Orca

[00:00] This is Orca. It is an agent orchestrator. I just fucking got that. Do you get it? Orca, orchestrator? Anyways, it’s from a company called Stably. They are Y Combinator backed, and this thing does more than I could possibly show in this video, but I wanted to show some really interesting things that it does. So let’s get into it. Okay, we’re here in Orca. We’re going to take a look at what’s called design mode. I’m going to be working on this project here, and I just want to show you really quick. I do have a setup script so that when it creates a new work tree, it’s going to run p npm install. So we’ll jump back to the app. I’m going to hit command N, and it’s going to create a new work tree. The agent is by default set to open code. I’m pretty sure I set that, but I can use any of these that I have installed. I’m going to say create work tree. So as soon as that runs, it’s loading up the open code, but it’s running through the setup first. Once that finishes, we’ve got our open code.

[01:04] And what I’m going to do is in the work tree, I’m going to run NX dev website, which is going to launch that application. So that’s running, I’m going to open up a new tab and I’m going to choose new browser tab. And I’m going to go to that application running inside of Orca in a browser tab. I’m going to split that to the right. And I’ve got open code here on the left. Now in the browser tab, I’m going to click on this grab page element. You can see it highlights the DOM elements as we go over them. I’m going to highlight this card container right here. And as soon as I click that, it copied it to my clipboard. So I’m going to come over here to open code. I can paste that in along with a prompt to make the description extend the width of the card. We’ll enter on that guy. So I got all of that markup. It’s in its own work tree. It’s going to go investigate the project and figure out what to do.

[02:00] Cool. So I made the change there. And if I open up this side panel, not only did we get a file browser. So if I want to go in and look at, you know, get ignore or whatever, we get a Monaco code editor right here in the application. We can make our own changes. We get a history of what our agents have been up to. And then we get our get status here. So I can see one change was made to that guy. I’m going to follow up on this. I’m going to jump to this page and I’m going to tell it to do the same thing here. So I’m just going to highlight this section, paste it in and say, do the same here. Cool. So it made the same change there. All of that’s great. Now I can come over here to these changes and I can say stage all, and I can create a commit message. I can also have AI create the commit message for me. So I’m going to try this out. I’m just going to say generate. So it created a commit message. I’m happy with that. I’m going to say commit and I’m going to say, create a PR. And if I jump over here to GitHub, I can see the PR that the AI created a commit message, but there’s no PR message, but that’s okay. And from here, I can go ahead and say, create the merge commit. And here on GitHub, I can see that the PR has been committed. So that’s the design mode. I’m going to delete this workspace and we’re going to move into something else.

[03:28] So now what I’m going to do is create a bunch of work trees and I’m going to assign them to different agents. So here we’ll do open code again. So we’ll create that work tree. It’s going to go through its setup. I’m going to come over here and say, create yet another one. And we’ll put this on anti-gravity. I haven’t used that forever. So we’ll create a work tree for that guy. And we’ll create one more. We’ll use PI. So each of these is in its own work tree. I can switch between them and I had to improve, go through and find some minor issues, things that I can tweak in the application. And I’m going to assign each of these agents to work on one of those tasks. So we’re going to have PI work on 001. I’m going to have anti-gravity work on 005. And I’m going to have open code work on 007. So we’ll run PI. We’ll run anti-gravity and we’ll run open code. And now what I can do is I can switch over here to this workspace board and I can see that each one is doing its thing and they’ll come up and we will review those.

[04:44] Okay. So open code claims to be finished. If I jump back over here, anti-gravity and ran out of their free tier sucks. So that did not get done. So I can just delete that guy and PI seems to be finishing up. This was like a big one. There was like nine steps, but then the meantime, I can review all the work that open code did. I can actually come in and make annotations on the diffs if I want and send those back to the agent. As long as I’m happy with that, I can stage it all and create the PR. We’ll jump back to our board. We’ll see how PI is doing. Cool. So it looks like it is wrapping up. And if I jump back over to the board, we can see that it is in fact done. I can stage all those, create a commit message, commit those and create a PR. And if I jump over to GitHub, there’s both of those PRs. So that really only covers a handful of features in Orca. This thing does so much more stuff. So I just had it run those tasks in parallel. And I could have said, you know, all of you work on the same task and I could have looked at the diffs and then pick the one I like best. You can do agent handoff where you have a sequence of tasks that need to be completed and one hands off to the next one or sub agents. There’s a ton of stuff in this application. I highly suggest you go check out Orca.