AI systems are taking on more consequential work with less human input. Watching them is not enough; they need to be understood, measured against what they were meant to do, and corrected when they drift.
We think of it as a progression: observe, understand, optimize, govern, control. Anosys today covers the first steps for agents, coding assistants and the products built on them. We are building toward the rest: the operational intelligence and control layer for AI-native systems.
We have both worked on large production systems where the data existed but the answers didn't. When an AI feature underperformed, connecting model telemetry, product usage and user behavior took weeks of engineering and data science work before anyone could say what had changed.
Agents make that problem bigger. They choose tools, hand work to each other and take different paths for the same request, and a run can finish cleanly while the result is wrong. Most tools look at one layer: traces, evals, infrastructure or product analytics. The cause usually sits between them.
Anosys joins those layers so a team can go from "something got worse" to the run, the step and the change that caused it, and then measure whether the fix worked.
A trace that looks healthy can still belong to a bad answer. We judge AI by what it did for the user and the business, and keep the trace as evidence.
Every AI system fails differently. Teams define their own metrics, judges, detection logic, pipelines and dashboards instead of living with a fixed set of charts.
Agents will read observability data more and more. People still set the goals, review the evidence and decide when to step in.
OpenTelemetry, REST and plain SDKs. Your data comes in through standard paths and stays queryable by you.
Anosys was founded in 2024 and is based in Santa Monica, California. The two founders have worked together for about twenty years, since their graduate studies.
Angelos Lazaris (CEO) has spent fifteen years in data science and machine learning, from building fraud detection and analytics products to leading data science for consumer AI products used at very large scale. He holds a PhD in electrical and computer engineering from USC. Moysis Vafeiadis (CTO) has spent fifteen years in software architecture, distributed systems and cloud infrastructure, and built most of the Anosys platform.
We use Anosys on our own agents, including the AI Assistant inside the product.
We are hiring engineers and researchers who want to help build the control layer for autonomous AI systems.
View Open Roles