Now
Current interests
A lightweight snapshot of the areas I’m actively exploring. This page is intentionally easy to edit rather than automatically generated.
- Coding-agent reliability — verification, failure handling, durable state, and clearer human decision boundaries.
- Skill interoperability — making reusable agent procedures discoverable without coupling them to one harness.
- Agent evaluation — evaluating behaviours and trajectories rather than relying on plausible-looking outputs.
- Model and harness routing — choosing enough capability for the task while keeping cost, latency, and uncertainty visible.
- Context engineering — explicit project truth, decision continuity, provenance, and repository intelligence.
- Lightweight orchestration — finding where state machines and deterministic controls beat another agent framework.