Building the system for AI at work adds a second bill
Your Copilot invoice used to be predictable: a flat per-seat fee, easy to forecast, easy to defend to the CFO. Microsoft's September 23, 2026 announcement, Building the system for AI at work, changes that math for anyone scaling agentic AI.
We think flat per-seat pricing never matched how unevenly teams actually use AI, so a second cost layer isn't shocking on its own. What matters is whether you can see it coming before the invoice does.
In that same September 23 post, Microsoft says advanced agentic work will run on usage-based billing, tied to the work an agent completes, stacked on top of the existing per-user subscription.
That's what changed in Microsoft AI at work on September 23: Copilot now runs two meters, not one.
Sources: Building the system for AI at work | AI at Work Blog
Model choice replaces the single-vendor platform
The same announcement reframes the platform itself. Microsoft says its AI platform now gives customers access to models from multiple providers, including Anthropic and OpenAI, rather than running on one company's model alone, a core piece of this Microsoft AI at work announcement.
Judson Althoff, Microsoft's Chief Executive Officer of Commercial Business, frames that choice as a stance against lock-in, in his own post announcing Microsoft Frontier Company.
“Customers shouldn’t be locked into a single model any more than they should be locked into a single technology vendor.”
— Judson Althoff, Chief Executive Officer, Microsoft Commercial Business · Source: Microsoft Frontier Company: AI engineering that amplifies and protects your intelligence - The Official Microsoft Blog
We're skeptical that 'model choice' alone moves the needle for a buyer until Microsoft states which products and editions actually expose it; the September 23 post doesn't spell that out.
Multi-model access is a platform decision, not a usage guarantee.
Sources: Building the system for AI at work | AI at Work Blog
Who this hits first
Microsoft also says it is embedding Forward Deployed Engineers directly with customer teams, as part of how it wants to support this new agentic push.
So when it comes to Microsoft AI at work, who is affected first? On a straight read, it's IT admins and finance teams managing budgets, not end users directly, since they're the ones who must set spending policies and model access before anyone else notices a change.
Separately, Microsoft's July 2, 2026 announcement of Microsoft Frontier Company describes a $2.5 billion investment. The September 23 post doesn't explain how that program connects to the Forward Deployed Engineers it mentions. That figure is Microsoft's own account of its program, not an independently verified count.
- IT and platform admins who manage billing and spending policies
- Finance and procurement teams tracking a new variable cost line
- Teams building or operating agents in Copilot Studio
- Leaders deciding how fast to scale agentic work
We'd push back on reading embedded engineers as a sign your own organization is 'AI mature' - that's Microsoft's go-to-market motion, not evidence of how your people actually use AI.
Sources: Building the system for AI at work | AI at Work Blog, Evolution of the Copilot pricing model | Microsoft Community Hub, What's new in Agent 365 - September 2026 | Microsoft Community Hub, Microsoft Frontier Company: AI engineering that amplifies and protects your intelligence - The Official Microsoft Blog
What to check before you flip on usage-based billing
A companion pricing post Microsoft published September 25, 2026 says usage-based billing stays off for enterprise customers until an admin creates a spending policy in the Microsoft 365 admin center.
A separate September 30, 2026 product update says cost-management support for agents built in Copilot Studio is planned for October 2026, so that visibility isn't fully in place yet.
Before flipping that switch, work through four checks first:
Do that mapping before the bill does it for you.
Sources: Evolution of the Copilot pricing model | Microsoft Community Hub, What's new in Agent 365 - September 2026 | Microsoft Community Hub
What the announcement doesn't say
Building the system for AI at work stays quiet on several questions a leader would ask. It doesn't state which editions, countries, or hosted and government plans, such as Microsoft 365 G7, are covered.
- Which editions and regions carry the new billing model
- Whether hosted or government plans are included
- How Forward Deployed Engineers relate to the separate Frontier Company program
- The exact usage-based rates, which sit in a separate pricing post
Ask your account team directly which of your editions and regions this actually touches; the announcement leaves that open rather than hinting at an answer.
Don't mistake this rollout for proof of adoption
None of what Building the system for AI at work changes tells you whether your own employees are using AI well, or using it at all. A platform and pricing shift from Microsoft is a vendor decision, not a measurement of your workforce.
If you want that picture, you're looking at what employees report about their own usage, skills and confidence, not an independently measured productivity number. That's the honest limit of any adoption survey, including AI Adoption Score, which shows how your teams actually use AI and where the gaps sit.
Our bet is that the organizations getting real value from Microsoft's rollout are the ones who already know where their people stand on AI skills and trust, not just where the invoice lands.
Fix your billing governance because of this announcement. Measure your own people's adoption separately, because Microsoft's rollout can't tell you that.
Frequently asked questions
Who should own this decision inside my organization?
A CIO or Head of AI typically owns platform and model choices, while finance owns spending caps and alerts. A separate owner should track whether employees actually adopt what's rolled out, keeping vendor rollout decisions distinct from internal adoption tracking.
Should I negotiate terms before turning on usage-based billing?
The announcement doesn't address this directly. As a practical step, loop in procurement before you enable usage-based billing: credit consumption rates and model access are variables worth clarifying with your account team ahead of time, not after your first invoice arrives.
Are Forward Deployed Engineers an extra cost?
The announcement doesn't state that either way. It mentions embedding Forward Deployed Engineers with customer teams as part of its partnership approach, but gives no detail on cost, headcount, or who qualifies for the engagement.
Should I pause agentic rollouts until I understand the new billing model?
The announcement doesn't call for a pause, and neither does this article. The practical move is to set spending policies and ownership first, then scale deliberately rather than stopping rollouts entirely or moving ahead without billing guardrails in place.
Sources: Building the system for AI at work | AI at Work Blog
