Viren Baraiya：把 AI 智能体的规划与执行分开，安全运行生产系统

核心内容

Orkes 联合创始人兼 CTO Viren Baraiya 提出把智能体分成负责规划的“大脑”和负责确定性执行的“手”。发布方介绍将讨论重点放在长期运行、事件驱动和会产生实际操作后果的智能体，说明仅有一次模型调用为什么不足以支撑生产系统。

演讲涉及审批关口、幂等性、操作记录与持久化工作流，以及把智能体生成的计划转为 Conductor 工作流的做法。示例围绕 SRE 场景展开，分别呈现推理与执行循环，让计划变化与执行约束各有明确位置。

Viren Baraiya, co-founder and CTO of Orkes, proposes separating an agent’s planning “brain” from its deterministic execution “hands”. The publisher’s description focuses on long-running, event-driven agents that cause real-world side effects, and why a single model call is insufficient for production operation.

The talk covers approval gates, idempotency, action logs, and durable workflows, including compiling an agent-generated plan into a Conductor workflow. An SRE example illustrates separate reasoning and execution loops, giving changing plans and execution constraints distinct roles.

介绍依据 · 发布方说明 · 完整内容待核读

编辑导览，依据原始来源 ↗，整理于 2026-10-09。中英文均由本站概括整理。内容介绍与下方原话、主张的逐条核对分开标注。

生产环境不能容忍非确定性扩散到系统副作用中：必须将非确定性的大语言模型限定为负责规划的“大脑”，而将具体任务执行完全交给具备持久化、审批关口与幂等保证的脚手架“双手”，使智能体本质上成为运行时动态编排的后绑定 Saga 工作流。

原话

2逐字核对过；核对了说话人和位置

the responsibility of a non-deterministic agent is to plan, is to plan what should happen next, not really to do things. Um and then Harness is the one who does actually execution. And And this is important for uh various reasons, right? One being that Harness is a deterministic piece of code

自动字幕 · 已核对Viren Baraiya▶ 00:08:09

When you think about agentic harnesses, they are essentially late bound sagas. And what I mean by that is that they have a very finite and deterministic set of tools uh and the task that they can operate on. Instead of putting them together upfront, the agent is kind of proposing and building them at runtime.

自动字幕 · 已核对Viren Baraiya▶ 00:10:21