00:00Hey everyone, this is Christina from OpenAI
00:02Welcome to Agent Builder 101
00:05Agent Builder is a new visual tool for building AI workflows
00:08You connect nodes and create agents without writing any code
00:12So you can start from templates or build your own from scratch
00:15And it also comes with built-in eval so you can test and understand how your agents perform
00:22When you're ready, you can export the workflow as code or drop it straight into your product
00:26Basically, it's your all-in-one space to design, test, and launch AI agents visually and fast
00:31So, today we're going to show you how to build your first agentic workflow, a helpful travel agent that will help you either build an itinerary or look up flight information
00:42So, we're starting here in the OpenAI platform
00:47Every workflow starts with a start node where you can set input variables or state variables
00:51And today for the travel agent, the defaults are great
00:57Next, I'll connect a classifier agent
00:58I'm going to be building a specialized itinerary agent in addition to a specialized flight agent
01:03And so, I want to first determine which agent I should route to
01:06So, I'll call this classifier
01:10So, you are a helpful travel assistant for classifying whether a message is about an itinerary or a flight
01:23And here I'll specify the output format as JSON
01:26And I'll add a property called classification which will have two options either flight info or itinerary
01:38Great
01:39Next, I'll add in an if else node um to branch based off of the classification
01:45So if put pars.classification classification is flight info then we want to branch to the flight agent and otherwise we want to branch into the um the itinerary agent
02:01So from the flight agent we'll create a new node and we'll call this flight agent
02:09You are a travel assistant
02:12always recommend a specific flight to go to
02:19Use airport codes
02:23And here I'll also give it access to web search so I can have the most up-to-date information about flights
02:32Great
02:32For the itinerary agent, I'll build um an itinerary agent with a new agent node
02:42You are a travel assistant
02:45So build a concise itinerary
02:51Great
02:51I think that's everything we need to get started with our travel agent
02:52So here in run preview, I'll ask it what should I do in a day in Tokyo
03:01And I can see the message going through the workflow that we just created
03:03going through the classifier agent, deciding which category it should go to, seeing that I asked about an itinerary, and then passing it on to the itinerary agent, whereas come up with a concise one-day itinerary for Tokyo
03:15Looks like a great day
03:17Now, for the flight agent, I actually want a richer experience for showing flight information instead of just plain text
03:24And so, I can do that by going and building a widget in our widget studio
03:27Here I've actually already designed um a widget for showing flight information to go from one location to the next with um all the details about the flight
03:36So I can just click download um to download this entire template and then bring that directly to the agent um that we just created here
03:45I'll add it in as the widgets output format
03:48I'll upload this flight widget that we just pulled
03:54Preview it
03:54Everything looks great
03:54And I actually want this to be a bit more customized
03:58Um, and so I will tell it to choose a background color creatively based on the destination
04:12And I'll also ask it to include time zones A.M
04:16or PM
04:21Great
04:21Let's test it out
04:26SFO to Tokyo on October 7th
04:30So here again, we see it moving through the classifier agent um determining whether I'm asking about um a flight or an itinerary, deciding that I'm asking about a flight agent um or deciding that I'm asking about a flight, searching the web, finding a flight for me, and then showing it to me in this rich interactive way
04:46In this case, it's decided that yellow is the color of Tokyo
04:51Um, and so showed that as the background color
04:56Great
04:56So I've now built an agent that I'm happy with and I can publish it directly
05:00Let's call it travel agent here
05:03And I now have a fully deployed agent that is ready to go
05:05Um, I can either use the agents SDK
05:08And here you can see that this is quite a bit of code um for me to manage myself or I can simply take this workflow ID and put it in my product directly using jacket
05:19So that's all
05:19Hope that was helpful and please leave any feedback in the comments and don't forget to subscribe to OpenAI devs for more product updates
05:27Thank you.