By Microsoft’s Work IQ APIs reach general availability on June 16 — three days from now — and with GA comes consumption-based billing via Copilot Credits. If you’ve been building enterprise agents on top of Microsoft 365 data during the free preview period, the meter starts Monday. Here’s what you need to know before it does.
Work IQ Is Not Microsoft Graph
The most important clarification upfront: Work IQ is not a replacement for Microsoft Graph. It’s a layer on top of it. Graph is an access layer — it returns raw data (emails, files, calendar events) and leaves interpretation to you. Work IQ is an intelligence layer. The difference shows up clearly when you compare the questions each can answer.
Graph can tell you which emails exist in a mailbox. Work IQ can answer “what did we decide about the Q3 budget?” — reasoning across emails, meetings, Teams chats, and files simultaneously. That’s a different class of problem entirely.
Under the hood, Work IQ runs across three integrated layers: a Data layer that unifies signals across M365, a Memory layer that maintains persistent cross-session understanding, and an Inference layer that enables agents to reason and act. Ten generic tools collapse what would otherwise be hundreds of individual Graph operations. Microsoft claims 80% fewer tokens consumed versus conventional M365 API approaches, and 54% better recall than single-shot RAG.
A2A, REST, or MCP: Which One Do You Need?
Work IQ exposes three integration patterns at GA, each suited to a different orchestration model:
- A2A (Agent-to-Agent): For orchestration scenarios where you want to delegate M365-specific subtasks to a dedicated Work IQ agent. Returns Copilot-quality responses with citations. Best when your primary agent is not M365-native.
- REST API: Standard server-side integration for workflows that do not follow MCP conventions.
- MCP (Model Context Protocol): The most direct path if you’re working with Claude, Cursor, or VS Code-based agents. Microsoft redesigned the remote MCP server specifically for this GA release.
If you’re approaching from Azure AI Foundry, you’ll likely interact with Work IQ through Foundry IQ — which aggregates Work IQ alongside Azure SQL, uploaded documents, and web search behind a single endpoint. Most serious enterprise builders standardize here rather than connecting to Work IQ directly.
Billing Starts Monday: Model the Costs First
Work IQ uses Copilot Credits — consumption-based, billed through your M365 tenant. Microsoft’s scenario estimates: $0.20–$0.40 per light agent interaction, $0.30–$0.75 for medium-complexity tasks, $0.50–$1.50 for heavy workflows. Per tool call, you’re looking at roughly 0.1 Copilot Credits (about $0.001). That sounds small until agentic loops start stacking calls.
The key operational detail: billing is off by default. An admin must explicitly enable it in the Microsoft 365 Admin Center before your agents incur any charges. Microsoft is also releasing a cost management dashboard mid-June for spending limits and per-user/group credit monitoring. Set those limits before enabling billing — not after you see the first invoice.
One constraint: a Microsoft 365 Copilot license is required. Work IQ is not available as a standalone purchase.
The Governance Layer Is the Point
Work IQ ships with a Rego-based policy engine that evaluates every request against resource paths, methods, user identity, and content. Agents cannot access data the requesting user cannot access — there is no privilege escalation path. Full tool-call audit trails surface in Microsoft Defender via Advanced Hunting queries.
This is worth taking seriously. Building equivalent governance from scratch — permission scoping, audit logging, rate limiting, payload inspection — is months of work for most teams. Work IQ includes all of it by default. The governance layer is not an obstacle to building with Work IQ. It’s the reason you’d choose Work IQ over a raw Graph-plus-custom-RAG stack.
Three Things to Do Before June 16
If you’re already in preview, three steps before Monday:
- Confirm your protocol. A2A, REST, or MCP — match the pattern to how your agent is orchestrated, not just what’s easiest to implement first.
- Model costs against real query volumes. Run a test workload in preview, use Admin Center usage data, and project what your interaction patterns cost in Copilot Credits before billing activates.
- Set spending limits before enabling billing. Microsoft’s cost management controls in M365 Admin Center let you cap spending per user and group. The public preview repo at github.com/microsoft/work-iq has PowerShell scripts to automate tenant setup.
Work IQ is one piece of a broader Microsoft IQ platform announced at Build 2026 that also includes Foundry IQ, Fabric IQ, and Web IQ. The June 16 GA is the production milestone for the piece that handles organizational context. For enterprise agent builders, that’s the piece that changes what’s possible.
