Tag
AgentEval is a system for evidence-bounded assessment of stateful agents, carrying contract-bound assessment records through tool execution, evidence capture, report qualification, and case maintenance to distinguish satisfied, violated, and unresolved obligations.
Vercel 宣布 Drives 正式可用，为在 Vercel 上运行的有状态 AI agent 提供文件/存储层，配合其此前提出的「brain、hands、files」现代可扩展 agent 架构理念。
微软发布 ThinkingBox-Bench，包含 507 个状态化企业工作流任务，每个任务独立运行 20 次并以终端数据库状态评分，揭示 pass@20 与 all-20 排名几乎完全相反，且 67% 的失败轨迹在完成式代理下会被误判为成功。研究显示发现能力与可重复性对模型的排名差异巨大，Kimi-K3 覆盖最广但重复率低，Claude Opus 5 重复稳定性更高。
TwinCheck is an inference-time verification policy that enhances stateful tool agents by using evidence-grounded negative-twin comparisons, significantly improving task success rates in benchmarks like BFCL V4.
Neo4j's Will Lyon presents a recorded session on designing stateful AI agents with memory and context graphs, covering pitfalls and integrations with Salesforce Agentforce and Databricks.
WorldLines introduces a benchmark for long-horizon embodied household assistance, featuring memory QA and embodied task planning with partial observability, and proposes ObsMem, a visibility-aware memory framework.
Yohei Nakajima predicts that 2025 will focus on agent memory and 2026 on stateful agents, arguing that better memory is the key unlock for truly improved AI agents.
A tutorial from Google on building long-running AI agents that can pause for days, survive restarts, and resume without losing context using the Agent Development Kit (ADK), with code and step-by-step guidance for enterprise workflows like new hire onboarding.
A tweet asking if anyone has seen an elegant primitive for implementing stateful agents, decision traces, and context graphs.
Yohei Nakajima shares his growing understanding of stateful AI agents.