ServiceNow built an AI Control Tower. UiPath has 950 customers orchestrating 365,000 processes through Maestro. The management layer for AI agents is shipping from every major vendor. The authorization layer is not.
ServiceNow just launched what it calls the AI Control Tower — a centralized command center to govern, manage, secure, and realize value from any AI agent, model, and workflow on a single unified platform. UiPath's CEO says 950 of its customers are developing AI agents that orchestrate 365,000 processes through its unified Maestro platform. Trace, a Y Combinator-backed startup, raised three million dollars to build knowledge graphs that decompose tasks and route them to AI agents or human workers. Microsoft is reportedly planning to bundle Agent 365 — centralized agent management, discovery, administration, and telemetry — into an E7 license at roughly $99 per user per month.
The management stack is filling in. Fast.
The Dashboard Without a Door
Every one of these platforms solves the same problem: visibility. Which agents exist, what they are doing, how well they are performing, which tasks they are assigned to. ServiceNow calls it a control tower. UiPath calls it an orchestration platform. Trace calls it a knowledge graph. Microsoft calls it centralized administration. The vocabulary varies. The capability converges. You get a dashboard. You get telemetry. You get the ability to watch.
What you do not get — in any of these platforms — is verification that a specific human approved a specific action. Not presence verification. Not biometric binding. Not cryptographic proof of intent. The control tower has monitors but no doors. It can tell you an agent accessed a database, drafted a financial filing, sent an email to a client. It cannot tell you whether anyone authorized those actions — because none of these platforms ask.
The Proliferation Math
UiPath's 950 customers are orchestrating 365,000 processes. Basis AI — an agentic accounting startup — just reached a $1.15 billion valuation automating financial statements and tax filings for seven of the top 25 U.S. accounting firms. ServiceNow's AI Platform lets thousands of agents coordinate actions across enterprise workflows. The rate of agent deployment is no longer linear. It is multiplicative. Each new platform multiplies the number of contexts in which agents act autonomously.
And each new platform deploys agents into those contexts through the same mechanism: a permission check at provisioning time. A role assignment. An API key. One-time authorization for ongoing activity. The assumption — that what was authorized at deployment remains authorized at execution, that the scope does not change, that the agent's behavior at hour one resembles its behavior at hour one thousand — has a name in security literature. It is called the confused deputy problem. The deputy was authorized once, to do one thing. The authorization persists while the thing the deputy does changes.
The Governance Mirage
What makes the current wave different from simple agent deployment is the word 'governance.' ServiceNow's documentation describes the AI Control Tower as a tool for governing agents. UiPath's platform references 'agentic automation security and governance.' These companies are explicitly positioning their products as the answer to the question: how do you control AI agents in production?
The answer they provide: observe, orchestrate, manage, report. Telemetry, dashboards, audit logs, workflow routing.
This is governance the way a traffic camera is governance. It records what happens. It does not prevent what should not happen.
The distinction matters because governance implies control, and control implies authorization. But the control offered is descriptive — what agents did — not prescriptive — what agents may do. The dashboard shows the agent sent the email. It does not show that the human responsible for that email approved its content, timing, and recipient with any form of verified intent. It cannot. The platform was not designed for that question. The platform was designed for the question before it: which agent did what.
Eighty percent of surveyed organizations reported risky agent behaviors including unauthorized system access and improper data exposure. Only 21 percent of executives reported complete visibility into agent permissions, tool usage, or data access patterns. The control tower is being built to address the visibility gap. The authorization gap is a different problem, at a different layer, and the dashboards do not reach it.
The Stack That Keeps Filling
The agent infrastructure stack now has five active layers. The perimeter layer — what agents can access — where Palo Alto Networks spent $400 million acquiring Koi for agent endpoint security and completed its CyberArk acquisition for identity and privilege governance. The identity layer — who this agent is — where Microsoft Entra, CyberArk, and Veza compete. The orchestration layer — how agents coordinate — where Trace, ServiceNow Agent Fabric, and UiPath Maestro operate. The financial trust layer — can this agent transact — where t54 Labs, Visa, and Google build payment rails. The management layer — how you watch what agents do — where ServiceNow AI Control Tower, Microsoft Agent 365, and UiPath dashboards live.
Five layers, billions of dollars deployed, dozens of vendors competing. Each layer is real infrastructure solving a real problem. And each layer implicitly assumes that some other layer handles verified human authorization — the specific, per-action confirmation that a known human approved a known action with biometric or cryptographic certainty.
None of them do. The assumption that someone else handles it is the architecture.
Every management tool deployed into this gap — every new control tower, every new orchestration platform, every new governance dashboard — widens it by one more system where agents act with observed-but-not-authorized autonomy. The stack fills in. The hole persists. And the confidence that the hole has been addressed — because the dashboard exists, because the governance label was applied, because the control tower is watching — grows with each deployment.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)