Discovery Phase: Creating an Inventory of Hidden AI Agents
Discovery Phase: Creating an Inventory of Hidden AI Agents
I asked a finance manager last week: “What is artificial intelligence doing in your company?”
The answer wasn’t clear. “I don’t know. We all use ChatGPT, Claude. There’s some system in Finance. Audit reports seem to go out automatically—I don’t know how. Someone finds anomalies in a database. No details.”
To map an organization, you need information. But where does it come from?
Traces: Budget, Users, Data Flows
To discover hidden AI agents, you start with observation. Three simple traces exist:
Budget Traces: Open your company’s cloud spending. Are there API calls, an OpenAI API account, or a payment to a SaaS tool? Every dollar spent points to a running process. If Finance’s manager says, “I don’t know why we spend this much every day,” then an automation is running.
User Traces: To find out how much ChatGPT, Claude, or Copilot each person uses, start by asking a manager. In your company’s identity provider (OIDC, SSO) logs, how many employees signed into an AI tool? If the number is high, it’s a signal.
Data Flow Traces: Look at your company’s databases, CRM, and storage systems. Check who last modified which tables. If a system automatically creates files or writes edits, it’s a sign of an agent.
Questions: Knowing What to Look For
Pick 20-30 employees in your company and ask them these questions:
- “When you do your job today, which of the tools you use have you filled with artificial intelligence? Which one?”
- “Which files are created automatically? Someone doesn’t write them by hand, and suddenly they’re done?”
- “Don’t you get a daily or weekly report, but you never asked anyone to make it? Who’s doing it?”
- “Does a tool you use in a request give you results? Does it read long text, create summaries, do things?”
The answers to these questions map out hidden AI use.
The Human Side: Define Roles
Once you find the hidden AI system, the questions change:
- Which department is responsible for this system?
- Who started it or maintains it?
- If there’s an error, who gets notified?
- Is this system tied to another one?
Every hidden system has a human owner. You’ll find them. Because when it’s no longer hidden, responsibility becomes clear.
Write Down Your First Inventory
Create a table. Four columns:
| System | Purpose | Owner | Risk Level |
|---|---|---|---|
| Finance - Report Automation | Create monthly PDF reports | ? | Yellow |
| Data Team - Anomaly Detection | Find inconsistencies in database | ? | Red |
| HR - CV Summary | Summarize job applications | HR Manager | Green |
The table won’t stay empty. By filling it in, you escape from secrecy.
Why Does This First Move Matter?
What if you don’t do it? The system grows. The owner forgets or leaves. Within hours it breaks. The company is harmed. No one is responsible.
The first move breaks this cycle.
About the Post: A hidden AI inventory is as dangerous as unwritten rules. With discovery, inquiry, and a simple table, you can map the institutionalized shadow AI in your organization. Organizational structure and responsibility emerge from this discovery table.