Airia: securing agents inside the enterprise

Spencer Reagan works at Airia. Luca Berton talked to him at AGNTCon + MCPCon North America 2026 in San Jose on 23 October 2026. His one-line description: “Airia is an agentic AI orchestration platform. We help discover and secure, optimize and govern agentic interactions.” The part of the platform he is proudest of is its MCP gateway and MCP integration. Gartner’s recent Critical Capabilities report for AI security and governance, he said, placed Airia “the farthest in the visionary section”.

Why enterprises need it

Airia’s mission, he says, is to help enterprises take advantage of agentic workloads, and “really actually use AI in the enterprise”. That brings challenges in security, governance, roles and RBAC: knowing which agentic workloads you have, securing each connection an agent has, and above all the attack vectors, which are new. “Unlike anything we’ve had in the industry over the last couple decades.” It is, in his words, a brand new world, an “empty white canvas”, and Airia has leaned into it.

From inside the house

For decades software was deterministic. Generative AI is not, and he compares it to a new employee more than to traditional software. Security used to keep the bad stuff outside the perimeter. With agentic workloads the risk is already inside, and the call is coming “from inside the house”.

Two things make that dangerous together: a malicious query or payload can reach the model, and the model can call tools across your systems. That combination “is extremely dangerous”, and because the agent already sits inside the organisation, its traffic looks trusted.

Discover, normalise, constrain

Airia first discovers every agentic workload, then normalises the data, so guardrails can apply by agentic interaction, by user or by use case. Does every agent have an identity, Luca asked? “Correct. Or at least some sort of attribution or provenance that you can use when you start applying policies.”

Many policies are like email forwarding rules: if this, then that. The layer above, which Airia calls agent constraints, uses AI analysis of an agent’s prompt and tools to decide what it is supposed to do. Then “you fill that negative space with blockers and barriers.” An agent that searches the internal issue tracker and emails you its alerts should not email outside the company or search the wrong sections. Airia can also restrict a tool call’s parameters, for example to certain email domains.

One place, not a hundred laptops

Luca compared it to the deny rules in Claude Code’s settings file. That is similar, Spencer said, but it lives locally: “if you’ve got 100 developers”, every one of them needs the restriction, and a change has to be made a hundred times. Airia moves it off the endpoints into its proxy and gateway layers, where IT and admins control it in one place. The users “don’t even have to worry about it. They’re just safe.”

When the alarm bells go off

Airia’s customers are large enterprises. Coding agents led the way; now HR, legal, sales and marketing use the same tools, knowing even less about the restrictions. When an organisation moves from developers to knowledge workers, “that’s immediately when the alarm bells go off and they need some security.”

Regulated industries such as finance add their own rules, and Luca raised Europe, where they vary between states. Where the data, the organisation and the users are decides what applies, and “that’s our job to understand what the regulations are across different geos”, so a customer hits a guardrail the moment it tries something non-compliant.

How to start

The site is airia.com, spelled A-I-R-I-A and pronounced like the measure of square footage. It has a demo link and a free trial. Beyond security, he says, there is “a whole other world of user enablement”, helping end users use AI: a subject for another conversation.

More

The episode with Spencer Reagan is being produced and goes out on YouTube first. The day it was recorded: AGNTCon + MCPCon North America 2026: day two.