Name: OpenAI 與 Anthropic 聯手推出「代理式 AI」基礎標準！ (OpenAI and Anthropic launch Agentic AI Foundation standards)
Uploaded: 2025-12-12T14:58:43.000Z
Duration: 7 min 14 s

過去簡單式

I saw this news and I was like, okay, this is pretty interesting stuff.

我看到這則新聞，就覺得，嗯，這東西還蠻有趣的。

to 不定詞

It's called the Agentic AI Foundation, OpenAI, Anthropic and Block all backing it to start.

這個聯盟叫做 Agentic AI Foundation，一開始是由 OpenAI、Anthropic 和 Block 在支持。

And the purpose here is to keep the AI agent ecosystem from becoming just a big massive mess of splinteredness and fragmented silos and all that kind of stuff.

這樣做的目的是為了避免 AI agent ecosystem 變成一團混亂、支離破碎、各自為政的局面。

Basically make it so that the agents can work together and across different, you know, all these different platforms and services and everything speak the same language.

基本上就是讓這些 agents 能夠互相合作，跨越不同的平臺、服務，並且能夠溝通無礙。

I guess it open sources the plumbing behind the scenes for AI agents.

我猜這就像是為 AI agents 開源了後端的基礎設施。

And we've talked a lot in the past about Anthropic's model context protocol, MCP.

我們之前也聊過很多 Anthropic's model context protocol，也就是 MCP。

That's part of this OpenAI's agents MD spec block's goose Agent framework, which I'm less familiar with.

這也是 OpenAI's agents MD spec 和 Block's goose Agent framework 的一部分，這個我比較不熟。

All of them are a part of this and in fact they're all donating those technologies to the effort as a part of the standardization stack.

它們都是這個計畫的一部分，而且實際上，它們都將這些技術貢獻出來，作為標準化堆疊的一部分。

So those projects are now going to live under Linux foundation governance.

所以這些專案現在都會在 Linux foundation governance 下面運作。

Anthropic is not going to have sole control over MCP anymore.

It's just kind of part of this big collective bucket.

這就像是這個龐大集體中的一部分。

我個人覺得這樣聽起來很不錯。

we had an interview with Microsoft, what, three weeks ago, I think now, where they are also trying to bring some sense to the agent world by giving them ID cards, basically.

我們大概三個星期前採訪了 Microsoft，他們也試圖為這個 agent 世界帶來一些秩序，基本上是給它們 ID cards。

One is it's not, it's not a standard until everybody signs up to it.

一個是，如果不是大家都有共識，那就不算標準。

So we do need to hear whether Google's there and Microsoft, does it come along with OpenAI at this point?

所以我們需要知道 Google 在不在，還有 Microsoft，以及 OpenAI 現在有沒有加入？

Google, yes, Other members Google is already.

Google 有，其他成員 Google 已經在裡面了。

Google, Amazon, aws, Bloomberg, Cloudflare, those are a few of the other names that I saw.

Google、Amazon、aws、Bloomberg、Cloudflare，這些是我看到的一些其他名字。

I didn't know that was a little bit more.

我不知道，這有點超出我的預期。

所以接下來就是看誰能成為標準了。

The other interesting thing that strikes me is I remember back in the day, you know, we would talk about spaces.

另一個讓我覺得很有趣的是，我記得以前，你知道，我們會談論空間。

There was namespace on the Internet and there's kind of a web space and, and

網路上有 namespace，也有類似 web space，然後

它們是這個東西裡的一個維度。

And so they're going to need protocols and standards to be able to operate properly.

所以它們需要協定和標準才能正常運作。

And I think it's a good thing if they get together at the beginning and recognize that their agents are going to have to talk to everybody else's agents and you can't really have a closed agentic world.

我覺得如果它們一開始就聚在一起，並且意識到它們的 AI agents 需要跟別人的 AI agents 溝通，而且你不能真的有一個封閉的 AI agent ecosystem，這會是好事。

You've got to be able to work with others.

And does it matter where an agent starts?

Like obviously all these companies, you know, I'm sure Anthropic wants to capture as many users as possible.

像是很明顯這些公司，你知道，我確定 Anthropic 想要盡可能多地捕捉使用者。

ChatGPT, OpenAI, they want the same thing.

ChatGPT、OpenAI，他們也想要一樣的東西。

But if you've got an agent and it knows how to speak the same language as kind of your competing AI systems, is there something for everyone to benefit from having that agent on your system integrating with your, your system?

但如果你有一個 agent，而且它知道要怎麼跟你的競爭 AI 系統說一樣的語言，那你的系統裡有這個 agent 並與你的系統整合，對大家都有好處嗎？

Like, it's interesting to figure out the scale of this.

像是，弄清楚這個規模很有趣。

You know, just generally agents aside, AI as a whole, there's a lot of talk about, about whether the web gets redesigned for that.

你知道，撇開 agent 不談，就 AI 整體來說，大家都在討論，關於網路是否會為此重新設計。

There was a paper I saw this week that tried to put forward standards around maintaining data separately from pages, separately from what we think of as stories in ways that are amenable to AI and agents.

這週我看到一篇論文，試圖提出標準，讓資料能獨立於網頁、獨立於我們認為的故事之外，而且還能被 AI 和 agent 使用。

So it changes that whole structure, number one.

所以這會改變整個結構，第一點。

Number two, at some point we're all going to be making our own agents.

到了某個時候，我們都會製作自己的 AI agents。

We used to call them programs, we used to call them apps.

以前我們叫它們 programs，叫它們 apps。

But in a sense now we're all going to be empowered to say, hey thing, go out there and do this for me, do this for me.

但現在某種程度上，我們都能說：「嘿，請幫我出去做這件事，做那件事。」

So it's not going to, I think the scale is boggling then.

所以，我覺得這個規模會讓人難以置信。

It may be coming from an agent factory that we kind of subcontract to.

它可能來自一個我們外包的 agent factory。

Google makes a nice agent tool and I use their tool to make the agent, but

Google 做了一個很棒的 agent tool，我用他們的工具來製作 agent，但

It's going all over the web and it's doing whatever I want.

它會遍佈整個 web，做我想要的任何事。

And it's not just going to the thing we know of as the web.

It's going to an agent world where it's intended to go agent to agent and it never goes through a display space such as the web, a space that was intended for human beings.

它會去一個 agent world，在那裡 agent 可以互相溝通，而且永遠不會經過 web 這種為人類設計的顯示空間。

It says I'm going to go straight to United Airlines and do this and come back and there's no such thing as a website then and then.

它說我要直接去 United Airlines 做這件事然後回來，那時候根本沒有網站這回事了。

So how do I know that my agent is going to speak well and act properly with all these other agents?

那我怎麼知道我的 agent 會不會跟其他所有 agent 溝通良好、行為得體呢？

Probably I'm going to have to use one of these companies platforms or protocols to, I don't want to say make it, I just want to instruct it to do that.

可能我必須要用這些公司的平臺或 protocol，我不想說「製作」它，我只想訓示它去做。

But there's going to be an unlimited number of agents out there going off and doing tasks and the web kind of becomes clearly it has always been a subset of the Internet.

但會有無限數量的 agent 在外面執行任務，而 web 就像是網際網路的一個子集，它一直都是。

But now it becomes very clearly a smaller subset of the Internet, proportionally, I think, because there's this activity layer going on all the way around there.

但現在它變得更清楚是網際網路的一個更小子集，我認為比例上是這樣，因為有這個活動層在周圍進行。

It's hard to get my head around and I want to think more about this, but

我有點難以理解，我得再多想想，但是

I think this is a MCP and this are good step forward.

我認為這是一個 MCP，而且這是個很好的進步。

But I think like mcp, it probably needs a protocol name at some point, or maybe it's a whole bunch of protocols they'll build.

但我認為就像 MCP 一樣，它可能需要一個協議名稱，或者他們會建立一堆協議。

I mean, this is also, you know, I'm sure kind of getting ahead of any regulatory pressure that might arise in this.

我是說，這也是，你知道的，我確定這有點超前了任何可能出現的監管壓力。

You know, avoiding a scenario where they might be compelled to open up later versus now.

你知道，避免了他們以後可能被迫開放，而不是現在就開放的情況。

They have to create this kind of neutral third party so that they're not accused of antitrust.

他們必須創造這種中立的第三方，這樣他們才不會被指控壟斷。

That's why they're all going to run behind this because they're going to say, oh, no, I'm not.

這就是為什麼他們都會跟著這個走，因為他們會說，哦，不，我沒有。