前几天，有人邀请老金做一场AI分享。我正好想试试dot，就让它准备8页PPT，按20分钟安排内容，每页配上讲稿。
dot是ChatGPT的个人AI助理，有自己的云电脑，也能调用我已经连接、授权的电脑。
我全程都拿着手机。 我让dot从我授权它访问的旧聊天里找素材，按这次的主题和听众重新整理，讲成普通人能听懂的话。我不用先把历史记录翻出来，再一份份复制给它。 我电脑里本来就有一套做PPT的私有Skill，也就是平时用的制作方法，这套还没开源。我就在手机上跟dot说，按这套来做。 dot去电脑里找到Skill，把模板、参考资料和脚本也一起复制到了云端。 重做前，dot问我先看一套还是两套预览。我选了一套，先看整体效果。
软件原型那页，居然画了个杯子。好在dot检查时发现了，把杯子换成软件界面，也一起修了分工图里的字和偏暖的背景。 我在手机上看过新版，回复可以，dot才接着做正式文件。 做好的8页PPT和逐页讲稿发回原来的聊天，我在手机上收到了文件。 除了单独的讲稿，每页PPT的备注里也有对应内容。 前面定了简约、典雅，后面改杯子、改背景，都在原来的聊天里接着说。 不用换一个任务，再把主题、听众、风格交代一遍。
dot把整理内容、制作页面、检查文件分开安排，收回结果给我看，需要我拿主意时再问我。 老金一直很关注multi-agent，也就是多个Agent分工协作。比如做一款产品，产品Agent弄清给谁用、解决什么问题，技术Agent负责做出来，营销Agent考虑怎么让合适的人知道它。
还得有一个Agent安排先后、汇总结果。
Agent分工以后，别把协调的活又留给我。
我很早以前设计、分享元框架时，就在考虑谁负责、什么时候推进、做完谁检查。 后来我把这套设计做成了开源项目Meta_Kim，放在GitHub上。 我用元做项目时，几个Agent就在并行干活：一个修程序，一个做独立审查，还有一个负责汇总结果。 grokbot官方展示的多Bot协作，也是各自有职责，再由一个Bot协调，遇到授权和关键判断时回来找人。我觉得这和元很接近，单说理念，我更喜欢grokbot。
要追社区里的新消息，再把热点做成视频，我也会更看好Grok，毕竟它背后接着社区。
但我之前在群里给出的个人排序，还是 dot > grokbot > Muse。
我跟ChatGPT聊过很多东西，有些没公开，本地还有用Codex、Claude Code做的项目和私有Skill。
dot能接着这些东西做事，对我就更顺手。
写在最后
我当初设计元，关心的就是一件事怎么交给一群Agent，又不用我逐个去盯。
这次准备分享，dot用上了我以前的聊天和电脑里的PPT方法，把制作和检查分头安排。我拿着手机提要求、看预览，该改什么就接着说。
我更愿意花时间拿主意，少花时间来回交代。
换成一个要做几周的项目，我还希望隔几天再回来，它还能接上进度，沿用已经定好的方法，把后面的活接着分下去。
谢谢你读我的文章。
如果觉得不错，随手点个赞、在看、转发三连吧🙂
如果想第一时间收到推送，也可以给我个星标⭐～谢谢你看我的文章。
开源知识库（实时更新 交流群）：tffyvtlai4.feishu.cn/wiki/OhQ8wq…
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
A few days ago, Kim was invited to do an AI share. I just wanted to try dot, so it could have eight pages of PPT, 20 minutes of content, each with a text.
Dot is ChatGPT's personal assistant, owns his own cloud computer and can call the computer I've connected and authorized.
I've been holding my cell phone all day. I asked dot to look for asset from the old chat that I authorized it to visit, to rearrange it along with the theme and the audience and to speak as if ordinary people could understand. I don't have to turn the record over and copy it. I had a private Skill, which I used to use for PPT, which is not open. I told dot on the phone to do this. Dot went to the computer to find Skill and copied the templates, references and scripts to the cloud. Before redoing, dot asked me if I'd read one or two previews. I've chosen a set of results first.
The prototype page of the software drew a cup. Fortunately, during the dot check, the glass was replaced by a software interface, together with the textual and warm background of the division of labour. I saw a new version on my phone, so I can answer and dot goes on with the official document. Eight pages of finished PPT and page-by-page speeches returned to the original chat, and I got a file on my phone. In addition to individual statements, there are references to PPT on each page. It's short, elegant, and then it changes the cup, changes the background, and it goes on in the chat. There is no need for a new mandate to repeat themes, listeners, styles.
Dot arranges for the organization of content, the production of pages and the checking of documents, and takes back the results and asks me when I have an idea. Kim has always been concerned about the multi-agent division of labour. For example, to make a product, the product Agent knows who to use and what to solve, the technology Agent makes it, and the marketing of Agent considers how to make it known to the right person.
There must also be an Agent sequenced and aggregated results.
After Agent's division of labour, don't leave me the task of coordination.
When I designed and shared the framework long ago, I thought about who was responsible, when to move forward, who to check. And then I turned this into an open source project, Meta-Kim, on GitHub. When I did the project in dollars, several Agents worked in parallel: one to fix the process, one to conduct an independent review and one to summarize the results. The multi-Bot collaboration, officially displayed by Grokbot, is also a separate responsibility, coordinated by a Bot, who comes back to find people in case of authorization and critical judgement. I think it's close to Won. I prefer to gookbot.
I'm going to follow the news from the community and make a video of the hot spots, and I'm going to watch Grok better, after all.
But I've sorted the individuals in the group before, but dot > rootbot > muse.
I talked to ChatGPT about a lot of things, some of them undisclosed, local projects with Codex, Claude Code and private Skyll.
Dot can do something with this stuff. It's easier for me.
At the end.
The only thing I care about when I design the dollar is how to give it to a bunch of Agents without me looking at it.
This time it's ready to share, and dot used the PPT method of my previous chat and computer to divide production and inspection. I'll take my phone, I'll read the preview, and I'll say what I have to change.
I'd rather spend my time making ideas and less time talking back.
Instead of a project for weeks, I also hope to be back in a few days, and it will be able to keep up with the progress, follow the methods already set and continue the rest of the work.
Thank you for reading my article.
If you think it's good, you can order it, read it, forward it.
If you want to get a delivery first, you can give me a sign. Thank you for reading my article.
Open source knowledge base (real time update group): tffyvtlai4feishu.cn/wiki/OhQ8wq...
Auto-fetched and translated by AI Atlas. All rights belong to the original author.