Slack Code Brings Multiplayer AI Coding to Teams (Full Transcript)

[00:00:00] Speaker 1: Do you want to see multiplayer coding in Slack? Watch this. Welcome back to Slack School. My name is Mike Reynolds. I'm your host. I'm part of the Slack team here at Salesforce. And today I am super excited to show you Slack Code. Slack Code is an amazing new tool that lets us have multiplayer coding natively right inside of Slack with agents that are able to write code for us. But in a way that you've never seen before, this is going to be worth it. So what is Slack Code? Slack Code is a special channel that an agent creates and manages where they're in charge of the channel and what happens in it. We get to play along, giving them direction. They use that information to then go build code for us. What's great about this is they've got two really fancy tools at their disposal. They can create a code diff and then actually show it to me right in Slack. They can also create HTML renders and show me those HTML renders, again, right in Slack. And this is all multiplayer. That means as I write my code, I can then vibe code with my agent. My other devs can be in there with me, my architects, my admins, the BAs, the project management, leadership, people who are going to be UAT testing this. They can all see that HTML preview. So maybe they're not going to see the code diff and understand it, but they can see visually what is happening before I even build my first PR. Now in the example that I built today, we're going to go pretty quick because I'm going to have my agent do essentially everything for me. So the agent is actually going to take my request, build a pull request and a branch, right? They'll build a branch, add their code to that branch, make a pull request, and then show me that diff and a preview. As long as I ask a question that requires a preview and as long as I ask a question that requires code, the agent is in charge here. I'm just going to be prompting them and then hopefully it works out. Let's take a look and see what that looks like. So what I've done here is I've at mentioned my agent, whom I called Jolly Roger. I've at mentioned Jolly Roger and I've said, hey, can you create for me a lightning web component that I can add to my case record pages? I want this component to give me the customer's sentiment kind of as a score. Now I didn't provide a ton of information, but I did enough to justify that we need a code channel because I've directly said to my agent, hey, I want you to write me some code. And in the agent's code, it knows every time I ask it to write code, it's got to do that with a code channel. So I can see here that Jolly has created a code channel and has given me, I've got this little block, right? This block is a link to my code channel because we want to leave breadcrumbs so that if other people go back in time and they want to see how we got here, they can do that pretty easily. I'm going to go ahead and click on this block and it's going to take me up here to my code channel. And again, I have a nice breadcrumb for myself. It says this is context from this code requests channel, which was where I started. And then it repeated my initial comment. I can see Jolly has already put an emoji response on this. Now I've set Jolly, my agent, Jolly Roger, I've set Jolly up to always create a code channel if I directly prompt Jolly to write some code. I've also said that if you're going to write code, you've got to have a branch. If we're going to move forward with that code or the code is done, then we should create a pull request on that branch. These are sort of deterministic guidelines that I've set up into the agent. The agent also knows that it should show me a code diff every time it writes code, even if that's all new code. So the diff is going to show predominantly green, new lines being added or whole new classes, a whole new component in this case. I should also be given an HTML render. Of course, anything visual is going to have the opportunity to have an HTML render. But if what you're writing is not visual, it's just a trigger or something along those lines, you're modifying an Apex class, we're not going to see a preview for that. This also doesn't have to be Salesforce related. I could be asking Jolly to build me an entire website. It doesn't really matter. Jolly is a coding agent. So Jolly can do just about anything. So let's take a look and see what we've got. First I can see that Jolly has put a code block in here and now I've got a little component preview. I can also see that Jolly responded up here. Before I even click in here, I want to take a look at this new section up here in the top right corner that says artifacts. When I click on artifacts, it's going to show me the two special things that Jolly has the ability to do for me. It can show me the literal new code that it wrote, which I can see here. The entire component bundle appears to be here. That's good. And I can view all these lines the same way that I would look at them in GitHub, which is great. I can also see a component preview. When I click on the component preview, it's going to show me literally everything that it built. Now, of course, it's gone a little bit above and beyond here because it's giving me how to deploy this. Here's a customer sentiment if it was positive, neutral, and then negative. It's also built in this idea of confidence. And then, of course, because the persona of my agent is a pirate, I'm getting some kind of pirate speak built into it. This is all really cool, but before I move on, I want to touch on something. I've made Jolly Roger obviously a pirate. This is actually a good demonstration of something that you should be doing too. Yes, a pirate is silly, but you can see the ways in which the personality of a pirate has been injected into the results. The coding agents that you work with shouldn't just be a coding agent. Think about having a coding agent with a particular persona. So, for example, I might want a senior developer coding agent. I might also want another agent that is a senior architect. And a very, very senior admin. And then I could have all of those agents participate in my development process. The reason for that is that I'm going to have a certain perspective that comes from an agent that's designed to focus on InfoSec or information security. That agent is going to have a particular goal in mind that is not necessarily purely what I asked it to do. It's going to be thinking about how do we make sure that the thing that we built was safe, not necessarily that it was the most beautiful or the most functional. And by working with agents that have these personalities, I'm going to have an end result that is a lot better than just one agent that I didn't really give a personality to. It's just some LLM that is really good at code. That's fine, but we can do better. Let's look at what Jolly actually said. I'll open this up. Here again, it's got a good overview of exactly what's happening, says what it built here, and it's given me some instructions with a link to the pull request. That's pretty cool, because like I said, it did create a branch and a pull request. I'm not going to bother looking at that. We've all seen pull requests before, but I do want to show you another one that I've already done earlier. When I take a look at this one, I can see comments from another user, Alex Brewer, and Zoe Maxwell, and in here, those comments informed this birthday countdown, where again, I got information about how to deploy it, and then I've got a way to preview different scenarios to see exactly what everything looks like, and that's really pretty cool. And again, that's informed by this larger conversation where everybody was participating. This is true multiplayer coding with agents right in Slack. Well there you have it. Multiplayer coding with agents, and your whole team is alive and well right inside of Slack. Being able to see a code diff without going anywhere else, being able to see an HTML preview without going anywhere else, my entire team working together in the same place, seeing the same things. Just think of what you can do when you connect other agents into this, or bring your entire team together. There's no more guesswork. It's all just laid in front of you, easy to understand. What could be better? Let us know what you thought of the episode, or let us know what you want to see me cover next by joining me in the Slack community at slackcommunity.com. Hop in at the Slack School channel and tell us what you thought. We'll see you next time. Oh, don't forget to like and subscribe. And hey, keep at it, you're doing great. Oh, come on, record. Record.