This Next-Generation AI Agent Creation Platform Leaves GPT in the Dust

This State-of-the-Art AI Agent Creation Platform Is Transforming Customer Service

This Revolutionary AI Agent Creation Platform Is Set to Disrupt the Industry

In this article, I’m exploring a groundbreaking new AI platform that no one is talking about — meet Zapier Central, the next big thing in AI automation. While recently all of the buzz has been about custom GPTs, Zapier has been silently working away on their answer to OpenAI’s agents, and an answer that could mean trouble for the entire GPTs ecosystem. But could Zapier Central really be the future of how businesses adopt AI, or is it just another stale automation platform that struggles to create real value for its users? And more importantly, if it is any good, how can you use it to boost your business or the businesses of your clients with AI?

After diving into Zapier Central over the past few days, I have discovered that the Zapier team is definitely onto something here. In this article, I’ll be breaking down what Zapier Central is and what makes it a potentially better alternative to OpenAI GPTs in many use cases. Then, I’ll be whipping up some live builds to demo some of these super simple yet stupidly powerful use cases that anyone can create on Zapier Central. Finally, I’ll be sharing with you how you can become one of the first experts on this new platform to boost both your business and your clients’ businesses, regardless of your skill level.

We strongly recommend that you check out our guide on how to take advantage of AI in today’s passive income economy.

What is Zapier Central?

Here on the site for Zapier Central, they say that you’re able to work hand-in-hand with AI Bots. The easiest way I found to explain it is a mixture between GPTs, where you’re able to interact with a conversational assistant or an agent and tell it what to do, and then it can call custom actions and use knowledge, etc. You’ve got the knowledge, the actions, and the prompting — the usual GPT stuff that we’ve talked about quite a lot. But what Zapier has done is a very interesting aspect of also including some of the autonomous automation and workflow automation steps that they’ve done on the platform prior. It’s a unique combination of GPTs mixed with a centralization of all of your Zaps for a particular function.

Scrolling down a bit, the vision that they have for Zapier Central is the next step towards evolving automation for the AI age. Central puts you in the driver’s seat to automate even more work with the help of our friendly AI Bots. What Central really is, is a conversational layer that allows you to utilize all of the thousands and millions of possibilities that you can create with Zapier Zaps and the integrations they have with over 6,000 different apps. Extremely exciting stuff.

Live Data, Instructions, and Behaviors

Here’s a little bit of a demo. You can connect live data, which is similar to knowledge for your GPTs. As you can see here, you can add a data source. There’s Google Sheets, Google Docs, and also Notion as options. The interesting thing about this that you’re going to see in a second is that it is actually dynamic. So whatever the Google Sheet is updated, you’re then going to have access to that updated data, which is something we haven’t really seen — knowledge bases that are dynamic and regularly updated with the latest information.

Live data is one part. Then we have the instructions or the prompting. When they open this up, there’s an instructions tab which instructs your Bot on how to use the knowledge and also how to use the actions. Finally, they have what’s called behaviors, which are best understood as actions for custom GPTs or tooling, etc. This, of course, is directly integrated with all of Zapier’s thousands of integrations.

The Biggest Advantage for Zapier Central Over GPTs

As you may have noticed, these are the same three ingredients that I’ve been banging on about for the past couple of months. If you can figure out this AI agent skill set, which right now is knowing how to prompt, knowing how to add data, and knowing how to connect these tools, then you’re able to build AI agents on basically any of these platforms because it’s the same skill set. I hope you’ve been paying attention and have built up these skills of knowing how to prompt, knowing how to connect different actions, because it’s the same process here — tying all of these three aspects together into these AI agents that you can use internally in a business.

The biggest advantage for Zapier Central over GPTs, in my opinion, especially for beginners, is that there is no code involved at all. You don’t have to add in these tools separately or use a schema. I’m sure if you’ve tried to get Zapier actions to be put as custom actions for your GPTs, you’ll know just how difficult that process can be. With Zapier Central, there is no code involved at all. All you need to do is go on as it shows here, click on “Create new Behavior,” give it some instructions, set up a trigger, and then set up what it’s supposed to do. You’re able to create these awesome, powerful custom actions without ever having to write any code at all.

Down here they’ve got a little bit more stuff. I think this all-in-one AI workspace is really the key point here. It centralizes all of your automations and AI features that you use throughout the business into one hub, which I guess is why they called it Zapier Central. It’s a pretty fitting title.

Live Demos of Building on Zapier Central

If that hasn’t got you excited about building on Zapier Central, then the next section where we’re going to jump into some live demos of building on Zapier Central will do the job. I can click on “Try Central today.” Of course, you need a Zapier account for this.

Once we’re in Zapier Central, we have this familiar kind of chat GPT interface, and it immediately gives us a new untitled bot. It tells us, okay, we can add behaviors, we can add data sources. A good starting point for us, I think, would be to show some of the live data and being able to chat to spreadsheets and CSV data.

Querying Spreadsheet Data

If I add a data source and go to a Google Sheet, it’s connected by my email to all of my Google Sheets. Super easy to do that. Now I’ve got this sample sales data that I’ve got off Kaggle just to show you how even large datasets can be used, and we can query our datasets to get some interesting insights.

To give you a quick look at the data, there’s a ton of stuff here. There’s the sales, there’s the quantity of orders, there’s the date — most interestingly, which I want to do a little bit of analysis over. Let’s see, there’s a lot of

2003, 2004

dates. Let’s give it a bit of a tricky question to analyze that information.

I can say, “Can you get the sales by month for 2003? Include the number of units sold and the total value month by month.” What it’s doing here, similar to ChatGPT, is that it has its own code interpreter. As you can see, it’s actually converted the spreadsheet into a database, and it’s writing a database query on this. This is some really impressive stuff, and it’s been kind of a fantasy for a long time — being able to convert a spreadsheet into a database. Now we’re writing SQL queries, going from text to SQL. I’m sorry, I’m getting a little bit nerdy here, but it is very impressive what they’ve been able to do here with converting it to a database. Now we can query the database and get some really powerful insights.

Here you can see the output of the data retrieved, and then it’s using that data to write the answer. Here we can see the data for 2003, month by month, units sold, and total sales value. Very, very cool stuff already.

I can say, “Can you compare that month by month to 2004? E.g., Jan 2003 vs Jan 2004.” This capability alone — being able to chat to spreadsheets in a way that’s very dynamic and can pull in literally dynamic data as it gets added to the spreadsheet — is an awesome new use case. I really recommend you guys jump in and have a play around with this because this is something that businesses have been wanting to do for a long time. Having it put on a platform like this for us is some awesome news for beginners and pros alike.

Here we can see again, it’s written a query for the database it’s created for our spreadsheet. It’s done it perfectly. It’s comparing the 2003 values month by month with the 2004 values. Dynamic spreadsheet data added as a knowledge base is so much better than the knowledge bases we have for GPTs. I’m sure you have tried to put CSV data or spreadsheet data as a knowledge document for your GPTs, and it’s just absolutely terrible at using that data because it doesn’t have a system setup like this to be able to create a database out of it, then query the database and use that intelligently in the answer.

That’s how we can chat to spreadsheets with dynamic data, which is cool enough. Then we have the typical chat to documents in the same way that you’d upload a document to a knowledge base, but again, it’s going to be dynamic.

Querying Google Docs Data

If I go to Google Docs and select a Google Doc here, maybe the support documentation for my company would be a pretty good one for this. Because if I update the support documentation, I don’t have to go in and re-download it and re-upload it into my agent. It’s just going to automatically do that. I can get my team to go in there, modify the support documentation as necessary, add new questions and answers and FAQs, and it’s going to automatically be pulled through to Zapier Central here.

Interestingly, this brings us to some of the quirks of this platform in its current form. As you can see here, I’ve uploaded the information. I’ve said, “How do I change my password?” which is included in the support documentation. Just in order to change your password, however, it’s not developed by Zapier, and it just starts making it up based on its existing training data.

This standard chat interface is not very useful in many ways, and in order to get it to do the things you want to do in many cases, you’ll have to go to “Create Behavior” and say, “Answer questions from the document provided.” Then if I go to “Trigger” and go “When I message the bot,” add the trigger, turn that on, and now if I go back and ask how to reset my password, finally, after a pretty long wait, I can click onto this thread and see that it has gone through and added the code block to select the document first of all, and then to answer the question from the information in the document.

It doesn’t quite work how you’d expect it in these simple use cases, but I believe, from my playing around, this is set up this way so that when you have tons and tons of documents, like 20 or 30 of them included, this system of using a lookup and looking up in the database to find the specific documentation file that you’re looking for or the knowledge base document you’ve provided, it’s going to actually look that up first and then put it into this thread.

So when the actions are performed, this is the interesting part where you combine a regular GPT style with what I think Zapier is trying to do, which is this threading style. Now I can converse back and forth within this thread and ask questions about my document, and I’m going to get the answers in there.

For dynamic knowledge bases, you can also connect Notion here. I have had some issues with getting it to answer correctly, so I think there might be a few bugs, but I’m sure they’ll buff those out in no time.

Creating a Notion Assistant

Here I’ve made a bot called “My Notion Assistant,” and what I want to do is set up some behavior in this. “Create a Behavior” — instructions are a good place to start. I could say, “Create a new Notion page when I say ‘create a new Notion page.’ Ask me for the content for the page, but come up with a page title yourself that represents the content appropriately.”

Now, usually, you’ll be getting suggestions here. I’m not sure why it’s not suggesting to me. This can be a little bit confusing, but just stick with me here. “When I message the bot” is the trigger. Then I want to use my phrase here, which was “create a new Notion page.” I can provide a specific keyword or phrase that I want to trigger this action: “create a new Notion page.” I want to add that as a trigger.

Then I want to create a new page. Here are my suggestions here. That’s nice. For the parent page, I want the AI to select this. This is really the magic of Zapier Central, which you’ll see a bit more of, but we have the title. I want the AI to generate the field for the value, and I also want it to come up with the content based on my input.

That’s all set up. I can test the behavior now. It’s going to open up a thread here. “To proceed with creating a Notion page, I need to know what content you’d like to include in the page.” Here you can see it’s triggered. This is the test trigger that it’s just inserted the phrase that I’ve included in the setup.

Let’s jump back to my spreadsheet chat and I want to put this data into a new Notion page. Let’s see if it copies across. Okay, so I have to go back into the thread here. Oh, let’s see how it goes with that. We got a little bit of a bug here where it didn’t actually send me a reply. Normally it will send back the link to the Notion page itself, but I followed up, and it seems like it’s already created it.

Sure enough, over on my Notion, it has created this. It’s given the title itself that it’s generated based on the content, and I’ve got a Notion page. Maybe it was because it was in the testing environment.

If I try again, “create a new Notion page,” try it again without the testing environment. Now it’s going to pick it up, create another thread, and see it started talking about some random stuff here. That’s because I didn’t come here and switch the action on. That’s my bad, apologies.

If I want to call this again and try to get it to properly create the Notion page, I’ll say, “create a new Notion page,” create a new thread for me. Hello? Seems to be bugging out here. Let me refresh the page. Maybe a little modification here.

I’ve just changed it to say “Trigger when I say ‘create a new Notion page.’” If I go “create a new Notion page,” there we go. So make sure in this, you use “trigger.” It seems to be the keyword that gets it to work well. “Trigger when I say ‘create a new Notion page.’” Now it’s going to be asking me in the thread for the content I want to include.

Let’s grab the sales data again, head back to the Notion assistant, go on to our thread, and paste it in. It should give me the confirmation message that it’s created the page successfully. “Proposed page title.” Now it’s going to create the Notion page. “Please hold on for a moment,” and it’s performing the action as expected.

There we go. If we drop this down, it has included all the information that it used to create it, and it provides us a link to the report here. If I click on this, I can open up my Notion, and there we have it.

Strange thing is that it’s kind of just putting it anywhere. I think if I go in here, it’s down here, and this one is inside the other one I just did. So back here, if we look at the “Notion Create Page,” I can’t even modify what the parent page is here. It’s not perfect by any means, but it does show you that this is a GPT-style experience with these thread mechanics that allow you to set up certain behaviors to happen upon a certain phrase being sent.

Combining GPT-Style Interactions with Background Automations

This particular human-in-the-loop style interaction with the assistants is not all Zapier Central does. It’s intended to combine both the GPT style where you are interacting with your agent, with your typical background workflow automations that you set up on Zapier, which might be triggered by some action like a Slack message or a Google Sheet row being added.

Here’s an example I’ve been working on, just having a play around scratching my own itch with this stuff. It’s a community control center. I have my community, and I thought, “Why not try to create some kind of central hub for managing it and put Zapier Central to the test?”

I have a link to a spreadsheet that is synced 8 minutes ago, so it’s actively re-syncing the data. I’m always able to ask questions about my community. But more interestingly, I’ve been able to set up some behavior where, when someone tries to join my community, a new row is added to the spreadsheet that it’s watching, and that is the trigger.

No longer am I having to ask a message to my assistant to trigger it. It’s able to use a traditional Zapier trigger. Then I’m getting it to send an action after that, which is to send a direct message. The magic of Zapier Central is that the AI itself, as I said, is a layer on top that can connect all of these things.

The instructions here will tell the AI how to autofill the rows. As you can see here, for the message text, I need to generate some kind of message that I want to send to myself on Slack. But what I’m telling it by my instructions is, “When a new row is created, send a Slack message to me that is a one-sentence summary of the new member’s details.”

If I go over to my Slack here, you’ll see that Zapier has been sending me messages, and these one-sentence summaries — obviously, I’m going to have to blur these out, but I promise you that there are one-sentence summaries as I told it to. Over on Zapier Central, it’s triggering the automation based on a row being added, and then it’s using the AI in the instruction to determine how it fills out the rest of it. It’s like this magical glue that can stick your automations together without having to get this direct lining up and matching up of the inputs from one to another.

As I said, this combines the conversational aspect that we’re used to with GPTs with these kind of background and autonomous automations. If I go to “While I was away,” we had 17 runs. Now I get to see everything that’s been happening in the background of my community while I’ve been asleep, while I’ve been away. I can keep tabs on all of the automations that are firing. I can set up new ones.

#### Scheduling Automations

Another cool thing you can do is go to “Create Behavior.” I can set a trigger to schedule. If I want to schedule something that I want done every week, which might be to look at all of the spreadsheet rows that have come in, all the new members that have tried to join, do some analysis, and then send the analysis via email to whoever or to send it in here.

There’s so much cool stuff that can be done now that we’ll be able to combine both the human-in-the-loop aspect of AI agents and talking to them and telling them what to do with these backend and background automations. That’s really the magic of what Zapier has done here with Central, and I’m a big fan.

### Limitations and Areas for Improvement

There are a few limitations that the platform is suffering from. I think the conversational aspect of pulling in context from the threads and answering questions correctly — there’s still a lot of work to be done there. It is quite slow. I’ve had to obviously cut it out for this article, but some days I’ve been sitting here for 30 seconds or more waiting for it to trigger an action or something like that.

There’s definitely room for improvement on the speed side, but this is quite clearly intended to be an internal use kind of platform and tool for managing your systems, managing your operations, managing the different departments, and centralizing. Like the community example, centralizing everything to do with my community — all the automations, all the reporting, all the questions I want to have over the data. It’s a really, really cool way of centralizing things.

It’s not so much as we’ve been familiar with with other agents of maybe creating a GPT and deploying it somewhere, putting it out on an Instagram channel or putting it on WhatsApp and things like this, because it doesn’t have to be interacting with customers. I think the speed can kind of be forgiven if you’re using it internally for now. Not terrible, but could definitely be better.

One thing that would make this a lot more powerful is being able to visualize the data. It’s currently not able to, like ChatGPT, use the code interpreter to generate a graph. So when I was looking at the data, I would have liked to see the 2003 line against the 2004 line. It’s a lot easier to understand and do data analysis when you can actually visualize it as well. I’m sure they’ve got that on the roadmap, but it doesn’t have that right now.

### Where Does Zapier Central Fit in the AI Automation Landscape?

Now that we know Zapier Central has got some pretty cool stuff going on, the question is, where does it fit into the AI automation landscape? How can we adopt it? How can we include it in our own businesses? How can we use it and sell it to our clients?

To me, I think Zapier Central is going to play a key role in the space in a number of ways. Firstly, it has completely removed the need to code in order to create these AI agents. We used to have that integration step of, okay, have a custom action or have an API that I want to put onto my GPT as a custom tool or a custom action, and there was this ugly OpenAPI schema thing that beginners had to do that wasn’t very beginner-friendly in many ways.

What Zapier Central has enabled us to do is to directly hook into all of Zapier’s apps and the actions and triggers that you can use, seamlessly, no code. Not a single line of code has been written. This could really be the democratization of AI agents in many ways. Now people can come in without any coding background, and they can start to build these things either for their own businesses or sell building them as a service to other people as well.

When you factor in that Zapier is making major moves on things like Zapier Tables for storing data, they have Zapier Interfaces, they have Zap Canvas as well — their ecosystem is growing at a rapid rate. This means if you are someone who’s looking to get into AI or building these kind of automation systems and beginning to make money by making businesses better with them, Zapier may be an interesting opportunity for you to focus on just one ecosystem.

There is so much available in this AI space right now. If you’re an AI agency or you’re a freelancer, just picking one platform like Zapier and saying, “Look, they’ve got so much stuff. I’m just going to specialize in that,” that could be a great place to start, especially if you’re a beginner and you want to utilize no-code tools as much as possible.

### Try It Out for Yourself

If what you’ve seen today seems interesting at all, I highly recommend you go on to Zapier Central and check it out. Try to build just one thing. The best way to learn this, at least in my experience, is to scratch your own itch. Try to build something that’s handy for you in your own life or for a friend or for your parents or something. Just try to find a use case that you can apply this tool to, to help make your life better. That “aha” moment might just be the thread that you end up pulling to a particular career in this AI automation space.

I’m super excited to see what kind of stuff you guys are building on Zapier Central and being able to find better use cases than I have so far. If you’re excited to get stuck into Zapier Central and have got something out of this article, please hit down below and leave a like on it. It really supports me to continue making free content like this.

Of course, if you want to stay up to date regarding future updates on Zapier Central, how to build GPTs, and other tips and tricks that I’m finding for you to make money and build businesses in the AI space, be sure to subscribe while you’re down there so that you don’t miss any of my future content.

If you’re interested in learning how to start an AI business in 2024, I’ve just released a full guide, an hour and a half long, that I highly recommend you watch. But aside from that, guys, that’s all for the article. Thank you so much for reading, and I’ll see you in the next one.

We strongly recommend that you check out our guide on how to take advantage of AI in today’s passive income economy.