We Don’t Want Them Ungoverned
Why We Invested in Onyx Security
An AI agent can now discover a tool on the internet, call it, and take action inside a production environment—all without a human in the loop. That sentence would have been speculative two years ago. Today it’s a slow Tuesday night.
This is the biggest platform shift of my career in technology, and it is happening with unusual speed. Adoption is coming from both directions at once: pushed by leadership because the productivity gains are obvious, and pulled by employees because the tools are already too powerful to ignore. For enterprises, that combination is powerful and unnerving. The mandate is clear, but the guardrails are not.
Agents are already spreading across the enterprise: inside agent-building platforms, coding and security tools, endpoint environments, and homegrown AI applications. On the leading engineering teams, agents are already executing far more work than humans. Soon that will be true across every function.
Houston, we have a problem.
Agents don’t just generate text. They speak to external systems. They execute code and terminal commands, take actions against production environments, use unaudited tools and skills pulled from the internet, and in some cases communicate directly with customers. They blur foundational assumptions around identity, permissions, review, and accountability. They are becoming a new operational layer inside the enterprise—and a new risk surface at the same time.
Companies need a way to govern these systems without crushing the productivity gains that made them valuable in the first place. That requires new infrastructure. The old approaches were built for deterministic software, static identities, and human-driven workflows. Agents are none of those things.
This is also a hard technical problem. Policy has to scale across dynamic, evolving systems. Intent has to be made legible, not just activity. Agent behavior has to be monitored and sometimes redirected in real time.
And as agents proliferate, deployment demands a common orchestration and routing layer that works across a fragmented landscape of models, tools, permissions, and applications. No one has built this yet. The companies that do will define the next era of enterprise security.
This is why we’re excited about Onyx.
Maxim Bar Kogan and Gil Elbaz are building the control plane for AI. Onyx sits across the enterprise, giving organizations a way to govern agents and AI applications at scale. Onyx trains its own models and builds bespoke agents to understand AI reasoning in real time—moving enterprises past surface-level monitoring toward genuine and scalable comprehension of agent behavior. The problem they’re solving spans security, trust, alignment, and cost. The challenge of a rational policy across all that breadth is very big.
This is a team of special technologists. Maxim finished his CS and math degree while still in middle and high school and won gold in Israel’s national computer science competition. Gil, at twenty years old during his military service, started an AI data company whose first customers included Apple, Google, and Microsoft. We at Conviction surround ourselves with researchers who have fully internalized the exponential curve of AI progress, and yet Maxim regularly surprises me with how far ahead he is willing to act on it. Onyx builds, fast, for capabilities that aren’t here yet but will be soon. It’s the only way to help our customers get ahead of the curve and enable their transformation.
The Onyx team has the qualities that matter most in this era: speed and understanding. They are building a killer go-to-market team, operating in close partnership with customers, and already showing traction with Fortune 500 companies and AI-forward organizations—including amongst discerning customers in critical infrastructure and financial services.
We want agents supporting our banking, energy, and technology companies. We don’t want them doing so ungoverned.
At Conviction, we are an AI-native fund backing the founders building this next generation of companies that matter. We think this is one of them. We are proud to partner with Maxim, Gil, and the Onyx team, and excited to work again with our longtime friends Gili, Hila, and Cyberstarts.
The agents are ready. Onyx is making sure we are too.