A live lab for measuring autonomous agents
SPARK members are writing the frameworks for how agents should be selected, coordinated, evaluated, and governed — like James Massa's SCORE-AI. This summer, SPARK provides a live laboratory for those ideas. On Recursiv's agentic AI platform, student teams and members implement, run, and measure autonomous agents on real workloads.
There's a place for you whether you want to bring a problem, test a framework, lend domain expertise, or co-author the results.
The landscape
A fast-moving set of standards is forming for how autonomous agents should be governed and orchestrated — AIUC-1, AARM, and the orchestration frameworks SPARK members are actively writing. The program opens by mapping that terrain, so every team is working from the same picture before anyone deploys an agent.
The live lab
Recursiv is the lab. Teams deploy swarms, apply orchestration patterns, and measure what actually happens.
Deploy agent swarms
Stand up teams of agents on real workloads and apply the orchestration patterns members are designing.
Reviews + audit trails
Every run captures performance reviews and a full audit trail of what each agent did and why.
Observable + reproducible
Results are observable and reproducible, so findings can be independently checked rather than taken on trust.
The experiments
Three tracks, each chosen so the result can be trusted.
Known-answer
The result is established up front, so the group can independently verify the platform's judgment.
Member-defined
Real industry problems brought by SPARK organizations from their own domains.
Public-impact
At least one high-visibility problem where the stakes are real and the work matters beyond the lab.
How to get involved
Pick how you want to plug in. Most participants do more than one.
Contribute a problem
Bring a real problem from your domain for the swarms to take on.
Test a framework or pattern
Implement and stress-test an orchestration framework or pattern in the live lab — SCORE-AI and others welcome.
Lend domain expertise
Help define what success looks like and verify results in your area.
Co-author the results
Findings are published; the people who shaped and verified them are named on the work.
Be part of it
The June 24 Special Session kicks things off with a live demonstration. Come see the lab in action, then tell us how you want to contribute.
jack@recursiv.io · jshort@ucsd.edu