Designs and builds the harnesses that enable AI agents to act reliably in the real world: tool interfaces, environment adapters, execution sandboxes, and verification loops that connect language models to live systems (exchanges, blockchains, and enterprise infrastructure). 6+ years building production agent harnesses and orchestration pipelines on rig-core (Rust-native agent framework), pairing self-reasoning, self-correcting agents with strict pre-execution verification (PEV loops), sandboxed execution contexts (SignerContext), and cryptographically signed, auditable outputs. 15+ years in software engineering with a testing-first foundation (TDD, BDD, ATDD, 100% automated coverage on production systems), applying the same rigor to building test, evaluation, and grading harnesses for agentic AI systems.
Why did the neural network break up with its training data? Because it felt used and said, “I think we should see other distributions.