Governing the Agent Fleet
On June 9, KPMG and Microsoft announced that Agent 365 and Copilot would go to all 276,000 KPMG professionals across 138 countries. Every writeup led with the agents. They buried the part that actually matters.
The product is not the agent. It's the layer that registers, scopes, watches, and retires the agents. Agent 365 lists at $15 per user per month standalone, or bundled into Microsoft 365 E7 at $99. What used to be a six-figure consulting engagement is now a per-seat line item.
That repricing is the whole story.
Why a management plane, not an agent
One agent is a demo. A thousand agents is an operations problem. The moment you have a fleet, you need to answer three questions, continuously:
- What agents are running?
- What can each one actually touch?
- Who is watching them?
Agent 365 answers those by giving every agent an identity, enforcing least-privilege permissions, monitoring behavior, and managing the full lifecycle from deploy to retire. It treats an agent the way a mature IT org treats an employee: provisioned, scoped, logged, and eventually offboarded.
I have been saying this for a while
This is the thesis behind everything I'm building. The hard, valuable, durable part of enterprise AI is not building an agent. It's governing a fleet of them.
KPMG also plans to sell the governance stack to its clients as part of its Trusted AI framework. So a Big Four firm just validated that agent governance is both an internal necessity and a product line. That's the same bet I'm making with Aether Ops, from the other end of the market.
When the consultancies productize the thing you're building, you're either too late or right on time. Given that 91% of organizations are already running agents and only 10% have a real strategy for managing the non-human identities behind them, I'd argue it's the latter. The fleet is already here. The control plane is the work.