AI-powered workflow automation with Zapier co-founder Mike Knoop - YouTube

Okay, up next we have founder demos. Mike from Zapier, which rhymes with happier, is going to show us a little bit of what they've been working on over at Zapier. Hey, everyone. I'm Mike, one of the co-founders of Zapier. Thank you, Sonia and Pat, for inviting me back this year to show us something new. One year ago, I was here, we were launching our V1 of AI Actions on Zapier, an API that sort of translates natural language into executed API calls. And a bit of an update on that. Over the last year, we've executed over half a million of those on behalf of our users. And to date, we've been launching all the tons of other stuff AI related on Zapier. And to date, we've run over 50 million AI tasks on Zapier. If you're not familiar with Zapier, task is basically something that's completely automated on your behalf. It's completely hands off.

好的，接下来我们有创始人演示环节。Zapier的创始人之一Mike将会向我们展示一些他们在Zapier上所做的工作。大家好，我是Mike，Zapier的联合创始人之一。感谢Sonia和Pat邀请我今年再次回来展示新的东西。一年前，我在这里，我们刚刚推出了我们在Zapier上的AI Actions V1，这是一个可以将自然语言转化为执行API调用的API。关于这个项目的更新，过去一年，我们代表用户执行了超过50万次这样的操作。至今，我们在Zapier上已经推出了大量其他与人工智能相关的项目，并且已经在Zapier上运行了超过5000万次AI任务。如果你对Zapier不熟悉，任务基本上是在你的代表之下完全自动化的事情。它完全无需人为干预。

We've heard a lot about agents today. And clearly it's happening. In fact, a lot of it's happening right now on Zapier. These are totally 100% automated. A lot of those 50 million tasks are using Zapier classic look something like this. This is sort of a canvas view, right? This is usually how people set up and build Zap's lots of structure. There's a high degree of learning. There's a big learning curve. Lots of configuration, nitty gritty stuff, you got to get right. And as a result, this sort of puts a huge sort of depression on the amount of people who can successfully use and adopt Zapier. So that's what I've been working on the last six months. Excited to give you all one of the first public previews of a new product called AI Bots inside of Zapier Central is the name of the product. And this is a sort of ground up reimagining of what we're calling AI automation. Entire goal is to help more people in the world be able to successfully use, figure out, discover and actually adopt automation using AI. So let's show a demo of this thing. Right now that you probably looks pretty familiar, you know, standard chat by, you can sort of come in here and say hello, and it will talk back to me. It doesn't do anything right now. This is sort of a blank slate bot. So let's customize it. We're going to build a new behavior. Behaviors are how you sort of customize what these AI Bots can do on your behalf. Now these Bots can, you can customize them to like sort of change the behavior when you talk to them.

今天我们听到了很多关于代理的信息。很显然，这是正在发生的事情。事实上，很多事情都正在Zapier上发生。这些都是完全自动化的。那50百万任务中有很多是在Zapier经典版上进行的，就像这样。这是一种画布视图，对吧？这通常是人们设置和构建Zap的方式。这需要高度的学习。有一个很大的学习曲线。需要很多配置和琐碎的东西，你必须搞定。结果，这使得成功使用和采用Zapier的人数大大减少。所以过去六个月，我一直在致力于这个。很高兴给大家展示一个新产品的首次公开预览，这个产品叫做AI Bots，位于Zapier Central产品中心。这是我们所谓的AI自动化的重新设想。整个目标是帮助更多人成功地使用、发现和采用AI来进行自动化。接下来让我们展示一下这个东西的演示。现在你应该觉得很熟悉，你知道，标准的聊天界面，你可以在这里输入你的问题，它会回答你。现在它还什么都不做。这是一个空白的机器人。所以让我们来定制它。我们将建立一个新的行为。行为是你如何定制这些AI机器人代表你做什么的方式。现在这些机器人，你可以定制它们，来改变当你和它们交谈时的行为。

But I think a more cool thing to show off something more new novel is how to control the spot doing stuff on your behalf in the background, even when you're sort of away from your computer. So I'm going to take a second to type in a quick prompt I have prepared. Maybe we'll do this. All right, so let's do schedule every morning. Get the list of today's meetings on my cheek. Well, summarize, define short, bullets, emojis, send the summary to me in Slack. And we'll do this. Add a fun automation inspired quote at the end. Alright, lots of typos, but you all know that's resilient to that. So here we've got the bot starting to suggest how to automate this. So in this case, we're going to use this suggested schedule trigger. It can happen every day, every morning.

但我认为更酷的事情是展示一些更新颖的东西，就是如何控制在你不在电脑旁的情况下，背景中的东西自动进行操作。所以我要花一点时间输入我准备好的快速提示。也许我们会这样做。好吧，让我们每天早上安排一下。得到今天的会议列表在我的微信上。总结，定义简短，用符号，表情发送给我在 Slack 上。我们将这样做。在最后添加一个有趣的自动化引发的引语。好吧，有很多错别字，但你们都知道我们能对此有所适应。所以这里我们有机器人开始建议如何自动化这个过程。在这种情况下，我们将使用这个建议的日程触发器。它可以每天、每天早晨发生。

Let's say, sure, 5 a.m. earlierizer. And we'll add that as a trigger. And now what we're going to do is we're going to equip this AI bot with a set of available actions. So the way to think about this is you're sort of giving this AI bot permission to act on your behalf and do things in the real world. For this use case, the two that got suggested that are relevant, we're going to do this Google Calendar find event. We want it to be able to find all of the calendar events. And I'm going to actually choose a specific calendar. We'll come back and I'll actually talk a little bit more about this in a minute. We'll add that. And then we want it to be able to send a Slack message. In this case, let's not do channel message. We'll do a direct message to me. Got that. And then here, I'm also going to choose a specific value for me and Slack. And add that. And we're all set. All right, this bot set up. It's done. I can turn it on and it will work tomorrow morning.

让我们说，好的，提前到5点。然后我们将把这个作为触发器。现在我们要做的是，给这个AI机器人配置一组可用的动作。所以想想这样，你在某种程度上给这个AI机器人允许代表你行动，并在现实世界中做事情。对于这个用例来说，建议两个相关的是，我们要做的是谷歌日历查找事件。我们希望它能够找到所有的日历事件。我将选择一个特定的日历。我们一会儿会回来，我将详细谈一谈这个。然后我们希望它能够发送Slack消息。在这种情况下，让我们不要发频道消息，而是给我发送直接消息。已经搞定。然后在这里，我还会选择一个特定的值来给我和Slack。添加上去。我们已经全部设置好了。好了，这个机器人已经设置好了。我可以打开它，明天早上就能够工作了。

Alright, if you're like most users who use app, you like to test things and make sure that they work the way they do. So we'll go ahead and click test behavior. And this is kind of one of the first cool things is we kind of introduced this concept of threading, which is how to kind of group context together for a single my task that the bot is doing. I'll try to pull this guy over. Is this legible? I can also zoom in a little bit. There we go. A little more. So there we go. Went ahead and pulled off my calendar. It's giving me a preview of what it's going to retrieve and that all looks good. So we'll say, yep, that looks good. Go ahead. Now one of the cool things about this bot while this is sort of processing here is you notice how you like sort of picked certain values for the calendar and the slack. So that's not required.

好了，如果你像大多数使用这个应用的用户一样，你喜欢测试功能，并确保它们按照预期工作。所以我们会继续点击测试行为。这是一个很酷的东西，我们引入了线程的概念，用于将上下文组合在一起以完成机器人正在执行的单个任务。我来尝试拉这个小家伙过来。这看得清楚吗？我也可以放大一点。对，现在好多了。这样我们就可以继续了。我已经成功拉取我的日历了。它给我展示了将要检索的内容预览，一切看起来都很好。所以我们会说，是的，看起来不错。继续吧。现在，这个机器人的一个很酷的功能是，在这个处理过程中，你注意到你选择了日历和slack的特定值。但这并不是必需的。

One of the things these bots do out of the box is they will sort of guess everything, right? This takes advantage of a lot of the AI actions technology we built over the last year. Out of the box, they will try to infer and guess and do things completely unprompted zero shot. However, one of the things we learned from real users is that doesn't always work. Oftentimes you want a degree of control over how the bot actually makes its decisions in order to increase the sort of scope of use cases. I think one interesting thing I found is like not every user needs that though. There's actually a lot of use cases where you might just say be fine with sort of these bots, you know, making decisions completely autonomous on your behalf. But there are a lot of use cases and it sort of opens up the aperture, the amount of use cases you can do if you can the users can get more sort of more deterministic control over how the bot makes decisions and what it does.

这些机器人的其中一项功能是它们会随机猜测一切，对吧？这利用了我们在过去一年中建立的许多人工智能技术。它们会尝试推理、猜测，并完全自发地做事情，零提示。然而，我们从真实用户那里学到的一点是，这种做法并不总是奏效。许多时候，你想要对机器人如何做出决策有一定程度的控制，这样可以增加使用案例的范围。我觉得有趣的一点是，并非每个用户都需要这样做。实际上，有很多使用案例，你可能会觉得这些机器人完全自主地代表你做决定就可以了。但是有很多使用案例，这样做成了开放的选择，如果用户能更加确定地控制机器人如何做决策和它的行为，那么可以做更多的使用案例。

Alright. So it looks like a good calendar event. Sure. Let's go ahead and say go. It does like to check in a lot. So that's on our roadmap to tune that down. Other sort of fun surprises I think that we've learned from these AI bots. Frankly, we're still learning a lot about what these bots sort of can do. There's a lot of cool like latent capabilities built in. Even that I've seen, I think Andrew mentioned this morning, there's a lot of self-healing capabilities. You know, these are sort of hooked up in those like agentic loops.

好的。看起来是一个不错的日历事件。当然。让我们继续前进吧。它确实喜欢经常检查。所以我们计划减少这种行为。其他一些有趣的惊喜，我觉得我们通过这些人工智能机器人学到了。坦率地说，我们仍然在学习这些机器人可以做什么。它们内置了许多酷炫的潜在功能。就连我自己也看到了，我记得安德鲁今天早上提到过，它们有很多自愈功能。你知道，它们被连接在那些智能回路中。

So we've been able to observe like in the bot running into dealing with things that classically would just break zaps because they're sort of super rigid, right? Like if you had a workflow hooked up to Google spreadsheet and somebody came in your team and they changed the name of the column in the spreadsheet, classically like a zap, which is break, it'd be stuck. You know, you get an air email, you have to come back and like completely debug it and figure out like, well, what happened here? And you always have to go through the entire learning curve of how to learn automation, how to learn reading structured that canvas view in order to figure it out.

所以我们已经能够观察到，在机器人处理事情时，传统上会导致工作流程中断的问题，因为它们有点过于刚性，对吧？比如，如果您将工作流程连接到Google电子表格，然后您团队中的某人更改了电子表格中列的名称，传统上像一个“zap”就会中断，陷入困境。您会收到一个错误电子邮件，然后不得不回来完全调试它，并找出发生了什么事？您总是不得不经历整个自动化学习曲线，学习如何阅读结构化的画布视图，以便弄清楚。

And one of the cool things with these natural language systems is they can sort of debug automatically and sort of loop and fix it for you. Alright. So this is giving us a preview of the message. Let's take a quick look at that. There are a couple links in here. Those don't actually look quite right though. Those don't look like they're slack formatted. I think those they're rendering nicely. Just from experience, I know that's not the slack link format. So let's go ahead and tell it to. Can you use the slack message format? Let that get done. Oh, here's our automation quote. Automation is like a good cup of coffee, wakes you up with the possibilities of the day. All right. Can you use slack message forming? We'll give that one more step and see if this is able to take our feedback, which by the way, this is also another very important part of what we found from users is they want to give these bots feedback and say, hey, you did this good, you did this bad. Here's how to fix it and correct it.

这些自然语言系统的一个很酷的功能是它们可以自动进行调试，并且可以循环修复错误。好的，所以这给了我们一个消息预览。让我们快速看一下。这里有几个链接。但是它们看起来不太对。它们不像是slack格式的。我认为它们没有渲染得很好。从经验上看，我知道那不是slack链接的格式。所以让我们告诉它。你能使用slack消息格式吗？让它完成这个任务。哦，这是我们的自动化引述。自动化就像一杯好咖啡，让你在一天的可能性中清醒过来。好的，你能使用slack消息格式吗？让我们再进一步，看看它是否能接受我们的反馈。顺便说一句，从用户那里我们也发现另一个非常重要的部分是，他们想要给这些机器人反馈，告诉它们，“你做得好，你做得不好。这里如何修正它。”

One of the cool things is these bots can actually learn from the feedback instructions in order to update how they run in the future. So you can see this one actually choosing to use a tool called updating instruction. So this is going to take my feedback that I actually gave it, feed it back into the sort of instruction. So the next time when this is running live and sort of completely autonomously, it's able to use that instruction to actually work. One other maybe fun bit of trivia, why name these things bots? When we were looking at some of our usage data of Zapier Classic over the last 10 years, one of the major themes that we saw is people tend to call their zaps bots.

Over a million people have named their zaps with the word bot in it. So we decided to steal that name too. It seemed like a pretty good one. It seems like a very colloquial way to refer to what these systems are. Okay, those links all look much better. That's using the right formatting there. Yep. All right. And the very last step, hopefully, is going to be yes and go. Can talk a little bit about the roadmap while we're waiting for this one to complete.

超过一百万人在他们的zap中使用了“机器人”这个词。所以我们决定也把这个名字拿来用。这似乎是个很好的名字。这似乎是一个非常口语化的方式来描述这些系统。好的，那些链接看起来都好多了。格式使用的很正确。是的。好的。在等待这个任务完成的时候，我们可以稍微谈一下路线图。

I think one of the biggest lessons learned we've been getting real feedback from on this stuff is how much consistency and reliability matter for AI automation. One of the big feedback learnings we're seeing is, I think I showed it to you in the thread, where we give feedback to the bot natural language. This is one of the biggest themes and trends we're seeing with how users interact with these systems is they almost think about these individual bots as their own individualized like training. So the way they give them feedback, when they give them thumbs up, when they give them thumbs down, when they type them feedback, there's a very strong perception that, hey, I'm sort of making my individual instance of the software significantly better.

我认为我们从这些真实反馈中学到的最重要的教训之一是AI自动化中的一致性和可靠性有多么重要。我们正在看到一个重要的反馈是，我想我在对话中向你展示过，我们向机器人提供自然语言的反馈。这是我们看到的用户如何与这些系统交互的最大主题和趋势之一，他们几乎把这些个别机器人看作是自己个性化的训练。因此，他们提供反馈的方式，当给他们点赞、当给他们点踩、当给他们输入反馈时，有一种非常强烈的感觉，说：“嘿，我正在显著改善这个软件的我的独立实例。”

So that's one of the big things on our roadmap over the next couple of weeks is to be able to incorporate and take all of that feedback and actually make these things better online. And there we go. It's sent to show up Slack bot. I'll render the links a little bit funny, but there we go. It's after your assistant from 218. Okay. So that is a very quick look. Let me see if I go back to the behavior and pop it open. There we go. Here at the end, use the Slack message formatting summary for sent to the user is that sort of updating process where the bots are able to incorporate feedback right back into their instructions.

因此，在接下来的几周里，我们在路线图上的一个重要内容是能够整合并利用所有的反馈意见，实际上让这些在线内容变得更好。然后，它被发送给了Slack机器人。链接可能会有点奇怪，但这就是情况。这是从218年开始的你的助手。好的。这只是一个非常快速的查看。让我看看如果我回过头去行为并打开它。在这里，在结尾处，使用Slack消息格式概述发给用户的内容就是那种更新过程，其中机器人能够将反馈意见直接融入他们的指令中。

So that's Zapier and Bots on the nutshell. I think one of the cool things is I just showed a demo with two apps on Zapier, but obviously the magic of Zapier is we support what's the number now over 7,000 integrations on Zapier and Zapier Central works with all 7,000 out of the box. So you can sort of connect any pairwise combination you want or even have these bots do multiple actions at once. This is not a fan or believer in wait lists.

这就是Zapier和机器人的精髓。我认为最酷的一点是，我刚刚在Zapier上展示了两个应用程序的演示，但显然Zapier的魔力在于我们支持超过7,000个集成，并且Zapier Central能够与所有7,000个集成直接配合。因此，你可以连接任意两个你想要的组合，甚至让这些机器人同时执行多个操作。我不喜欢等待名单的粉丝或信徒。

So this is available today. You can go to central.zapier.com and log in with an existing Zapier account or sign up for a free Zapier account and give it a go and excited to see what you all build with Zapier's new ad bots. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you.

所以今天这个功能已经可以使用了。你可以前往central.zapier.com，使用现有的Zapier帐户登录，或者注册一个免费的Zapier帐户来尝试一下，我很期待看到大家如何利用Zapier的新广告机器人进行创作。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。谢谢。