The Work IQ API reached general availability on June 16, 2026. It carries no separate subscription and no per-user licence; consumption is billed through Copilot Credits. Administrators must enable pay-as-you-go billing in the Microsoft 365 admin center for any tenant where custom agents call Work IQ. Without that configuration, the agents do not bill — they fail.
This is a small operational requirement with an outsized capacity to break a project at exactly the wrong moment. Development and testing frequently occur in a tenant where billing was configured once by someone who has since moved on, and the production tenant is a different story entirely. The failure appears at deployment, not during build.
What Work IQ is for
Work IQ exposes organizational work context — the signals about how people, content and activity relate inside a tenant — to custom agents. An agent that can reason over that context produces materially better output than one operating on a prompt alone. In practical terms, it is the difference between an agent that can answer "who owns this project and what changed last week" and one that can only answer from what the user pastes into it.
For organizations building agents in Copilot Studio or through the API surface directly, Work IQ is likely to become a default dependency rather than an optional enhancement, which makes the billing prerequisite a standing requirement rather than a one-off.
The three-step check
First, confirm pay-as-you-go billing is enabled in the Microsoft 365 admin center for every tenant where agents will run, including non-production tenants used for testing. This requires an Azure subscription to bill against, and the linkage between the Microsoft 365 tenant and that subscription is the part most commonly missing.
Second, confirm the Azure subscription behind it is one that will not be decommissioned, has an owner who understands what it is now funding, and carries a budget with alerts. A consumption dependency billed to a subscription created for a proof of concept is a known pattern and a bad one.
Third, establish attribution. Copilot Credit consumption at tenant level tells you the total but not the cause. Instrumenting agents so consumption can be traced to a specific workflow is what allows a cost conversation to be about a workflow rather than about AI in general.
The wider pattern
Work IQ, Copilot Cowork and Copilot Studio agent capacity now all meter through Copilot Credits. That consolidation is genuinely helpful — one unit of account rather than three — but it also means a single pool of consumption is being drawn down by workloads owned by different teams. Without attribution, the finance conversation degrades into "AI cost went up" with no ability to identify which workload moved.
Organizations getting this right treat Copilot Credits the way a mature Azure practice treats compute spend: tagged, attributed, budgeted per workload, and reviewed on a cadence short enough to catch a runaway before it becomes a quarter's variance.
What to do before your next agent goes live
Write down the expected credit consumption for the workflow, based on a measured pilot rather than an estimate. Set a budget at a multiple of that figure that represents an acceptable overrun rather than a comfortable ceiling. Configure the alert. Name the owner. Then deploy. The whole exercise takes an afternoon and it is the difference between a controlled AI programme and an uncontrolled one.
How Lorexus engages
We configure the billing linkage and budget structure, instrument agents for per-workflow attribution, and run the measured pilot that produces a defensible consumption forecast before production deployment. Book a free 15-minute call with our senior engineers.