Experiment Pattern
Experiments need an exit ramp
Most prototypes should teach something and disappear. The useful ones need a path into shared infrastructure, support, and governance.
Higher education technology leadership
Making technology useful in higher education.
I lead technology, infrastructure, and digital workplace teams at UC San Diego, turning emerging tools into dependable services people actually use. Right now that work centers on practical AI through TritonAI. This site collects the field notes, architecture, talks, and product experiments behind the question of what should be temporary, what should become durable, and what should scale.
Operating idea
Run experiments deliberately, keep what proves durable, and move the surviving patterns into shared infrastructure, governance, and scalable practice.
UC San Diego field work
Moving from promising tools to dependable services, shared practice, and campus trust.
Working questions
- 01How should IT prepare the institution for a shift from applications people use to agents that work on their behalf?
- 02How do we decide which AI experiments should be retired, repeated, or scaled?
- 03How does IT govern agentic systems that can act across tools, data, and workflows?
Current focus
AI as institutional infrastructure: TritonGPT is scaling from a flagship service to a multi-campus platform, with several peer institutions currently in evaluation. We're focused on building the deployment models that allow peer institutions to adopt this vertical AI approach safely.
Updated May 24, 2026How it connects
Institutional inputs become useful outcomes.
Knowledge, people, and tools route through governed institutional AI into the everyday work of service, instruction, and research.
System map
Useful AI moves from experiment to evidence to scale.
01
Experiment
small bets, prototypes, local workflows
02
Evidence
usage, feedback, risk, fit
03
Durable Layer
platforms, APIs, governance, patterns
04
Scale
shared services, adoption, support
Field notes
What survives the experiment becomes the pattern.
Decision Signal
Evidence decides what survives
The work is separating enthusiasm from durable value: who used it, what changed, what risks appeared, and what should scale.
Architecture Note
Scale is a design constraint
A durable pattern has to survive outside the pilot team: documentation, ownership, funding, privacy, and repeatable operations.
Evidence and conversation
A publication index, not a logo wall.
The media page keeps the longer record. These are a few entry points into the public conversation.