Recursive Harness Engineering
Harness engineering first, then recursive improvement.
Production AI systems · Berlin
I design production AI systems across software engineering, agents, evaluation, retrieval, security, and infrastructure—turning traces and failures into controlled experiments, reliable releases, and safer behavior.
My Recursive Harness Engineering series starts with the foundations: build a harness, evaluate its behavior independently, and test changes. Later chapters will automate the search for improvements and investigate whether experience from earlier rounds helps on new tasks.
Automate the experiment, not the authority to declare it safe.
Harness engineering first, then recursive improvement.
In order. Start at the loop.
Equations from AI papers, read symbol by symbol.
Most read
Read the equations in AI papers: find the definitions, shapes and sampling, then check on small numbers. Worked through attention, with a notation reference.
The same support tasks as an agent, an agent with approval, an agent in a workflow, a router and a workflow with no agent, compared on one test set.
How task requirements shape an agent harness: choosing control flow, tools, code execution, state, permissions and evidence before choosing a framework.
About
I’m Slava Dubrov, a Staff AI System Engineer with a doctoral degree in AI diagnostics. I work across software engineering, agent systems, evaluation, retrieval, security, and production infrastructure.