Zapier AI Agents: The Future of HR Automation (Workshop)

All right. Hey everybody. It sounds like we are live. I am so excited to get to spend time with you today as you filter

in. If we haven't met before, uh my name

in. If we haven't met before, uh my name is Emily. Maybe. My pronouns are she

is Emily. Maybe. My pronouns are she her. Today I am wearing rainbow glasses

her. Today I am wearing rainbow glasses a black shirt, I have long brown hair and I am here with you to talk about Zapier agents specifically from the

perspective of somebody who works in a people role, an HR role. And today we are actually going to be building together. So by the end of our time

together. So by the end of our time together, a very short period of time you're actually going to be able to have deployed an AI agent using Zapier. I am

with you not as somebody who's super technical. I'm not an engineer. I'm here

technical. I'm not an engineer. I'm here

with you as a senior learning designer from Zapier. So someone just like you on

from Zapier. So someone just like you on the people team who knows that our roles are technical. They just maybe haven't

are technical. They just maybe haven't traditionally been viewed from a technical lens. So let's get started

technical lens. So let's get started with talking about what we're going to cover today. Our agenda for the day is

cover today. Our agenda for the day is starting with talking about what agents are, where they fit into this AI and automation uh continuum. Then I'm going

to share with you a use case example, a story of success from HR and people automation. Give you a quick overview of

automation. Give you a quick overview of the tool we're actually going to use.

And then we're just going to build. And

at the end, there will be some time to answer questions. So, if you find

answer questions. So, if you find yourself uh wondering about something in the midst of our build, go ahead and log it in the Q&A portion of the uh the

screen that you're on with me. And for

starters, I want to make sure we tackle what agents actually are because today we're going to be building one. On your

screen, there's this spectrum. On the

left hand side, which if you can see the colors, is the more orange side. we see

something called determinism. And on our long spectrum, determinism means something that happens the same way, the way that you've determined every single

time. So, it's going to be predictable.

time. So, it's going to be predictable.

It's going to execute identically every time. If you're a Zapier user or you've

time. If you're a Zapier user or you've used automation before, determinism is something that looks like when a new email comes in, then I get a Slack

notification. There's no question that A

notification. There's no question that A will cause B and it will look the same every single time. All the way over on the other side of the screen on the

right hand side, the purple side, we see inference. So think of this as a long

inference. So think of this as a long spectrum.

Inference is the opposite of determinism. It is going to happen a

determinism. It is going to happen a little differently every time. There's

creativity or thought that will influence it and cause your outcome to change but also the process might change. And that is where agents live.

change. And that is where agents live.

You can see on the screen there's a picture. Agents they have a workflow but

picture. Agents they have a workflow but they're making decisions within that workflow and it's going to change the process and the outcome. Now, Zapier has

powerful tools that can make any of these possible from deterministic workflows to AI in the workflow, a workflow with some agentic properties all the way to building an agent like

we're going to do today. The important

thing to think about is that agents are autonomous. So, we're going to give it a

autonomous. So, we're going to give it a prompt. We're going to give it access to

prompt. We're going to give it access to some tools, but then the agent is going to decide how to accomplish that goal within the parameters you set. So, you

can do things like have it search the web or give it access to live business data to query. And because agents are not deterministic, that means they'll

run a little differently every time. And

that's great for work that involves some kind of thinking or analysis like writing reports or generating content or researching your competitors. Those are

the features and functionality that we're going to lean into together today.

So what is an agent? Simply put, it's a software that perceives its environment.

It makes decisions and it takes actions to achieve goals without the need for humans to hold its hand. all along the

way. And Zapier agents, Zapier being the

way. And Zapier agents, Zapier being the company I work for and Zapier agents being the tool that we all use. That's

an example of an AI agent builder. In my

opinion, it's like the easiest one to use on the market. That's my bias. With

Zapier, you can create as many agents as you want using plain natural language.

And you're going to give it one role.

So, think of your agent you're going to build today as like one colleague doing one job. You'll give it access to your

one job. You'll give it access to your apps and your data, and then you'll give it some instructions to do work for you.

Now, I'm going to pause for just a second. If you haven't signed up for a

second. If you haven't signed up for a free trial with Zapier yet, that's the tool we're going to use today. So, go

ahead and do that. Uh, I think our fantastic Leah is able to drop the link for Zapier agents if she hasn't already.

Uh, but we're going to make sure everyone's building in that same environment today. So, if you haven't

environment today. So, if you haven't had a chance yet, go ahead and sign up.

Now, I mentioned that agents after you give it a role, it's going to access your apps in order to do actual work with Zapier. Uh, some of the magic here

with Zapier. Uh, some of the magic here is that agents have access to more than 8,000 apps that you use at work. So, in

a minute here, we're going to set up our agent to connect with tools. And when

you do this for your own work needs, you could have it have access to tools like Slack, Bamboo, Teams, Gmail, Notion Koda, Workday, Asana. Zapier's products

like agents, they can connect those apps so they work together and do work without you having to be involved. So

that saves you more time for the most important work of people-to-people interaction.

Now, I've given you a whole bunch of information, but let's take a look at how connecting AI agents to our apps can solve real problems in HR and people

spaces. So, I'm going to show you a real

spaces. So, I'm going to show you a real agent that I have now.

There's a lot on this screenshot, but this is an example of a Zapier agent HR use case. So, what this agent does is it

use case. So, what this agent does is it creates a custom new hire welcome gift box for each new hire. You can see there's no code on your screen, right?

This agent is using natural human language that I wrote and a few tools.

In this case, you can see up towards the top, Gmail is connected. And with just that amount of information, it's acting as an autonomous teammate. Now, how did

I build it?

The first thing I did, and this is the same picture just zooming you into a specific area. First thing I did is I

specific area. First thing I did is I asked myself what should trigger the agent. So that looks like filling in the

agent. So that looks like filling in the blank in a sentence. When blank happens then blank. Your first blank, the when

then blank. Your first blank, the when blank happens is your trigger. So in

this case, when someone responds to an offer email then something happens. So again, using natural, language,, we're, going to, fill, in

that blank. When someone responds to an

that blank. When someone responds to an offer email, then read their explanation of their preferences, their likes and dislikes, and browse our giftgiving

website to curate a custom welcome box for them. It's as simple as that. So

for them. It's as simple as that. So

again, you're going to see on the screen there's absolutely no code. We're going

to use some connections, some buttons some typing, but we're going to do it all with natural language using what I think is HR and people's superpower, our

incredible skills of communication.

So, with that in mind, it is building time. We are going to jump into the

time. We are going to jump into the product and we are going to build our Zapier agent in just a few minutes. If

you're someone who's working with two monitors, I'm going to encourage you now. This is a good time to set up one

now. This is a good time to set up one monitor with the window of me so you can see me, our video stream. And then on your other monitor, that free Zapier

trial account with agents or your Zapier account if you've already got one. That

way, uh, you can see what I'm doing. You

can follow along. If you're just working on one screen, that's totally fine.

Maybe make a side byside view. The

problem that we're going to solve today with our agent is a universal problem.

How do we maintain a record of our performance impact across the year? And

if you were in Brandon's recent HR by Zapier uh working session, you might have seen him preview a very similar agent.

Let's go ahead and get into agents. So

I'm actually going to navigate over to agents. My screen's going to look a

to agents. My screen's going to look a little different than yours because I build agents all the time, so I have dozens on my screen. Yours is probably going to have fewer or none. But

importantly, from the agents web page you should see a button in the upper left or the upper right that allows you to create a new agent.

And once again, we're going to use HR's incredible superpower here. Our

incredible superpower is our communication. We're going to

communication. We're going to communicate just like we would to a colleague that we were asking to do a job. What's our job? When blank then

job. What's our job? When blank then blank. For me, I work in a Slack work

blank. For me, I work in a Slack work environment. I get feedback in Slack

environment. I get feedback in Slack constantly all across the year, but I need a way to capture it and make that meaningful when it's time for performance reviews. Let me quickly show

performance reviews. Let me quickly show you. Here's a Slack channel where people

you. Here's a Slack channel where people are giving me feedback. And I want to make sure I don't lose this. This is so valuable for me to reflect on at performance review time. So when I talk

to my agent, the workflow I'm going to describe is when I react to a Slack message with a specific

reaction.

Then I want you to capture the feedback I've gotten on my

performance and I want you to store it in a Google doc. That way when my performance review

doc. That way when my performance review time comes up, I can simply reference that Google doc with my manager to have a scope on the whole year's progress as

far as feedback I've gotten. I'm going

to go ahead and with just this one sentence, I'm going to click create my agent.

You can see a new screen here and it's got some different information than we saw before. It's got a co-pilot here on

saw before. It's got a co-pilot here on the left. It's got instructions here on

the left. It's got instructions here on the right and it's got a title. This

time it named itself Slack Feedback Collector. Now, if you're someone who's

Collector. Now, if you're someone who's not working in Slack, you're working in a different tool for communication at work, you might have chosen a different tool, uh, or you might be building along

with me and using Slack, but just know that we're using Slack. In your mind you could substitute another tool that you might use. Perhaps you get most of your feedback in email. It would be okay

to swap these out. Now, when we're setting up our agent, you can see it's doing a ton of work for us behind the scenes. This is one of the magical

scenes. This is one of the magical pieces of working with Zapier's tools.

They have AI cooked right into them. So

for now, the agent is walking me through exactly what it's setting itself up to do. It's saying, "Hey, first thing I did

do. It's saying, "Hey, first thing I did was I built out your instructions. Now

I'm setting up your workflow trigger and I think it's Slack new reaction added." Now, you heard me. When I react

added." Now, you heard me. When I react to something in Slack, then I want you to catalog this feedback. So, I agree.

That is the trigger I want. And then

it's setting up a few new pieces for us.

But I'm going to ask you to focus for just a minute on this trigger right here. So when we build an agent, our

here. So when we build an agent, our trigger is what's going to start all of the behaviors to happen. Nothing is

going to happen without us interacting in some way. Whether that's uh triggering it by reacting to something whether that's getting a new email whatever our trigger is starts the whole

workflow. I'm going to set up that

workflow. I'm going to set up that trigger to make sure it's meeting my needs. And in this case, I want to make

needs. And in this case, I want to make sure of a few things. First, I'm going to authorize my Slack account so that it has access to my work environment. If

you haven't used Slack and Zapier yet you might have to set that up yourself right now. But if you have used Zapier

right now. But if you have used Zapier and Slack, you'll see your Slack account in the drop-down menus here. And you can just find your account and connect it.

So when I skim through these, I can see other co-workers here. I want to make sure to choose Emily maybe at Zapier.

And then I'm going to get a blue text field here that says show all options.

It's actually a selector. When I click on that, there are a few more pieces of information that my agent could ask for.

So, first my Slack account.

Then, oops, sorry, I shouldn't have clicked around and gotten fancy. Then I

need to choose which emoji reaction is going to trigger this agent. Let's head

back to Slack for just a minute so I can show you what I mean. Here's a piece of feedback that Cecilia left for me and I want to make sure that next year when I'm in my performance review, I can

reference it. I'm going to pick one of

reference it. I'm going to pick one of the emojis in my Slack environment and this is going to be the trigger for this agent to start running. In this case

I'm going to choose we have an emoji for feedback.

That's going to be the emoji that triggers my agent. So, I'll go back into my agent and I have to find that particular emoji. There it is.

You're going to look at your Slack environment and your emoji selections and pick one that you'll remember to always use to trigger this agent. So

ours might not look exactly the same here.

And then here's your first tip or trick of our session.

There are pieces of information that you don't need to give your agent because it'll look at all the data coming from Slack and it will make choices for itself. One piece that I do recommend

itself. One piece that I do recommend that you update is make sure to choose your own self as the person who triggers

this workflow.

Sometimes when you build an agent, if you use this particular trigger, the new Slack reaction added. There are times where you might want anybody to be able

to trigger it. I'll give you an example.

Hey everybody, I'm having this great HR professional development event. Anyone

who reacts to this Slack message will get an invitation. In that case, anyone who reacts to the Slack message, I would want to be able to trigger the workflow.

But in this case, I want to make sure I'm only storing feedback that I want to capture feedback for me. So, if anyone else uses this emoji in our workforce, I

don't want it to trigger. So, I'm going to go ahead and choose myself and my member ID. and I happen to know uh that

member ID. and I happen to know uh that I need to let a lot of um people load for a second. You can see 3,000 accounts before I see myself. If you have a

really really big Slack organization you might need just a minute to let your member ID load. But you're going to go ahead put yourself in as the member ID to select.

And then we don't need to fill in any of the other fields. So in summary, three fields. Number one, your Slack account

fields. Number one, your Slack account authorization.

Number two, oh, I didn't know I could do that. Number two, the feedback or I'm

that. Number two, the feedback or I'm sorry, the reaction that's going to help trigger your uh workflow here. And

number three, yourself as the member ID.

And importantly, we're going to hit save.

And now the other piece of the puzzle right? We've got our trigger and then

right? We've got our trigger and then we've got our instructions. This is sort of the the most important part. It's the

body of our agent, what it's going to understand that we want it to do. So

I'm going to read this and I'm going to adjust it to make sure it's meeting my needs. Let's read it. When a specific

needs. Let's read it. When a specific reaction is added to a Slack message capture the feedback content from that message and store it in a Google Doc.

That's, sort, of, the, summary., All right.

Step one, monitor for when a specific reaction emoji is added to the Slack message. Great. Step two, extract the

message. Great. Step two, extract the text content from the reacted message.

This will be the feedback. I'm going to give it a little bit more information here. Sometimes in Slack, uh there could

here. Sometimes in Slack, uh there could be a message with feedback for like six people in one Slack message, like compliments to the whole team. I want to make sure it's pulling out the stuff

that's relevant to me. Uh so I'm going to say this will be feedback for Emily.

Emily, maybe.

Next, it says get additional context like who sent the original message, the timestamp, and channel information.

That's fine. I would love for it to build more context. Then format the feedback entry with relevant details date, sender, channel, feedback content. Now, you saw how short what I

content. Now, you saw how short what I wrote was, right? One sentence. And AI

took my prompt and is giving itself all of this information. Computers speak to computers. So, it's giving itself all

computers. So, it's giving itself all the context it needs to do a really good job. And I didn't even have to formulate

job. And I didn't even have to formulate this. I'm going to leave that just the

this. I'm going to leave that just the way it is. That sounds great to me. Now

here's the part where we're going to have to do a little bit more clicking.

It says, "Step five, append this formatted feedback to a designated Google Doc for performance tracking."

And then it inserted a tool. Now, if

your uh sort of AI co-pilot didn't insert your tool, what you're going to do is you're going to go to insert tools, upper right hand corner of the

instructions box, and you would either select Google Docs append text to document or you would click add tools and you would find the tool you're

looking for. So, for me, I use Google

looking for. So, for me, I use Google Suite. So, Google Docs, sheets. I also

Suite. So, Google Docs, sheets. I also

use Zapier tables to store information.

But if you're using a different product like Air Table, all you need to do is search for the tool you're looking for and you can select it and all of its actions. In my case, again, I'm going to

actions. In my case, again, I'm going to use a Google Doc. I can find all of the actions that my agent can take on a Google Doc. Uh, for instance, it could

Google Doc. Uh, for instance, it could find a document for me or create a new document. In this case, I want it to

document. In this case, I want it to append the text to a document. So

that's the tool I'm going to insert.

And I insert it right inside the instructions uh so that the agent knows when and where to activate it. Now

there's a little gear icon here that lets us adjust this tool. I'm going to click that and double check that it's choosing it's

using the document that I actually want.

In this case, I have a blank performance tracker set up to capture this information. And it's called the

information. And it's called the performance tracker document for HR demo. And it looks like that's the one

demo. And it looks like that's the one it's selected. But if it didn't, you

it's selected. But if it didn't, you could go ahead and search for any Google Doc where you want to capture this information. If you don't have one yet

information. If you don't have one yet because we're doing the demo live together, that's totally fine. You can

make one now. And same thing, you'll have to connect your Google Doc authorization just like you did with your Slack account if you've never used it before. But then you just search by

it before. But then you just search by name for the document that you're looking for.

You can leave all of the other fields blank because your agent is smart enough to fill in everything it needs.

And then you can click save.

Now I'm going to test this agent and see if it's working. And in order to do that, I'm going to press the test button in the lower right hand corner. And the

screen's going to change just a little bit. The instructions are going to

bit. The instructions are going to collapse to the left and there's going to be a testing window here on the right. So I'm going to press test in the

right. So I'm going to press test in the testing window and I can watch my agent think. So, the

first thing it's doing is it's looking for a piece of feedback where I used that trigger, that reaction. And I can see right in the testing window that it

found that feedback from Slack, that feedback from Cecilia that I reacted to right there.

And now it's walking me through its plan. It's going to append the feedback

plan. It's going to append the feedback and it's asking how does this look? And

again, the place it's going to put it is this blank document.

I'm going to click approve.

And then we'll see how it looks when it puts it in our document and make sure we like it. We can keep talking to the

like it. We can keep talking to the agent and adjust it before we turn it on live and make sure it's working exactly the way we want.

All right, let's check our document.

All right. So, it is showing us the way it's going to format the content, but I want to make sure it's actually processing the feedback.

So, let's go ahead and talk to it for a minute. Can you try with the actual

minute. Can you try with the actual feedback from Slack instead of placeholder content?

The nice thing about agents is troubleshooting doesn't look like looking at lines of code. It looks just like talking to someone in Slack.

Now I am going to do a little bit of adjustment here while I have you all live. There's something

that I've prepared in advance which is if you like me work at a company where our performance uh is tracked against

specific metrics like um a set of competencies or in our case we call them impact behaviors. You probably have them

impact behaviors. You probably have them documented somewhere and you can make this even richer by giving your agent context that's relevant to the company.

So in this case, I have this document with all of the impact behaviors that I'm measured against my performance at Zapier. And I want to give this to the

Zapier. And I want to give this to the agent and ask it to use it when it translates my performance into my tracker. So, I want it to read Slack

tracker. So, I want it to read Slack messages, compare it to those impact behaviors, and then when I get to my performance review, it'll be able to

look back at all of that information and I'll have it categorized by like how did I show up against my company's competencies. So, I'm going to go ahead

competencies. So, I'm going to go ahead and add one more thing into my agent here.

when I have it monitor for specific reactions and then extract the text content, I want to actually give it instead of having it look at the

original message timestamp and channel information, I'm going to say compare

the feedback in the in the Slack message text against our companies impact behaviors.

to help you summarize how my performance is showing up as impact. And then just

like we did before, I'm going to insert a tool again. This time, again, right there at step three, when I insert my tool,

I'm going to use Google Docs again because that's where my impact behaviors are.

And I'm going to use data integration which just means I'm going to integrate the impact behaviors right into the agent. It's going to act sort of like

agent. It's going to act sort of like its brain, like its knowledge source.

And I'm going to find my Google account. So, let's go ahead and use that. And the document I want to find is impact behaviors for this

workshop. A nice cleaned up version for

workshop. A nice cleaned up version for all of us.

I'm going to add that data source and you'll see a little yellow button that says it's syncing which means it's like kind of uploading all the information to its brain. If I ever go

change the impact behaviors I know at our company we revise them. The uh the sync will continue to happen. It will

keep syncing. So this can be dynamic data. It can be data that changes and I

data. It can be data that changes and I don't need to go like re-upload this document. The agent will still have

document. The agent will still have access to it. All right. And then I'm going to ask it to I'm actually going to simplify this even more. I'm going to uh

tell it to read the text feedback.

Compare the feedback in the message. Compare the feedback.

format the feedback with relevant information um and relevant details and then

connection to impact behaviors and again no code I'm just writing like a person and then append it to the Google doc for performance tracking. Now let's test

performance tracking. Now let's test again.

I'm going to click the purple retest button.

It will start the whole process over again using my Slack information that I reacted to.

You can see it's loading the content from that Google doc I gave it access to. And uh we can hear it like thinking

to. And uh we can hear it like thinking aloud. It talks to us as it works when

aloud. It talks to us as it works when we're doing the testing mode. Once we

actually turn this on, we won't see this again unless we come back and visit it.

It'll just work in the background anytime I'm using Slack.

All right. So, now it's analyzing against the impact behaviors or your competencies depending on what you connected. It's giving me a preview. It

connected. It's giving me a preview. It

looks like I need to do a little troubleshooting because it's still giving, me, placeholder, data., All right.

And it it is ingesting the information correctly from the impact behaviors.

Here's what I'm going to do. I'm going

to give it one more chance to make this even simpler. I'm going to say

even simpler. I'm going to say read the feedback compare the feedback, and then I'm going to ask it.

Yeah, add it to my tracker document.

Let's try making this even simpler.

Thanks for bearing with me. It's always

dicey, to, do, it, live., All right,, we're testing it one more time. Let's see how we do.

All right. Now it's summarizing our results.

Okay.

Oh, the live demo gods don't smile upon me today.

All right, we're going to try this one more time. We're going to do something a

more time. We're going to do something a little zany. Stick with me, everybody.

little zany. Stick with me, everybody.

We are going to try just turning it on.

I'm going to turn my agent on and I'm going to see if giving it real data to react to helps kind of nudge it along.

This is a safe use case for that. So

I'm going to go ahead and use my ReactG and I'll show you a little behind the scenes of one more thing, which is the activity tab.

So, behind the scenes on the activity tab, this is where you can do any troubleshooting or adjusting like you're seeing me do right now. Uh you can see

what it's working on. So, I just reacted today and we've got a uh a run of the agent in progress here. And if I check here,

neat, I can see the new Slack message that I just reacted to. And I can see it ingesting that. Now it's reading the

ingesting that. Now it's reading the behaviors, the impact behaviors we use our competencies. Um, and this is sort

our competencies. Um, and this is sort of like the uh interface you would use to go back if you've ever used Zapier's Zap workflows before. This is how you

would watch a Zap run, except it's the way you do it with agents. So, we go to our activities tab instead of our configuration tab. And then we can click

configuration tab. And then we can click into any one of our agents runs and see either where it may have gotten stuck where it successfully executed, and we

can watch it think together. Let's see.

It's adding a detailed analysis with impact behaviors breakdown to our document. We've got our old placeholder

document. We've got our old placeholder data there.

While that's all right, fantastic. While

that's running, I will say uh classic live demo fashion. We had a little bit of a hard time together. Uh but it is super simple to build your agent. It's

just as simple to tweak it. So, I could go ahead and have a conversation with it here and say, "It looks like you're

sharing placeholder data in the doc, but

you gave me an ex excellent analysis here. Can you help me improve

here. Can you help me improve my agent prompt?"

So, if I scroll up, you can see it actually read the feedback and then gave me uh information about how I matched against the impact behaviors. But in our

tracker, it put placeholder data. So, I

can actually talk it through troubleshooting just like I would talk to a person and say, "Hey, you're still putting placeholder data in as a variable. I'm not really sure why. Can

variable. I'm not really sure why. Can

you help me improve it?" And in this case, the agent was like, "Yeah, I totally did that. I see the issue. I'm

using placeholder variables instead of the actual data. Let me go ahead and improve the agent. So now it's extracting the data we want. It says

it's running it again. Let's go ahead and give it one more chance.

Okay. Can you run it one more time?

And then I promise we'll switch to Q&A with our last few minutes. I want to make, sure, we, have, at least, 10, at, the end.

All right, let's see. Moment of truth.

It betrayed us.

All right, I'm going to go ahead switch back to configuration.

I'm going to check on one thing in front of all of you my wonderful participants here. Uh, and

I am going to actually stop screen sharing. Here's what I'll do. I'll stop

sharing. Here's what I'll do. I'll stop

screen sharing. We'll see if we can come back to this in our last minute here.

Share.

There we go. Leah, can I invite you up for any Q&A and then we'll see if we can revisit our tricky little agent here in a minute.

Hey., Hey., Um,, all right., First, round, of clap emojis for building live. Um, I

think this is always fun. Uh, you got a ton of love in in the chat. So, uh

yeah, let's let's jump into some of the questions. Um I think

questions. Um I think see um so one is from McKenzie. Uh we

have this here. How does your team share ideas for agent use cases?

Oo that's such a good question. Right

now um we have a few like base questions that we ask ourselves at Zapier that help open us up to like automating. And

sometimes automation looks like an agent, sometimes it doesn't. But some of those questions are like, is the process tedious?

Is the process repetitive? Because if

you think about it, we're taking a few minutes to invest in building an agent or a workflow. So, we want to make sure it's either something that's so big and tedious, it's worth that investment once, or it's something we do enough

times in small chunks that when you kind of balance it out like five minutes to build my agent, 30 seconds every single day for a year, that's worth it to me to build the agent. So, we will often talk

to each other about like what feels tedious, what uh is repetitive or just plain annoying. Like if it's something

plain annoying. Like if it's something in your job that doesn't have to be a human task and it annoys you, please build an agent. And also, oh my gosh everybody, as soon as I stopped screen

sharing, it worked. So, I'm going to show you really quick. I'm taking over.

All I had to do, friends, was refresh.

So, sorry. Uh but here we go. it uh it ran perfectly. It went and captured my

ran perfectly. It went and captured my Slack feedback. It pulled in the impact

Slack feedback. It pulled in the impact behaviors that connected to that feedback. It said like where I could

feedback. It said like where I could grow based on my company's competencies and where I was shining. And then it even summarized key takeaways. So, uh I do want to share that was related to

this piece of feedback here that I reacted to. And if I go back to my agent

reacted to. And if I go back to my agent now that I've refreshed, of course, live demos, you can see complete here, you can see it working. And if I click in, I

can even like go back and read how it did that entire task. And if for some reason I wanted it to do it again or differently, once again, I can talk to

my agent in this box here. I don't have to get any engineers involved to do any coding. It's all human language, natural

coding. It's all human language, natural language between me and a computer. All

right, that's all. It's working. Back to

you, Leah.

All righty. So, we'll grab uh uh Leah's question here. Um so, I followed the

question here. Um so, I followed the instructions exactly, but I've got errors and testing the agents. What's

the best way to troubleshoot? Uh I think a lot of what you showed was great. Uh I

think turning it on helps always. You

know, when in doubt, turn it on and off refresh, um is always going to be key.

But uh any other tips and tricks as people are building agents?

So often when I jump in a live call to troubleshoot with someone, the first thing I do is say, "Did you ask your agent?" Uh anywhere where there's a text

agent?" Uh anywhere where there's a text field to talk to your agent, and I know they're in a few different places if you're in like the activities tab or the configuration tab, but I absolutely

recommend you talk to it. You either

take a screenshot of what you're seeing and push it to the agent by putting it in that text box or you say, you know here's what I'm seeing. You're not doing it right or that there's an error coming

through. What should I be doing? Because

through. What should I be doing? Because

it knows its own sort of like programming really, really well. And it

can help you get unstuck pretty consistently. Sometimes it will even fix

consistently. Sometimes it will even fix itself. If you tell it what's wrong, it

itself. If you tell it what's wrong, it will say, "I have adjusted the instructions." Uh, like you might have

instructions." Uh, like you might have seen when I was talking to it.

I love that. Um, yes. When in doubt, I always sometimes I'll even go to the agent and be like, "Hey, can you help me? Like, what's a good prompt? Like

me? Like, what's a good prompt? Like

how should I structure my prompts?" Uh

because I use an app called Super Whisper because I love I can't type as fast as I talk. And so, I will just have it record. I'm like, "Here's what I'm

it record. I'm like, "Here's what I'm looking to do." And I'll just like stream of consciousness. This is what I'm trying to build. And then we'll then start uh from that path. So, love that.

Uh, all right. Next one is from Jennifer. So, are there any data

Jennifer. So, are there any data security considerations when linking Zapier agents to company tools and documents, especially with a free account?

Absolutely. I love this question. So

um, when you're connecting your data to Zapier's tools, some of Zapier's tools let you choose which large language models you're using. So, for instance

today we built an agent. Uh, another

thing that a lot of you have probably seen us build with is something called an AI by Zapier step in a workflow. And

when you're doing that, you can choose like, do I want to use OpenAI or um, do I want to use Claude or Gemini? And if

you've got an agreement as a company with them, like an enterprise agreement you'll want to talk to your company's legal your leaders about like how where

is our data safe? How do we want to connect to other softwares? So, for

instance, if if like I worked somewhere that had an enterprise agreement with Gemini, I would want to make sure that I had my Gemini account connected to my Zapier workflows. Probably wouldn't want

Zapier workflows. Probably wouldn't want to use like OpenAI with no agreement.

So, when in doubt, ask your legal team.

Every company's going to be different and have different requirements and will have different tools where data is stored in different ways. So, please

please, please default to checking with your company before connecting to any software.

Alrighty. Uh let's see. We've got our next one here. If I can maybe answer it live. Uh it's from Amy and it's uh which

live. Uh it's from Amy and it's uh which HRIS or uh ATS systems does Zapier work best with?

Oo, work best with. Man, I'm not sure I can answer work best. We've got 8,000 apps that we connect to now. And uh I will say some folks had asked me in a

previous session like where can I see all those apps um and within agents uh or within zaps it looks a little

different but I'm going to say that uh Zapier apps there is a list of all of our integrations that's pretty easy to browse. They've got like a human

browse. They've got like a human resources tab. Um, and what you can do

resources tab. Um, and what you can do is before you ever connect to anything you can actually find the tool you're looking for. So, for instance, like

looking for. So, for instance, like man, I love Bamboo HR. I just can't get enough of it. I know we're always saying that. I want to see all of the different

that. I want to see all of the different actions I can take from Bamboo HR. Um, I

can go to zapier.comapps

look for the HR use cases along the lefth hand side and then I can actually click on the tool bamboo and browse all of the suggested ways to use it like

oh, there's one that's like grabbing time off requests. I know that data comes through. There's another one to

comes through. There's another one to create a database when reports change.

Another one for uh teaming and whatever like all of that stuff. So, I would recommend you you browse through the apps, look for the tools that are native to you and the way that you work um and

then check out how those tools are used on the website for some inspiration and to get you kind of like percolating on what to build.

Amazing. All righty, the next one is up.

So instead of just gathering my own feedback, can I build one agent that with a certain emoji that collects all feedback and then I can split it out per

receiver of the feedback? I love this because my core role is actually manager enablement. Uh so in L & D like my focus

enablement. Uh so in L & D like my focus is user experience for managers and I've helped several managers build databases where they are tracking feedback for all

their directs and then when it's time for performance reviews they can just sort by like which direct was this talking about. In that case, instead of

talking about. In that case, instead of using a Google doc where you are uh putting all of the feedback as it comes in, I would recommend using a Zapier

table or if you're a company that uses Air Tables or um Google Sheets because then you can filter um and so like later when it's impact review time, if you've

made or sorry, performance review time if you've made a full table of all that feedback, each row is a different piece of feedback and it is denoted by a

different employee. When it's time

different employee. When it's time you'll just make a view or a filtered view of like I only want to see the feedback pertaining to Leah so that I can kind of summarize everything I've shared about Leah or seen people share

about Leah this year. Um, so yeah, you can absolutely automate that. I think

that's a really powerful use case.

Uh, and as someone who's personally benefited from that workflow, I actually use that with my team. So, we just did our like H1 reviews. Uh, and then what's really cool is if you take Emily's like

impact behaviors, uh, and if you say you have a a teammate that is, you know, in IC3 and wants to go to that IC4, I'll put both behaviors on there and I'll ask like the agent to say, how does this

reflect from an IC3 and then how does this match to that IC4 behaviors that we want to see? And that helps frame up uh my growth conversations with them as well. So it's not I'm not just looking

well. So it's not I'm not just looking at it, you know, once every six months.

It's just a continuous conversation that we have. My added pro tip there.

we have. My added pro tip there.

I love that. I love that you you're living it so you can speak from like real experience.

All righty. Uh we got tons of questions here. So, next one. Um, how to best

here. So, next one. Um, how to best create templates so you can share agents with your team um and then they can edit uh for their own teams. This is such a good question. Okay

forgive me. I'm going to switch back to screen sharing for a second. So um there are several different and uniquely

amazing Zapier products. One is called chat bots. A lot of you might have

chat bots. A lot of you might have interacted with that. That's where

there's like one chatbot and everyone can talk to it. Agents is a little different than a chatbot where like I have an agent and I can talk to it, but if I want to share it with Leah, if I'm

like, "Oh, I nailed this. I think it's working so well. Leah's going to love it." uh Leah can't go talk to my agent.

it." uh Leah can't go talk to my agent.

So, that's one way it's kind of like private and locked down is like no one else is going to go be able to go to that like configure tab and have a conversation with my agent like I did

but I can still share it and make sure that people who want to use it can do that. Uh I can create a template. If you

that. Uh I can create a template. If you

can see where my mouse is in the upper right, um I can go ahead and create a template that then anyone can click on and it will make a copy just like a Google doc where like you can't edit mine but you can have your own and then

you'd go in and change the authorizations to your Google Docs, your Slack account. The other thing I can do

Slack account. The other thing I can do if I love this agent but I'm like I want it for another emoji like I want one for every one of my direct reports and I'm going to, react, to, the, little, picture, of their faces. uh I can just duplicate the

their faces. uh I can just duplicate the agent and make another one and all I would have to change is like the reaction details. So two ways that are

reaction details. So two ways that are helpful. One is creating a template to

helpful. One is creating a template to share with others. There's a big share button up at the top here uh where I can generate a link of a copy that other people can customize.

And then uh the other way is if I want to duplicate it for myself, there's a duplicate button.

All righty. Um, next one is, "What's the number one most common and useful use case for Zapier and HR for small to medium businesses?"

medium businesses?" Oh my goodness. I feel like this totally depends on what the parameters for your HR department are. I

feel like when you look at small businesses, HR wears so many hats and they're often doing like ops stuff and like there there are a lot of ways to answer this. I would say the most common

answer this. I would say the most common ways I've been seeing floating around lately are onboarding new employees. I

mean, what a thing that's ripe for automation. We have seen uh many

automation. We have seen uh many companies that say they're cutting like 15 hours per new employee just by automating information going out greetings going out, connections

happening. Um, so onboarding is one. The

happening. Um, so onboarding is one. The

other one is recruiting. It's another

one that's so ripe for automation.

Making sure nobody's going to slip through the cracks. Everybody's getting

followup, calendaring is scheduled follow-ups are scheduled, um, collating information about each candidate. Those

are both super super valuable uh, use cases for for HR and Zapier.

All right, going down the line here.

Um, all right. Uh, when do you build an agent versus buying an AI tool for a specific use case? Uh what AI tools does Zapier HR team use?

That is a wonderful question. Um I would build an agent whenever it is valuable to you to be able to customize what's taking place. So for instance, uh I used

taking place. So for instance, uh I used to have a tool called Reclaim AI. A lot

of us are probably familiar with it. Um

you know, something comes in in Slack, I need to get it done. I'll push it to reclaim or to-doist and then like that finds time on my calendar. that's fine

but it was only so customizable. So, I

actually ended up I no longer pay for that. I just built it in Zapier agents

that. I just built it in Zapier agents where just like you just saw whenever someone asked me to do something in Slack, I react to it with to-do and then I told the agent like when that happens

summarize what I'm being asked to do. I

actually gave the agent, this is my unlock here. I gave the agent my whole

unlock here. I gave the agent my whole job description of what I do and who I am and what I like to do for work, like how I like to work, which I couldn't do and reclaim. And then I said, based on

and reclaim. And then I said, based on all that, estimate how long you think this ask is going to take me. Look at my calendar. Here are my preferences. And I

calendar. Here are my preferences. And I

couldn't get that granular and reclaim right? I couldn't say like, I don't love

right? I couldn't say like, I don't love to have meetings after, you know, this hour, but I will. And I would love to have my work time scheduled at this hour, but I can't. you know, give all

that to an agent and then uh whenever I react to a to-do, it takes all that information, it puts it on my Google calendar. It's so I never ever fail to

calendar. It's so I never ever fail to follow up on something I was asked to do in Slack. It's like my superpower. I

in Slack. It's like my superpower. I

always follow through. Um so in that way, because we care about like being able to customize, I wanted to make sure it worked for me. I'm not going to pay for another tool when I can do it better

on Zapier with one agent. However, to

your point about like um someone asked about data privacy. If there is a tool where they have absolutely nailed uh like their data privacy, they're going to be able to do things at scale that I

can't do building my own agent. I might

actually consider using it. A good

example is like we are we pay and use Bamboo. We're not replacing Bamboo HR

Bamboo. We're not replacing Bamboo HR right now with Zapier's suite of tools.

There may come a day where our tools are like able to do everything we care about in Bamboo. Right now, we're perfectly

in Bamboo. Right now, we're perfectly happy to pay for Bamboo HR and then connect it to other tools using Zapier so that we don't have to pay for as many things.

I love it. I'm needing to go back and like look at the AI tools I have now and be like, can I build that in in in Zapier agents now? All righty. Um

here's one about the actual build. So

regarding this agent, will it go through all Slack channels or do you do you need to add those? Uh, I'm getting a question to provide a link to the channel name.

Such a good question. Oh man, these questions are giving such great uh like upleveling opportunities. Okay

upleveling opportunities. Okay switching back to my agents tab for just a second. When we look at the

a second. When we look at the configuration of our Slack trigger, one of the things it asked for was like, do we want to

pick a specific conversation to look for? We don't. By leaving these blank

for? We don't. By leaving these blank like we are saying to the agent, it doesn't matter which conversation I react to or which author wrote it. No

matter what, I want you to take these actions. Now, if you're getting a field

actions. Now, if you're getting a field that asks you like which channel in Slack and you leave it blank, that essentially means all channels. And

usually that's what we want. There are

occasions like my example of um react to this message to get a calendar invite.

We might want that to only be pertinent to people within our HR Slack channel.

So then we might want to say only this particular Slack conversation or only this channel. But generally speaking

this channel. But generally speaking for something like this where I want to be able to use it anywhere I get feedback in any channel, I'm going to leave those blank.

Now, as a followup, can that be in a private DM or does it need to be in a channel?

I love this question. That is a different trigger. So, um, if you want

different trigger. So, um, if you want to be like a huge nerd with me for a second, Slack treats channel messages, direct

messages, so private DMs, and chatbot messages as three different kinds of data. Uh, and so you would actually have

data. Uh, and so you would actually have a different trigger for that. I do have a workflow that does this same thing from a DM. Uh, and it looks different

than this one. uh because of the fact that like public channel messages are something that our agent can connect to automatically. If you want to do it from

automatically. If you want to do it from a DM, you would need to push it from a DM, which just looks a little bit different. And in that case, I actually

different. And in that case, I actually end up using a Zapier workflow and then pushing it to the to an agent. So once

you have your agent built, you can actually have it as a step inside of Zaps, which is super helpful. That's for

another day, but but that's the answer to your question.

Uh, love it. I know we're getting uh if you have any workshop ideas, feel free to throw them in the chat. I'll go

through them later because we want to do these monthly for you. I saw onboarding was another one. Um, Emily, I'm you're booked. Um, I'm just going to book all

booked. Um, I'm just going to book all of your time.

Um, all right. Uh, we've got seven minutes for questions, so let's run through them. Which tools do you use for

through them. Which tools do you use for performance reviews out of interest? Uh

that's something that Bamboo HR offers but feels like something we could possibly build in an agent.

Okay. Okay. What tool do you use for performance reviews out of interest? I'm

just rereading. Uh, what's something that Bamboo HR offers but feels like something an agent could do quite well?

Would love to hear if you Okay. Oh. Oh.

Oh, that's something that Bamboo HR offers. Yeah. Okay. Great question. So

offers. Yeah. Okay. Great question. So

this is a meaty one. Performance

reviews. Uh I am actually the person who builds our tooling for performance reviews. Uh we use at Zapier we use

reviews. Uh we use at Zapier we use Bamboo for employee data but we actually use small improvements as our

performance uh sort of like home for like one-on- ones um performance reviews ongoing feedback gratitude praise that

kind of stuff. Um and at the same time we join those together a lot with Zapier and the tooling that we've built that layers on top of those is again the need

to customize. So I have built 56

to customize. So I have built 56 different chat bots for the company to use that each chatbot is equipped with those impact behaviors the um the

competencies for your particular department's roles. And when it comes

department's roles. And when it comes time for performance reviews, you know that document that we just set up that has all the feedback across your year that you're getting in Slack, all you

have to do is copy that entire document and give it to the chatbot and it formats it into our performance review format for you. Um, and helps you through conversation. So, it's not

through conversation. So, it's not making decisions for you, it's talking to you. Um, it helps you narrow in on

to you. Um, it helps you narrow in on what you think are the most important things to surface to your manager. And

we have the same bots for managers to help them streamline their communication about their employees, how they can grow, what they're doing really well how they're having an impact. So in

summary, we've got like the baseline uh tooling that where we like store this information in an ongoing way. We use um Bamboo HR for employee data, small

improvements for the actual performance data, but all of the customization that we've built, we've built on Zapier using Zapier agents, tables, chatbots

workflows, canvas.

Amazing. Um, all right, here is a fun one. How do you consider equity in your

one. How do you consider equity in your PM automations and natural bias that is coming up in this cycle pre-automation?

Mhm. Uh so PM here for anybody who's not in performance related HR stuff, PM is performance management and uh thinking about equity and performance management

is like your starting point, right? So

in this case when we built the chat bots, uh I mentioned a second ago like the chatbot's not making decisions for you. It's actually right in its

you. It's actually right in its instructions that it may not evaluate your performance as far as like telling you like I don't think you're exceeding expectations or like oh that's you know

it it does not do that. What it does do is it summarizes back to you. Think of

like a coach right the coach doesn't tell you what you're doing well. They

say things back to you to help you get clarity. So that's how we have set up

clarity. So that's how we have set up our uh like scaffolding for performance.

We as a company have very strong ethical stances on what we believe are like the boundaries of AI use in HR. So for

instance, I feel really confident telling you that like AI, we're just not going to let AI make decisions. We're

going to let it help us summarize clarify, but our boundary is like people make decisions about people. AI is not making decisions about people. Um, as a

company, I would really encourage you wherever you work to think about a pair of things. Number one, making it so

of things. Number one, making it so clear what your boundaries are for what AI does and doesn't do. And when I say make it clear, I don't mean just to you.

I mean to your entire company. This

should be something any level of employee can answer if I ask them about it. Like what does your company believe

it. Like what does your company believe about what AI is never going to do for the company? The other thing is thinking

the company? The other thing is thinking about how we can think of building empathy with AI and against AI. So what

do I mean by that? Where does AI make space for us to have human empathetic moments? So like if it's great at

moments? So like if it's great at collating data, fantastic. Take that off my plate so I can have more one-on-one meetings about performance management like a human to another human. And also

where can I build in empathy into my AI products that I'm building? I'm building

this agent. How can I tell it more about me so that it actually approaches things with some knowledge of where I'm coming from with some empathy baked into it?

So, those are what I mean by like empathy within and empathy up against it. Um, so that was kind of a we went on

it. Um, so that was kind of a we went on a ramble there, but I think it was important to talk about.

Uh, no, I think it's great. And if you weren't part of Brandon's talk um before this, his AMA, he said something that really resonated with me is AI is the tool, not the outcome.

Yes. And I think this is a really good example of that of we want to make sure that we're building it in a way that we do have humans in the loop. Uh I mean who knows maybe five 10 years it'll be a little bit different. We can have all

these like guardrails and it can start making decisions for us but we're just not there yet. So I think having those human in the loops and then also like question it. Um I know uh when I am

question it. Um I know uh when I am going through like uh interview notes or review notes, I'll go through and I'll even ask it to to question like is there bias language in here? Are there things

that I'm not realizing I'm giving to to one employer or the the other? And so

those are things that I think it's great for it to be a mirror to reflect back on you. Um but yeah, I love that kind of

you. Um but yeah, I love that kind of building in that empathy.

Yeah, absolutely. The biggest

Oh, I'm sorry. I was just going to say I think the biggest danger is letting AI be a yes man instead of asking it to challenge you. Um and that's like we see

challenge you. Um and that's like we see that a lot in the news right now that that's becoming a dangerous area and I think we can all do our part to like give it the space through our prompting to really challenge some of our thinking

too.

Uh yeah, just to build on that, my pro tip, uh if you do use like a a chatbt you can go into your settings and I always have it in there and like challenge my thinking, challenge my

assumptions and ask me questions and make me like fight for my point of view because uh yeah, nobody wants a yes, a yes., Um, a, standard, yes., All right,, we

yes., Um, a, standard, yes., All right,, we are unfortunately at time. We still have a good couple of questions in here.

We're going to look to answer them and reach out to you oneonone. Uh, thank you so much, uh, Emily. Fantastic build. Uh

I loved the transparency and, you know when you build things, you will need to troubleshoot them. And I love that you

troubleshoot them. And I love that you created that space for us to to see how you build these things. Ah, thank you all for sticking with me. I know it's kind of awkward to watch me troubleshoot live, but it's part of the reality. We

don't have to be technical in order to get it working. So, thanks for sticking with me, everybody. All righty. Uh, as a reminder, we'll be sending out the recording later this week. And yeah, if

you have any ideas for future workshops and how I can book Emily's time, let us know. Send us a DM, reach out to us on

know. Send us a DM, reach out to us on LinkedIn. And yeah, thanks everyone for

LinkedIn. And yeah, thanks everyone for joining us. Have a great one. Bye

joining us. Have a great one. Bye

everyone.