What I'm working on /now.
Evaluating build-vs-buy across the AI stack, designing the next generation of prose-first agents, and writing about what actually ships in production AI.
live · 5 active threads
● Status feed · 5 threads
Building A reference architecture for prose-first agentic systems.
Evaluating Custom ML vs. managed services for document understanding.
Writing A field guide to prose-first agents for ops teams.
Reading Recent eval and agent papers; team-topology essays.
Open to Director / Head of AI conversations.
01Building
- A reference architecture for prose-first agentic systems — turning natural-language specs into deployable agents with retrieval, evaluation, and guardrails baked in.
- AI-driven QA tooling that closes the loop between failing tests and code suggestions.
02Evaluating
- Custom ML vs. managed services (AWS Comprehend, Azure AI) for document understanding at scale.
- Agent evaluation harnesses — what "passes" really means for non-deterministic systems.
03Writing
- A field guide to prose-first agents for operations teams.
- Lessons from taking AI from prototype to production in a regulated environment.
04Reading
- Recent eval / agent papers, plus anything thoughtful on team topology for AI orgs.
05Open to
- Director / Head of AI conversations where strategy, hands-on architecture, and team leadership intersect.
"Now" pages are inspired by nownownow.com — a snapshot of current focus, not an archive.
✉ get in touch