Microsoft is preparing to move another important piece of the Copilot Studio experience into usage-based billing.
Starting September 1, 2026, the grace period for existing agents and workflows created using the GitHub Copilot harness in Copilot Studio will come to an end. From that date, these agents and workflows will begin consuming Copilot Credits under Copilot Studio’s usage-based billing model.
For organizations experimenting with agentic AI, this is an important change to understand before the billing date arrives. While Microsoft says that no immediate action is required, administrators may want to review historical usage, identify active GitHub Copilot harness agents and workflows, and make sure their environments have appropriate cost controls in place.
The change is particularly relevant for teams that have been testing or building GitHub Copilot harness experiences during the current grace period, because usage that was previously non-billed will transition into credit consumption.
What Exactly Is Changing?
The change applies to GitHub Copilot harness agents and workflows created before August 3, 2026.
That includes agents and workflows created in Dev and Trial environments. Beginning September 1, their usage will count toward Copilot Credit consumption.
There are two sides to this consumption model.
First, maker AI authoring activities will consume Copilot Credits. This covers AI-powered activities involved in creating or working with the agent during the authoring experience.
Second, runtime execution will also consume Copilot Credits. In other words, the cost consideration isn’t limited to the person building an agent. Actual usage of the resulting agent or workflow can also contribute to credit consumption.
This distinction matters for organizations because an agent that appears inexpensive during development could generate significantly more consumption once it is actively used by employees, customers, or automated processes.
Which Agents Are Affected?
The policy specifically targets agents and workflows built using the GitHub Copilot harness.
However, not every Copilot Studio agent is affected by this particular billing change.
Microsoft says that standard harness agents and Copilot Chat harness agents are not affected by the September 1 transition described in this announcement.
That makes it important for administrators to understand which type of harness their existing agents use rather than assuming that every Copilot Studio agent will suddenly be billed under the same rule.
For organizations with a large number of agents, this distinction could become especially important when reviewing usage and planning future deployments.
The Grace Period Is Ending
The September 1 date essentially marks the end of a transition period.
Existing GitHub Copilot harness agents and workflows have been operating during a grace period in which their Copilot Credit usage was not billed. Once that period ends, the same workloads will begin consuming credits.
This doesn’t necessarily mean organizations need to rebuild or disable their existing agents.
In fact, Microsoft describes the announcement primarily as an awareness message, stating that no action is required.
The more practical takeaway is that organizations should use the remaining time to understand what they already have running.
If an organization has forgotten about an experimental agent created several months ago, September 1 could turn that previously invisible usage into measurable Copilot Credit consumption.
How Administrators Can Review Usage
One of the most useful resources available to administrators is the Power Platform admin center (PPAC).
Historical, non-billed Copilot Credit consumption data is available there. This information can give administrators a better understanding of how their environments have been using Copilot capabilities.
To find the relevant information, administrators can navigate to:
Licensing → Copilot Studio → Manage Agents
The historical information can help answer some practical questions.
Which agents have been active?
Which environments are generating the most activity?
Are development or trial environments still running agents that nobody uses anymore?
Are there agents that could see substantially higher usage once they move into a billed model?
These are the types of questions administrators should consider before September 1.
There is an important caveat, however: Microsoft describes the historical data as directional only. It should not be treated as a billing estimate or a guarantee of future costs.
Actual consumption after billing begins can vary depending on how agents are authored and used.
Why Historical Usage Matters
For many organizations, the biggest challenge with AI billing isn’t simply understanding the price of an individual operation. It’s predicting how frequently those operations will occur.
An agent used by a small development team may have relatively limited activity. The same agent, if deployed across a large organization, could have dramatically different consumption.
That’s why historical usage can be a useful starting point.
Administrators can look at previous activity and identify the agents that deserve closer attention. They can then combine that information with expected future usage to establish internal controls and budgets.
The goal isn’t necessarily to eliminate Copilot Credit consumption. Instead, organizations should make sure that consumption is intentional and aligned with business value.
Managing Copilot Credit Consumption
Microsoft recommends several ways administrators can manage GitHub Copilot harness consumption.
One approach is to allocate Copilot Credits to specific environments. Environment-level allocation can help organizations establish clearer boundaries around where credits are available and how resources are being consumed.
Administrators can also control access to the shared tenant credit pool. This can provide another layer of governance for organizations that want to prevent unrestricted consumption.
Monitoring is another key part of the process. The Power Platform admin center provides existing reporting and capacity management experiences that administrators can use to keep track of consumption.
For organizations using pay-as-you-go, Microsoft also recommends using Azure budgets and alerts to help monitor spending.
These controls become increasingly important as organizations move from AI experimentation toward production-scale agent deployments.
What Should Organizations Do Before September 1?
Although no mandatory action is required, a short review before the billing transition would be sensible.
Administrators can begin by identifying all GitHub Copilot harness agents and workflows that exist across the organization.
Next, they can review historical Copilot Credit consumption in PPAC and identify workloads with meaningful activity.
It is also worth checking Dev and Trial environments. These environments can sometimes contain experiments that were created for evaluation and later forgotten. If those agents remain active, their usage could become relevant after the grace period ends.
Organizations should also consider whether their current credit allocation and cost-management policies match their expected AI usage.
Finally, teams should make sure the people responsible for Copilot Studio understand that September 1 is a billing milestone—not necessarily a product shutdown or a requirement to migrate every existing agent.
A Bigger Shift Toward AI Cost Governance
The change reflects a broader trend in enterprise AI.
As organizations move from simply experimenting with generative AI to deploying autonomous and semi-autonomous agents, AI usage becomes an operational cost that needs governance.
Agentic systems can perform tasks repeatedly, interact with users, invoke tools, and execute workflows. That creates significant opportunities for automation, but it also means organizations need visibility into consumption.
For IT leaders, the question is increasingly moving from “Can we build an AI agent?” to “How do we operate AI agents responsibly and cost-effectively at scale?”
Copilot Credits and the Power Platform administration experience are part of that transition.
The Bottom Line
The key date is September 1, 2026.
From that day, existing GitHub Copilot harness agents and workflows created before August 3, 2026 will begin consuming Copilot Credits under the usage-based billing model. Both maker AI authoring activities and runtime execution are included.
Standard harness and Copilot Chat harness agents are not affected by this specific change.
There is no required action from Microsoft for customers simply because of this announcement. Nevertheless, administrators have a good opportunity to review their environments before the billing transition takes effect.
Checking historical Copilot Credit usage, identifying active GitHub Copilot harness agents, reviewing environment allocations, and establishing appropriate monitoring and budget controls can help organizations avoid surprises.
The September 1 change is therefore less about an urgent migration and more about visibility, governance and preparation.
For organizations rapidly expanding their use of AI agents, understanding consumption before production usage grows will be just as important as building the agents themselves.
