Empirik secures $21M to predict software outages with AI

Empirik is part of a broader trend of using AI agents not just to write code, but to run the production stack that code lives on. By positioning itself as an autonomous ‘traffic cop’ for infrastructure changes, Empirik is turning model-based reasoning loose on one of the most operationally brittle parts of modern software: config changes, rollout pipelines, and complex dependency graphs.

This matters for the race to AGI because infrastructure reliability is one of the hidden bottlenecks in deploying powerful models widely. When every feature team is wiring LLMs and agents into existing systems, the blast radius of a bad change grows. Tools that can predict and prevent outages become a necessary complement to scaling AI workloads. If they work well, they reduce the operational friction of experimenting with more capable agents and chaining them through more systems.

There is also a feedback loop angle: models that watch millions of real-world changes and their outcomes could become very good at reasoning about causal structure in complex systems. While Empirik’s data is proprietary, this class of application is exactly the kind that, if scaled, could give labs or partner companies unique, high-signal training data about how large sociotechnical systems behave under perturbation.