Treat Your AI Agents Like New Hires

By Siddarth Kotha

Most companies have spent years getting identity right for people. Every employee gets an account on day one, access that fits their job, and a manager who answers for it. When they change roles, their access changes. When they leave, it is turned off.

AI agents are now doing real work inside those same companies. They read email, open tickets, query databases, and call other systems. Many of them do it with far less oversight than we would ever accept for a new hire. Ask most security teams how many agents they have, who owns each one, and what each one can reach, and you will usually get a long pause.

The good news is that we already know how to fix this. We just need to give agents the same identity basics we give people.

Why agents raise the stakes:

An agent can act thousands of times an hour, and it does what its instructions and inputs tell it to do. That includes instructions hidden in an email, a document, or a web page it was asked to read. A person might stop and ask whether a request makes sense. An agent usually will not.

That is why identity matters so much here. When an agent is tricked or simply gets something wrong, the access it holds decides how much damage it can do.

Five basics that carry over from people to agents:

1. Give every agent its own identity. No shared accounts and no borrowed human logins. When an agent acts, the logs should show that agent, not a person or a generic service account.

2. Name a human owner. Every agent needs one person who answers for what it does, the same way a manager answers for an employee's access. An agent with no owner should not be running.

3. Grant only what the task needs. Scope access to the job in front of it, not to everything it might use someday. Prefer short-lived credentials that expire on their own over keys that sit in a config file for years.

4. Keep a person in the loop for high-impact actions. Let agents draft, sort, and recommend freely. Moving money, changing access, deleting data, or touching production should still need a human yes.

5. Watch it, review it, and know how to turn it off. Log what each agent does, review its access on a schedule, and have a tested way to shut it down quickly. When an agent is no longer needed, retire it, just as you would close an account when someone leaves.

Where to start:

You do not need a brand new program to begin. Start with an inventory and find the agents already running, including the ones business teams built on their own. Then put each one through the same joiner, mover, and leaver process you already use for people. Give it an identity, assign an owner, review what it can reach, and set a date to check it again.

AI will keep moving faster than most policies can. Identity is the one control that goes wherever the agent goes, which makes it the right place to start.

‍

Join the conversation. Connect with the community. Help shape what's next.

Siddarth Kotha is a member of the Kodi Connect Advisory Board and leads identity and access management for a global manufacturer.

The conversations shaping the future of AI don't stop here. At Kodi Connect, we bring senior technology, cybersecurity, data, and AI leaders together to exchange ideas, challenge perspectives, and explore what's next. Join us at an upcoming Kodi Connect executive networking event to take part in similar conversations with fellow industry leaders. Our Advisory Board is an extension of our in-person networking community, bringing together experienced leaders like Siddarth to share insights, influence important conversations, and keep the exchange of ideas going long after the event ends.