Anna Varosyan explains why production AI needs a trust layer outside the agent to enforce identity, delegation, policy, and verifiable decisions at the moment of action.
About Anna Varosyan
Dr. Anna Varosyan is Deputy CTO and VP of Engineering at Affinidi, with over 25 years of experience leading world-class engineering teams and strategic technology partnerships. Holding a PhD in Computer Science, she specializes in workflow-based formal verification, practical AI in identity management, and the decentralized technologies that enable it. At Affinidi, Anna builds privacy-focused products that let individuals, organizations, and their AI agents control their data and digital identity, and leads work on trust infrastructure for agentic AI. Her career spans leadership roles at Affinidi, HERE Technologies, and Synopsys, with deep expertise in software, reliability, and productivity engineering. A dedicated mentor and coach with numerous scientific publications, Anna is passionate about secure, impactful technology.
What motivated you to join the AI Builders Global Conference?
Absolutely. What excites me about AI Builders Global Conference is the strong focus on people who are actually building AI systems and dealing with the realities of taking them into production. We are at a fascinating point where AI is moving from systems that primarily answer to systems that can reason, decide, and act. That creates enormous opportunities, but also a completely new set of engineering challenges. I'm looking forward to exchanging perspectives with builders who are asking not only "What can we build with AI?" but also "How do we build it so that people and enterprises can actually trust it?"
What shaped your journey through technology and AI?
My background is in engineering and technology leadership, building and scaling products, platforms, and engineering organizations. My PhD work on formal verification left me with a lasting conviction: it isn't enough for a system to behave correctly; you need to be able to prove it did. What has always interested me is what happens when promising technology meets the complexity of real-world production. With AI, that gap is particularly interesting. Building an impressive prototype is becoming easier. Building an AI system that is reliable, observable, secure, governed, and trustworthy at scale is much harder. As AI has evolved toward increasingly autonomous and agentic systems, my focus has naturally shifted toward the intersection of AI engineering, AgentOps, identity, governance, and trust. For me, the interesting question is no longer simply whether AI can perform a task. It is whether we can confidently allow it to perform that task in a real production environment.
Why do AI systems need a trust layer?
Because I believe we are solving only part of the production AI problem today. LLMOps and AgentOps help us build, deploy, observe, and operate AI systems. But they mostly tell us what an agent did after the fact. As agents become more autonomous and start touching real systems, real money, and real data, that isn't enough. We need to answer another set of questions: Who is this agent? Who or what is it acting on behalf of? What is it allowed to do, right now? And can we prove afterwards why it was allowed? That is where I see the need for a Trust Layer, and the evolution from AgentOps to what I call TrustAIOps: deciding, at the moment of action and outside the agent, whether an AI agent may act, and proving it afterwards. If a rule lives only in the agent's prompt, nothing can enforce it. So the trust layer sits outside the agent, binds every decision to identity and delegation, and grades its verdict: allow, constrain, escalate, or block. Every decision leaves a signed, policy-versioned record that an auditor or partner can verify. Each era of Ops added a new kind of evidence. TrustAIOps is the first that turns evidence into a gate. Some pieces are still open, like knowing when and why an agent's or LLM's behavior has drifted from normal, and I think the whole community needs to work on them together.
Who should attend the AI Builders Global Conference?
I'd recommend it to anyone who is moving beyond experimenting with AI and thinking seriously about how to build and operate it in the real world. That includes AI and software engineers, architects, CTOs and engineering leaders, product leaders, founders, security professionals, and anyone working on agentic AI. The most valuable conversations now sit at the intersection of disciplines. AI is no longer just a model problem. It is becoming a systems engineering, product, security, governance, and trust problem. Bringing those perspectives together is exactly the kind of conversation I'm looking forward to at AI Builders Global Conference.