The birth of AMP, the incredible analogies with IAM, and the rise of HAM

On October 27th, a new and interesting Gartner Magic Quadrant saw the light: the “Magic Quadrant for AI Agent Management Platforms” (AMP).

Gartner defines AI agent management platforms “as centralized software platforms that manage AI agents independent of the execution environment. They unify security, governance, and FinOps controls across custom and marketplace-acquired agents. As enterprise AI adoption scales, organizations will increasingly need these tools to manage agent deployments, ROI, and costs.”

Sound familiar to IAM people, innit? Replace “AI agents” with “humans”, “execution environment” with “applications”, “FinOps controls across… agents” with “license costs across SaaS”, and you’re there.

The analysts who built this MQ are not the IAM ones, though. Two different groups. Who’s leading this new AMP MQ? The usual suspects: Microsoft, IBM and ServiceNow.

The MQ is well done, and I honestly learned a lot. Still, I could not find any mention of what I think is one of the overarching questions:

“How do you detect security and risk situations when agents operate on behalf of humans?”

Agents are not risky by themselves. They become risky depending on the human(s) using them, as agents inherit the human’s permissions.

And here’s the tricky part: agent governance (at run time or at audit time) needs to know who that human is (job title, job description, the org unit he/she is in) and what he/she can do (permissions on IT systems).

This whole data gold mine sits in good ol’ IGA platforms.

There’s just one way forward, then: correlate these two worlds, the old human IGA one and the new agentic one. That’s what we do at IAMONES. We believe the real risk is in personal agents, consumed via the employee workplace. No rocket science, right? Yep, no rocket science. Just pragmatism and common sense.

Alright then. You might now ask: “I’m running my good IGA platform. No plan to migrate to the cloud. Why can’t I simply stuff all these agent identities and permissions into my good ol’ IGA?”

You can’t, for many different reasons, but here’s the key one:

  • Agents are ‘programmed’ and ‘behave’ according to plain-language prompts (call them ‘skills’, ‘instructions’ or ‘intent’).
  • A traditional IGA platform (any of those you know of) thinks ‘structured’: roles/rules/ABAC/RBAC/PBAC and the whole shebang of *AC we’ve seen over the last 10 years.
Only an AI-native reasoning layer can correlate these two worlds. You can build one or buy one (such as ours, or one of the many others on the market now), but you won’t escape this requirement.

Maybe one day Gartner will publish a combined MQ? Can I come up with a funny name?

Let’s call it HAM: ‘Human Agent Management’.

My 2 cents from a Gen-X Identity guy.

Ask anything. Configure everything. In your own words.

Request a demo