Perplexity 用「Teammate」智能体并行调度 48 个沙盒渲染 Pantheon 3D 视频
Perplexity 的 teammate 智能体自己写 three.js 代码渲染万神殿，每帧 80 秒太慢，它就把任务拆到 48 个沙盒并行跑，3 小时出片。做 agent 编排或 AI 视频工具的可以看看这套并行调度思路。
Perplexity 工程师 Jeff Grimes 展示了一个 Video Production teammate 智能体的完整工作流：它自主抓取 Pantheon（万神殿）的测绘尺寸和近期照片，编写 three.
js 渲染器并加入程序化着色器，最终输出 60 fps、逐帧带 motion blur 的写实视频。每帧渲染耗时 80 秒，智能体把任务拆分到 48 个并行沙盒同时执行，总耗时压缩到 3 小时。
这个案例展示了智能体编排算力资源的能力：不是单次问答，而是拆解任务、并行调度、产出成品文件的端到端流程。与 Perplexity 此前用 Comet 浏览器智能体做网页操作的路线相比，这条 workflow 更接近视频生产管线，说明同一家公司在把 agent 从浏览器扩展到 3D 渲染等重计算场景。
对做 AI 视频工具和 agent 编排框架的开发者来说，48 沙盒并行是可直接借鉴的工程模式。
“Teammate orchestrated rendering across 48 sandboxes in parallel so that the whole process took only 3 hours” Jeff Grimes @jeffgrimes9 Perplexity recreated the Pantheon in photorealistic detail. I created a Video Production teammate that pulled survey dimensions and recent photos, wrote a three.js renderer including procedural shaders, and rendered the final video at 60 fps with motion blur on every frame. Each frame took 80 seconds to render, and the teammate orchestrated rendering across 48 sandboxes in parallel so that the whole process took only 3 hours.
This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Perplexity, Teammate, three.js, agents.
Source attribution: curated from Aravind Srinivas (@AravSrinivas).
