Agent discovery: useful, and not the same as governance

Agent discovery is the process of finding the AI agents operating in an organization's environment: scanning platforms, cloud accounts, code repositories, SaaS integrations, and network traffic to identify agents, the credentials they hold, what they connect to, and sometimes who created them. It is the agent-era version of a pattern the security industry has run many times: directory cleanup, privileged account discovery, shadow IT scans, secrets scanning.

The pitch is always some version of the same sentence: you cannot govern what you cannot see. That statement is true. It is also one of the industry's most expensive half-truths, because it assumes governance begins only after something has been discovered and registered.

What discovery gives you

Run an agent discovery tool and you will get, at best: that an agent exists, what it runs on, which credentials it holds, what scopes those carry, sometimes who created it, sometimes how it connects to other agents, sometimes what it has been calling.

This is useful. It tells you how large the surface is, where credentials are leaking from, and which agents should be shut down rather than governed. It gives the NHI and access programs something to work on. It should continue.

Why the inventory keeps coming first

The inventory comes first not because anyone believes it is sufficient, but because it produces a deliverable. You run the scan, you get a number, you put the number on a slide, and the number goes down over the next two quarters. Everyone can point at progress. Authority is harder to count.

What discovery cannot tell you

For any agent it finds, discovery cannot tell you what the agent is allowed to decide, on whose authority, what the limit of that authority is across everything it does today, what happens when it reaches the limit, or who answers when it decides wrongly. None of those are about existence or access. All of them are about authority. The discovery layer does not have a field for them because it is answering a different question.

Why discovery-first governance has a ceiling

Our position is the following. If governance is built on top of discovery, then governance only reaches the agents that discovery found. You cannot register what you have not found, you cannot assign a mandate to what you have not registered, and in an environment where agents create other agents, the gap between what exists and what has been catalogued widens rather than closes.

Agents are spawned by workflows and by other agents. They are created and destroyed inside a single task. Any inventory is out of date the moment it is written. A control that depends on a current inventory fails on precisely the agent you did not know about, which is the agent most likely to be the problem.

The action still happens whether the actor was registered or not. So the control has to sit where the action happens, and it has to evaluate the decision on its own terms, regardless of whether anyone has heard of the thing making it.

Govern the action, not the inventory

The alternative is to place enforcement on the surfaces where actions have to pass: the tool call, the API tier, a gateway in front of the system of record, a sidecar beside the workload. An action from an unregistered agent is evaluated because of where the check sits, not because the agent was known. Coverage follows the surfaces you put enforcement in front of, rather than the completeness of your inventory.

Under this model, discovery is an input. It helps you decide which surfaces to protect first and which agents to retire. It is not the perimeter, and it is not the precondition for enforcement.

Taking attendance

An organization that finishes its agent inventory, builds the dashboard, and reports that its agents are governed has done something closer to taking attendance. It knows who showed up. It does not know what any of them are permitted to decide, or who answers when they decide badly. The gap between those two states is where the next serious incident will happen, and the inventory will have looked clean the entire time.

Where BotAris fits

BotAris does not discover agents. It governs actions. The Decision Authority Enforcement Point sits in the action path and evaluates every consequential action against the Authority Model, whether the agent behind it was registered last quarter or came into existence four seconds ago. Discovery tools and BotAris work on different problems, and organizations generally need both.