I'm an engineer working on the AI Agent harness — the runtime layer that turns a language model into a dependable agent. My focus is the harness itself: the agent loop, step execution, tool dispatch, and context management across long-running runs.
I care about the parts that make agents dependable in practice: durable session state, resumability, sub-agent delegation, model and provider abstraction, and observability. A harness should be able to explain what it did — and recover cleanly when a step fails.
I also work on Ontology and AIP-based agent development, grounding agents in a shared semantic layer instead of guessing.
|
The loop that drives an agent: planning, step execution, tool dispatch, and context management.
|
Durable sessions with resume and fork, context compaction, and long-term memory retrieval.
|
|
Sub-agents, task fan-out, and orchestration across specialized agents with bounded reporting.
|
Modelling domain knowledge into a semantic layer that grounds agent reasoning in real data.
|