SELF-
DEFINING
SYSTEMS

University of Washington

Systems that define themselves from specification through deployment—and refine from operational experience.

Core thesis

01

Time to integrate

Hours to days, not months or years, for new workloads and technologies.

02

Scalable agency

Agents exploring architecture and integration paths in parallel, with safeguards for reliability.

03

Spec → code → ops loop

From spec to generated code to deployment. Operational experience feeds back, closing the loop.

What we're building

Self-Defining Operator

The SDO deploys and operates applications with agents and long-term memory—resolving issues faster by recognizing patterns from past incidents.

Blog: agents that remember →

VibeServe

VibeServe explores how AI agents can build serving systems tailored to different LLM deployment scenarios.

News

Mar 2026

Blog: The Long Game: How Agents That Remember Resolve Operational Issues Faster.

Read the blog post