Automated Design of Agentic Systems
What this work does
A meta-agent writes and refines executable agent designs, storing prior candidates and their evaluations to guide later proposals.
Connection to discovery
Treats workflows, prompts and tool orchestration as searchable representations instead of assuming the harness is fixed.
Evidence
The body separates representation, search algorithm and evaluation, and reports search and test splits plus transfer of selected designs.
Scope and limitations
A Turing-complete language does not imply feasible coverage of all agents. Validation-set search, tool costs and transfer conditions constrain the result.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.