Intelligence is not authority.
A model being capable of proposing an action does not mean it should be allowed to execute it. Capability and authority should be separate by design.
Zürich · Autonomous systems
I build AI agents you can trust.
I am interested in what happens when AI moves beyond generating answers and begins taking actions, exercising authority and participating in economic systems. I build around that problem through BelaZayka and CognoKratos.
Autonomous systems Bounded authority Agentic finance
01 / What I believe
AI agents need more than a smart model.
The model is only one component. Trust emerges from the architecture around it: what can act, under whose authority, with which limits, and with what evidence after the action happened.
Autonomy requires boundaries.
Identity, permissions, budgets, human approval, observability and deterministic controls are part of the system — not safety features added after the demo works.
Economic agents need machine-native finance.
As autonomous software begins buying resources, moving value and coordinating with other services, financial authority must become explicit, programmable and auditable.
02 / The work around the thesis
Two homes for the work.
The ideas live in two places. For engineering work and collaborations, start with BelaZayka. For open-source laboratories, the book and the community around them, start with CognoKratos.
Turning the ideas into systems.
BelaZayka is the engineering company I created around trustworthy autonomous systems. We work on AI agents, the infrastructure around them and — where autonomous software needs economic agency — the programmable financial systems that make bounded authority possible.
01 / AgentAutonomous systems
02 / ControlTrustworthy AI infrastructure
03 / EconomyAgentic finance
PrincipleCapability without uncontrolled authority.
IdentityWho is acting?
PolicyWhat is allowed?
AuthorityWho can commit?
EvidenceCan the system explain and prove what happened?
Making the engineering inspectable.
CognoKratos is the open-source side of the same thesis: runnable architectures, engineering laboratories and educational material for people building increasingly consequential agentic systems. The book is the textbook; the repositories are the laboratories.
Production agents
Trust boundaries, tool use, observability and human approval around a probabilistic runtime.
simple-agent-template ↗Cryptographic capability
Blockchain-backed capability boundaries that let agents request wallet operations without exposing private keys or custody.
arktos-wallet ↗Financial authority
Separating proposal, approval and execution so AI capability never silently becomes financial authority.
tauros-revenue ↗03 / Where the ideas come from
A short professional context.
The path that led me to the problems I work on today. A professional profile is available on LinkedIn ↗.
Agentic AI in regulated finance
Production agentic AI for financial compliance: governed agents, knowledge systems and evaluation, designed for regulated environments.
AI, blockchain and financial infrastructure
The intersection of AI and Ethereum: stablecoin and digital-asset infrastructure, and domain-specific LLMs for smart-contract engineering.
Bridging data science and software engineering
ML, data and distributed systems for businesses across Europe, treating models as production software rather than isolated experiments.
Selected external validation NVIDIA · Agentic AI ↗ NVIDIA · Generative AI / LLMs ↗ NVIDIA · Multimodal AI ↗