win a sales cycle to land — it lands the moment a Copilot admin enables it. That is a fundamentally different go-to-market than Glean or Whirl can match in 2026.

Framework #1: The 4-Engine Decision Matrix for Microsoft IQ Adoption

Picking which IQ engine to start with — and which to defer — is the most consequential architecture decision enterprise teams will make over the next two quarters. The matrix below maps the four engines to the use cases, team sizes, data prerequisites, and risk profiles where each is the right starting point.

Engine	Best For	Team Size	Data Prerequisite	Primary Risk	Quick-Win Use Case
Work IQ	Knowledge-work productivity, internal Q&A, document drafting	5,000+ M365 seats	Mature M365 adoption (>70% DAU), Purview labels applied	Privacy backlash if email/chat surfaced without communication plan	HR self-service for benefits, leave, expense policies
Fabric IQ	Numeric analysis, KPI agents, finance/ops dashboards	Any size with OneLake	Data lake populated with curated tables, semantic model defined	Hallucinated numbers if semantic model is incomplete	Quarterly revenue Q&A for sales leadership
Foundry IQ	Knowledge management, policy and procedure agents, IT/legal	1,000+ users	Documents tagged, ownership clear, retention policies enforced	Stale information if document refresh process is weak	Engineering runbook search and on-call triage
Web IQ	Competitive intel, news monitoring, research agents	Any	None — but allowlist required for compliance	Hallucination from low-quality sources, brand safety	M&A research, regulatory tracking, vendor due diligence

Decision rules:

• Choose Work IQ first if your bottleneck is internal knowledge friction — people asking "where is the latest version of X" or "who owns Y." This is where IQ's ROI is most defensible because the baseline (people asking colleagues) is so expensive.
• Choose Fabric IQ first if your bottleneck is decision latency — leaders waiting on analysts to pull numbers. This is where hallucination risk is highest, so insist on a curated semantic model before exposing agents to executives.
• Choose Foundry IQ first if you have a mature document estate and a specific high-volume support workflow. IT runbooks, legal policy Q&A, and customer-facing knowledge bases are the canonical entry points.
• Choose Web IQ first only as a complement to one of the others. Web grounding is a feature, not a workload. Lead with internal data, then layer external context.

Anti-pattern to avoid: turning all four engines on simultaneously and letting agents query whatever they want. Every successful Microsoft IQ rollout we've seen in the first 90 days starts with one engine, one use case, and one business unit. Scaling comes after the first agent ships measurable results.

Framework #2: The 10-Point Pre-Deployment Readiness Checklist

Use this checklist before committing to a Microsoft IQ pilot. Score each item 0 (not ready), 1 (partial), or 2 (ready). Total of 16 or higher = green light. 10-15 = pilot with caveats. Below 10 = do the prep work first.

Identity and governance:

1. Microsoft Entra identity governance is current. All users provisioned through Entra, sensitive groups defined, conditional access enforced. (0/1/2)
2. Microsoft Purview sensitivity labels are applied to at least 80% of documents in scope. Without labels, IQ cannot enforce data classification. (0/1/2)
3. A named data owner exists for every system IQ will index. Orphan data is the #1 source of permission leaks. (0/1/2)

Data hygiene:

4. SharePoint and OneDrive permission inheritance is audited. Broken inheritance is the #2 source of cross-tenant data exposure in early IQ deployments. (0/1/2)
5. OneLake (for Fabric IQ) has a semantic model defined with measure definitions, business glossary, and clear data ownership. (0/1/2)

Cost and consumption controls:

6. Copilot Credit spend caps configured at tenant, department, and user tiers. The cost management dashboard exists for a reason — use it on day one. (0/1/2)
7. Charge-back model defined. Decide whether IQ usage rolls up to IT, to business units, or splits. Defer this decision and you will fight about it later. (0/1/2)

Adoption and change management:

8. Executive sponsor named at the C-suite level for the first use case. No exec sponsor, no scale. (0/1/2)
9. End-user communication plan addresses what data agents can and cannot see. The first time an employee feels surveilled, the pilot dies. (0/1/2)

Measurement:

10. Baseline measurement for the target workflow exists. Hours per ticket, time-to-answer, escalation rate — pick a metric and capture it before deployment. (0/1/2)

Scoring guidance: Organizations scoring 16-20 typically reach measurable productivity gains within 6-8 weeks. Organizations scoring 10-15 should pilot in a single department with weekly cost reviews. Organizations scoring below 10 should spend a quarter on the prep work before turning IQ on. The hardest items to fix are #2 (Purview labels) and #5 (Fabric semantic model). Start those before you sign the contract.

Case Study: How a Fortune 500 Financial Services Firm Avoids the Pilot Trap

A North American financial services firm with roughly 60,000 employees was an early access participant for Work IQ in Q1 2026. The pilot scope: HR self-service for U.S. benefits questions, replacing a Tier 1 support queue handling 14,000 tickets per quarter.

Phase 1 (Weeks 1-3) — Foundation. The firm spent the first three weeks not on the agent itself but on Purview sensitivity labels. Only 41% of HR documents had appropriate labels at the start. By end of week 3, that number was 92%. This was the gating step. The lesson: if you skip the data hygiene, the agent will eventually surface a document an employee should not see, and the pilot will be shut down by Legal within 48 hours.

Phase 2 (Weeks 4-6) — Pilot agent in Copilot Studio. A team of three (one IT architect, one HR business analyst, one prompt engineer) configured the agent against Work IQ's Foundry IQ engine for documents and Work IQ's Context API for employee-specific data (job grade, location, benefit eligibility). The initial agent answered 67% of test queries correctly; after iteration on prompt patterns and tool definitions, that climbed to 89% by end of week 6.

Phase 3 (Weeks 7-10) — Soft launch to 4,000 employees in one business unit. The firm captured baseline data first: average HR Tier 1 ticket resolution was 18 hours, cost per ticket was approximately $34, and CSAT was 3.6/5. After ten weeks of soft launch:

• Ticket volume in the pilot population dropped 47%.
• Average resolution time on the remaining tickets dropped to 11 hours (escalations got faster because Tier 1 staff weren't drowning).
• CSAT climbed to 4.1/5.
• Copilot Credit spend was $0.18 per agent interaction, against a baseline ticket cost of $34. Net cost reduction per resolved query: 99.5%.

Phase 4 (Weeks 11-14) — Full rollout to U.S. HR. Production scale brought new failure modes: a spike in queries about a recently changed 401(k) provider exposed gaps in the document corpus that had been indexed before the policy update. The fix was a weekly content-freshness audit on the top 50 source documents, automated through a Power Automate flow against Work IQ's freshness signals.

What worked: narrow scope, owner-aligned business case, data hygiene as a precondition rather than an afterthought, and a willingness to ship a 67%-correct agent and iterate.

What didn't: the team initially tried to instrument cost-per-resolved-ticket as the KPI, but executive sponsors wanted simpler numbers. Final dashboard metrics were ticket volume, average handle time, and CSAT. The CFO got cost detail separately on a monthly review.

Twelve-month projected savings: $1.7 million on Tier 1 HR support alone. Payback on the IQ-related implementation cost: 4.2 months.

What to Do About It

For CIOs: Stand up a Microsoft IQ working group this month. Charter it to inventory which of the four context engines map to your top three agent use cases, score those workloads against the readiness checklist, and pilot one engine in one business unit before the end of Q3. Insist on Purview label coverage as the gating criterion — it will dominate the timeline. Resist the urge to roll out all four engines at once; the operational complexity is not where IQ pays back.

For CFOs: Get the Copilot Credit consumption model in front of finance leadership before any pilot ships, not after. Set tenant-level spend caps and require monthly reviews for the first six months. Build the charge-back model now — IT will not want to absorb IQ costs once they exceed $1 million annually, and business units will not want to inherit costs they did not budget for. The June 1 GitHub Copilot usage-based billing transition is a free preview of what unmanaged consumption looks like; learn from it.

For business unit leaders: Pick one workflow with a clear baseline metric — ticket volume, response time, sales productivity, deal cycle — and propose it as the IQ pilot. Vague use cases ("improve productivity") will not survive the first executive review. Specific use cases ("reduce HR Tier 1 ticket volume 30% in U.S. benefits queries") will. The case study above is a template — adapt the structure to your function.

For governance and security leaders: Use IQ's GA as the forcing function for cleaning up Purview labels, SharePoint permission inheritance, and Entra group hygiene. These items have been on every CISO's to-do list for years; the difference now is that AI agents will surface the gaps publicly. Better to find them in a pilot than in production.

Most enterprises will not adopt all of Microsoft IQ. Many will not adopt any of it — Salesforce, ServiceNow, SAP, and Glean all have legitimate alternatives for specific workloads. But every enterprise IT leader should have a written position by end of Q3 on whether IQ is the default context layer for their AI agents, or whether they will deliberately go elsewhere. The vendors who win the next two years are the ones who solved the data silo problem before the agents arrived. Microsoft just shipped the most aggressive answer yet.

---

Continue Reading

• Microsoft's 7 MAI Models: End of Single-Vendor AI Era
• Microsoft Scout Autopilot: Always-On Agent at $15 with Entra Identity
• AI Agent Access Control: Noma's MCP Lens on 65% of Breaches
• MCP Hits 97M Downloads: The AI Integration Standard Has Arrived
• Azure Agent Mesh + Windows Agent Framework: $15M TCO Story