Microsoft Work IQ GA — The Billing Switch Nobody Budgeted For

Microsoft flipped the billing switch on Work IQ today — June 16, 2026 — and every agent built against the preview endpoints now draws from your Copilot Credits balance. The free preview is dead. The meter is live. And Microsoft still has not published a per-credit rate card specific to Work IQ consumption. If you are running production agents against M365 data, this is the single most consequential billing event since the GitHub Copilot credits debacle two weeks ago.

TL;DR

• What: Work IQ APIs (A2A, MCP server, REST) exited preview and entered GA on June 16 — billing started simultaneously
• Pricing: No separate SKU or subscription — consumption uses shared Copilot Credits balance with no published Work IQ-specific rate card
• Controls: Per-tenant, per-group, and per-user spend caps available from day one — but admins must explicitly enable them
• Action: Set a hard spend cap in the M365 admin center before any production agent touches Work IQ

Microsoft Work IQ GA — What Happened

Work IQ is not Graph API v2. Graph returns raw data — emails, calendar events, files, chat messages. Work IQ sits above Graph as a reasoning layer that can synthesize answers across multiple M365 data sources simultaneously. Ask it “what did we decide about the Q3 budget?” and it pulls context from email threads, meeting transcripts, Teams conversations, and shared documents to construct a coherent answer. That is a fundamentally different capability than querying individual endpoints and stitching results together yourself.

At GA, Work IQ ships three integration patterns. The A2A protocol lets you delegate M365 subtasks to a Work IQ sub-agent — your orchestrator hands off a natural language request and gets structured intelligence back. The redesigned remote MCP server (via @microsoft/workiq npm package) means MCP-native tools like Claude Code or Cursor can point directly at M365 organizational data. And the REST API covers conventional server-side workflows for teams that are not on the agent train yet.

The admin controls are actually better than what GitHub shipped with its credits transition. The Microsoft 365 admin center includes a cost management dashboard where IT admins can review AI credit usage, configure billing, and set spending limits at three levels: tenant-wide, per-group, and per-user. Critically, an administrator must explicitly enable consumptive billing before any Work IQ API call succeeds. This is a meaningful difference from the GitHub Copilot rollout, which delivered billing surprises before adequate controls were in place.

Admins who enabled Work IQ preview access must now explicitly configure billing controls in the M365 admin center. Preview agents that were running for free will fail on API calls until consumptive billing is activated — but once activated, they will start consuming credits immediately with no grace period.

Why This Matters

The billing architecture is the story here, not the technical capability. Work IQ does not have its own subscription, SKU, or per-user license. Every API call draws from the same shared Copilot Credits balance that Copilot Studio uses. The pricing model splits into two components: a fixed per-invocation cost for Tools calls and a variable cost for Chat and Context calls. Microsoft references general Copilot Credits pricing at $0.01 per credit, but the specific credit consumption rates for Work IQ operations — how many credits a Chat call burns versus a Tools call versus a Context call — remain unpublished.

This is the same consumption-based model that burned GitHub Copilot subscribers on June 1. Developers reported that routine prompts and agentic workflows chewed through monthly credit allotments far faster than expected. Microsoft clearly learned something from that backlash — the day-one admin controls for Work IQ are substantially better — but the core problem persists: you cannot budget what you cannot predict, and you cannot predict consumption without published rates.

The comparison to the Microsoft 365 agent control plane is instructive. Microsoft has been building enterprise governance infrastructure for agents: per-tenant controls, approval workflows, usage dashboards. Work IQ’s admin tooling fits that pattern. But governance without pricing transparency is a leash, not a strategy. An enterprise CTO can set a $5,000 monthly cap — great. But they have no way to estimate whether that cap supports 10 agents or 10,000, because the per-operation credit math is not public.

There is a timing problem too. Four days before Work IQ went GA, the U.S. government issued an export control directive suspending access to Anthropic’s Fable 5 and Mythos 5 models for foreign nationals. Work IQ itself is unaffected — it operates within Microsoft’s own model infrastructure. But many teams building on Work IQ route the intelligence it returns through third-party frontier models for further reasoning or generation. If your agent pipeline depends on model access that is suddenly subject to export controls, this is a bad week to commit to a hard production launch. The model layer beneath your agent stack is less stable than it was a week ago.

Before enabling Work IQ billing, run your preview agents through a 48-hour instrumented test with aggressive spend caps ($50-100). Log every API call type (Chat, Context, Tools) and measure actual credit consumption. This gives you real data for a budget conversation instead of guessing.

The real competitive question is whether Work IQ’s reasoning layer actually delivers on its efficiency promises. Microsoft positions it as dramatically more efficient than conventional approaches to M365 data retrieval — fewer tokens consumed, better recall than naive RAG. These claims are plausible given the architecture (a pre-indexed intelligence layer should outperform ad-hoc retrieval), but Microsoft has not published verifiable benchmarks in any public documentation or blog post. Until independent testing validates the efficiency claims, treat them as marketing.

What makes Work IQ genuinely interesting — and why I think teams should watch this closely despite the billing uncertainty — is the MCP server integration. The @microsoft/workiq npm package means any MCP-native development environment gets direct access to organizational intelligence. This is not “query your emails from the command line.” It is “ask your dev tool a question about what your team decided last week and get a grounded answer.” That capability did not exist in any usable form six months ago. The .NET MCP governance work Microsoft has been doing makes more sense now — they are building the plumbing for exactly this kind of tool-to-enterprise-data connection.

The Take

I would not launch a production agent on Work IQ this week. Not because the technology is bad — the three-pattern integration surface (A2A, MCP, REST) is well-designed, and the admin controls are genuinely better than what GitHub shipped. But consumption-based billing without a published rate card is an open checkbook, and we just watched that exact model blow up in developers’ faces with GitHub Copilot Credits two weeks ago.

Here is what I would do: enable Work IQ in a sandbox tenant with a $100 monthly cap. Instrument every API call. Run your actual agent workloads for two weeks and measure real credit consumption. Then you have data — not guesses — for the budget conversation with your CTO. Microsoft will eventually publish the rate card; they have to. But the teams that wait for the rate card before experimenting will be six months behind the teams that started measuring now.

The deeper signal is strategic. Microsoft is making M365 organizational data — the messy, unstructured, cross-application context that lives in every enterprise — accessible as agent infrastructure. That is a moat play. No other vendor has this data. Build your agents to be Work IQ-aware, but build your budgets to survive the billing opacity. Set the spend cap before you ship. Not after.