Research

My Ph.D. at Carnegie Mellon focused on service-level fault injection testing and resilience in microservice systems. Lately most of my public writing applies the same instincts—coordination, failure modes, and what “safe to build on” means—to AI-assisted development, multi-agent workflows, and the product I’m shipping, Zabriskie.

What I’m Thinking About Now

• Multi-agent systems and distributed systems

  Multiple LLM agents on one codebase behave like a distributed system: shared mutable state, stale views, lost updates, and coordination failures that aren’t fixed by better prompts. I’m interested in how ideas from concurrency control, CRDTs, and fault models carry over (and where they don’t).

• Platforms, safety, and who ships code

  As more product work is done with AI assistants, the limiting factor becomes the platform: isolation, traceability, observability, and automatic mitigation so that “welding” new features doesn’t collapse the bridge.

• Zabriskie and community around taste

  A social product for people who care deeply about music—grounded in how real-world scenes (including the jam band world) already organize around shared experience, and what’s missing in today’s tools.

• Conflict-free replicated data types (CRDTs) and distributed computing

  Still the conceptual backbone: replication, merge semantics, and coordination-free programming—now as much for reasoning about agents and tools as for classic large-scale systems.

• Education

  Teaching and public writing on distributed systems ideas.

Research Lineage & Software

• Filibuster

  Prototype implementation of service-level fault injection testing—the topic of my Ph.D. dissertation—for finding resilience bugs in microservice applications before production.

• Zabriskie (in development)

  Current product work at the intersection of community, music, and mobile—documented in the Zabriskie / AI section of this site.

• Partisan

  A high-performance alternative distributed runtime for Erlang, originally built to run large-scale Lasp experiments; it influenced later work on fault injection and testing.

• Lasp (retrospective) and SyncFree (EU/FP7)

  Lasp was a coordination-free programming model on top of CRDTs, developed in the SyncFree project alongside systems such as Antidote. Partisan and Filibuster grew out of that line of work.