RELIC: building long-horizon agent organizations
A long-horizon testbed for LLM-agent organizations, including OrgEnv and professional profile graphs.
Organizations that persist beyond a conversation
I am leading a long-horizon research testbed for LLM-agent organizations. The question is how a group develops persistent organizational capabilities while working under constraints such as roles, skills, deadlines, tool access, fatigue and budget.
The focus is collective production over time: agents must work with shared repositories, documents and workspaces rather than treating each interaction as an isolated conversation.
Profiles, shared work and OrgEnv
Professional-profile graphs provide structured priors over roles, skills, memory and decision preferences. They make organizational heterogeneity more explicit than persona-style prompts alone.
OrgEnv supplies a startup-style product-development environment, with repositories, workspaces, customer events and evaluation of multi-day product-development outcomes. A perception–memory–planning–action–reflection loop operates under explicit time and resource constraints.
My role in the research
My responsibilities include problem formulation, benchmark design, profile-graph construction and experimental design. The planned comparisons examine homogeneous teams, role-only agents, aligned or shuffled professional profiles, and institution-formation mechanisms.
This early testbed work is part of RELIC, not a separate project. The current project page follows the research into persistent organizational state, executable protocols and human–agent organizations.
Loading comments…