AI agent adoption is creating a new control problem for technology leaders: agents do not operate only on the models they are assigned. They can also use skills, integrations and other add-ons that expand what they can access and do.
AIR has raised $50 million for a platform built around that problem. The company says its software can discover agents operating within an organization, continuously vet the skills and add-ons those agents use, and block unwanted behavior.
The issue shifts from model approval to operational oversight
Early enterprise AI governance often focused on deciding which models employees or applications could use. Agentic systems make that approach less complete.
An agent's risk profile can change as it gains access to new tools and services. A seemingly contained workflow may be able to retrieve data, call external systems or take actions through the skills and add-ons connected to it. For operators, the challenge is not simply approving an agent at launch; it is maintaining visibility as its capabilities evolve.
AIR's stated approach combines three functions: inventorying agents running in a company, examining their connected capabilities on an ongoing basis, and stopping behavior the organization does not want.
Why continuous vetting matters
A one-time review of an agent's configuration can become outdated when integrations change or new add-ons are introduced. Continuous vetting is therefore the central operational claim in AIR's model.
For security, IT and risk teams, a platform that identifies active agents could help establish a more reliable baseline: which agents exist, where they are operating and what extensions they use. That is a prerequisite for setting policies around approved connections and responding when behavior falls outside those policies.
For builders, the implication is that integrations may increasingly be treated as governed components rather than interchangeable implementation details. The ability to connect a skill or add-on could come with review and monitoring requirements, especially in environments handling sensitive systems or data.
What the funding signals
The $50 million raise points to investor interest in infrastructure around enterprise AI deployment, not only in the agents themselves. As organizations move from experimentation to broader use, controls over agent behavior and connected tools become a distinct product category.
The opportunity for AIR will depend on execution in environments where agents are distributed across teams and systems. Discovery needs to be broad enough to provide meaningful visibility, while blocking mechanisms need to be precise enough that security controls do not unnecessarily disrupt useful workflows.
What to watch next
Companies evaluating agent platforms should watch for practical evidence of how AIR handles three questions:
- **Coverage:** Which types of agents, skills and add-ons can it discover?
- **Policy control:** How do organizations define unwanted behavior and apply that definition across systems?
- **Operational impact:** Can it block risky activity without creating excessive friction for development and business teams?
The key takeaway for executives is straightforward: AI-agent governance is becoming an ongoing systems-management task. The relevant unit of oversight is likely to be not just the model or agent, but also the expanding set of capabilities connected to it.
