AI UNDERDOGSDAILY PICK
AI UNDERDOGS
AI 写了 12 个文件，然后呢？
Your agent made 12 files. Now what?
Workplane
AI Agent 每天往你本地丢 HTML
Markdown、半成品原型
想让同事看看？
打个压缩包扔 Slack，祈祷对方能打开
Your agent dumps HTML and half-baked prototypes into
your local filesystem every day. Want a teammate
to see it?
Zip it, Slack it, pray they open it
给 Agent 产物一个 URL
Give artifacts their own URLs
Workplane 不是又一个文件托管
每个产物自动生成独立 URL
带版本历史
人在浏览器里打开
Agent 通过 API 读取——同一个对象
两套受众
这才是产品思维
Workplane isn't file hosting. Each artifact gets its
own URL with version history. Humans open it
in a browser, agents read it via API
— same object, two audiences. That's product thinking
★ SIGNAL 1
CLI 优先，不挑 Agent
CLI-first, agent-agnostic
接入方式极简：一个 skill 文件加一个 token
Agent 从终端直接发布
不需要人插手
Claude、Codex
Cursor、Devin
MCP 都兼容
不是某个 Agent 的插件
是所有 Agent 的发布层
Integration is dead simple: a skill file plus
a token. The agent publishes straight from the
terminal — no human hand-holding. Claude, Codex, Cursor
Devin, MCP all work. It's not a plugin
for one agent, it's a publishing layer for
all of them
★ SIGNAL 2
同一份产物，两种读法
Same artifact, two ways to read
最妙的细节：同一个 URL
人打开是渲染好的页面
Agent 拿到的是结构化数据
不需要为「给人看」和「给机器读」维护两套东西
这个区分市面上九成工具没想明白
The clever bit: one URL, but humans see
a rendered page while agents get structured data
No maintaining two versions for 'human-readable' and 'machine-readable'
Most tools haven't figured this out
目前还是早期阶段，功能还在打磨
但它抓到了一个真实痛点：Agent 产出的速度已经远超人类组织的速度
中间这道缝总得有人来补
Still early, still rough around the edges. But
it nails a real problem: agents are producing
faster than humans can organize. Someone has to
bridge that gap
AI UNDERDOGS
给你的 Agent 产物找个家
Give your agent's output a home
Workplane
关注 · 每天发现更多 AI 神作
workplane.co