AI Agent Harness: 위임의 계약과 권한 경계
Architectural Design Decisions in AI Agent Harnesses — Hu Wei
paper-review golang kubernetes mongodb go agents concurrency streaming redis k8s memory retry deployment http verification multi-agent reinforcement-learning rag model-routing sqlite architecture observability evaluation embeddings race condition search consistency kafka backpressure transaction modularity asyncio cancellation security api rust kotlin coroutine distributed systems distributed lock sse goroutine spring boot indexes readiness git worktree configuration reproducibility multipart upload validation code-reasoning asynchronous-communication code-understanding agent-memory grpo bert language-model representation-learning survey harness orchestration text-classification gliclass zero-shot ner gliner encoder tool-calling idempotency failure recovery project-review retrieval colbert multilingual attention optimization react abortcontroller ime information-retrieval visual-document late-interaction compression vector-search provenance incremental-computation shell fault-tolerance head-of-line-blocking python materialized-view information-hiding software-design resource-lifetime subprocess backup restore toctou data integrity queueing minio logging refactoring testing cleanup tokio mcp gitlab ci rbac serviceaccount networkpolicy pvc cloudflare dns cache memory optimization data structure lock queue ttl vector db milvus data consistency array frontend cpu channel debugging performance file upload websocket realtime pub/sub regex virtual thread java21 docker build jib caching computer science thread ingress load balancer master node worker node statefulset interview