Make enterprise autonomy reliable and evolvable in production.
Noether builds control architectures for enterprise AI in production: replayable decision traces, deterministic verification at runtime, and continual reliability learning grounded in real workflows. Systems intelligence that makes autonomy inspectable, reproducible, and safely expandable.
Agents in production; reliability is reactive and we debug after the fact.
Same failure modes recur with no decision trace or replay, and postmortems don't scale.
A system of record for decisions becomes non-optional or operational debt compounds.
Systems that learn from failure operate at scale, or autonomy stays brittle and can't expand.
Research at Noether
Our research program spans four areas that together aim at operational superintelligence: enterprise AI that improves from experience under strict safety constraints. The central question is how fast agentic systems can learn from production at the pace of deployment.
01
Tribal Knowledge Systems
Knowledge systems that evolve with use. They ingest an enterprise's code, data, dashboards, experiments, and tasks, and maintain them as living artifacts instead of static documentation. The goal is institutional memory that compounds over time rather than decaying.
02
Enterprise World Models
Simulation environments with sufficient fidelity that agents can practice, accumulate experience, and improve through reinforcement learning off production. They provide the safe learning substrate for organizations running autonomous agents in high‑stakes settings.
03
Differentiable Knowledge Systems
Representations and algorithms that extend backpropagation-style updates into knowledge systems. The aim is to update not only model weights but the artifacts agents reason with: runbooks, policies, decision frameworks, and institutional heuristics.
04
Verification at Scale
Methods for evaluating agent behavior over long horizons, from single decisions to full‑system trajectories, and for closing the loop between economic outcomes and policy updates. Our work here includes TraceOps benchmarking and production verification infrastructure.
Team
We are hiring researchers with deep expertise in reinforcement learning and knowledge representation, with a keen interest in bringing their more esoteric aspects to real enterprise product surfaces in production. People who thrive at Noether take permissionless initiative, design their own experiments, and don't wait for structure to be handed to them.
- Reinforcement Learning / ControlSan Francisco or Remote
- Research Engineering (Infra + ML)San Francisco or Remote
We believe the next breakthroughs in reliable AI will come from systems that learn from production data—and we're building that layer. If you have aligned expertise and are excited by our mission, please get in touch.