A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence
What this work does
Organizes evolving agents by component, update timing, learning mechanism and application domain, covering model, context, tool and architecture changes.
Connection to discovery
Supplies a systematic map of mechanisms that can improve and retain discovery capabilities.
Evidence
The body distinguishes within-task and across-task changes, contrasts feedback mechanisms and reviews long-horizon evaluation limitations.
Scope and limitations
Surveyed task gains are not all discovery gains; reviewed systems use different feedback, budgets and persistence assumptions.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.