AWS re:Invent 2025 - Slack Platform: Build with AI-Powered Development (AIM111)

🦄 Making great presentations more accessible.
This project aims to enhances multilingual accessibility and discoverability while maintaining the integrity of original content. Detailed transcriptions and keyframes preserve the nuances and technical insights that make each session compelling.

Overview

📖 AWS re:Invent 2025 - Slack Platform: Build with AI-Powered Development (AIM111)

In this video, a Slack developer demonstrates how to supercharge AI-assisted development using Slack's platform and deploy AI agents. The presentation covers Slack's event-driven architecture, showcasing how multiple agents (GitHub Copilot, Claude, Cursor, Codex) can coexist in channels and be invoked like team members. A live demo shows GitHub Copilot automatically creating PRs and providing asynchronous updates. Key features highlighted include out-of-the-box UI, contextual APIs, WebSocket support for local development without endpoint setup, and the ability to build functioning Slack apps in under 67 minutes using AI-assisted coding tools. The speaker emphasizes Slack's platform simplicity, enabling rapid prototyping and user testing over lengthy development cycles.

; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.

Main Part

Supercharging AI-Assisted Development with Slack: Platform Overview and Live Demo

Hey, everyone. Thanks for coming. So I want to do two things today. I want to talk about how you can take your AI-assisted development and use Slack to supercharge and streamline it. I also want to touch upon how you can use Slack to deploy agents and build AI. I'll try to go over some high-level topics when I can. Does anyone in this room use Slack? Has anyone developed on Slack before? Okay, one or two. So I'll go over some high-level concepts.

Slack, the Slack platform came out maybe eight to ten years ago, and at its core, it's basically an event-driven platform. Back then, we called them bots, which were just non-human actors within the Slack workspace. Now we call them agents, which are bots with LLMs attached. The technology and the architecture is almost exactly the same. Over the past few years, we've found a really good match with the development that's been happening. It's a natural match of Slack and AI, and we've seen a lot of growth.

I want to focus on the far right here. I'm not going to talk about Slack usage. I'm going to really focus on developer conference and specifically on AI and what we're doing within Slack. Every company is complex. There's no one silver bullet or one solution that works. Companies make acquisitions, employees build things and leave. There's always a blend of solutions. Our approach to AI is trying to support all of that, whether it's out of the box, whether it's something you've had running on your servers for ten years, or whether it's a partner that you chose to work with. We try to build in a way and give you the tools to serve all those things.

I would say our biggest advantage is the UI. When you're developing AI, you should be focused on the new technology, the business value, and the solution. Slack gives you an out-of-the-box UI you don't have to worry about. We've been through some new things this year. Have you seen Work Objects? We're having some new multimodal stuff that's coming out. Probably one of the most popular aspects of working in Slack and building AI is context.

To have a good agent, what you need is to understand who's talking to whom, what are the relationships, how do people relate to each other, how channels relate to each other, how objects relate to each other, what happened last week, and what happened in the last five minutes. That's what it takes to have a good contextual agent that actually makes sense and actually serves a purpose. We give that to you through a set of APIs where you can call in real time. You can call on behalf of the user as well. When you're launching agents in Slack, you can understand what that user is allowed to see and only what that user is allowed to see, and then use that context for the agent.

We also launched some automation this year. Like I mentioned, every company is different. Sometimes you don't want to build. You just want to be able to do no code or configure. In that case, you can also automate that in Slack. Lastly, what we've been trying to do is take all these new UI elements you see around the internet on different AI products you use and blend them into the Slack platform, but do it in a way that doesn't increase developer load or cognitive work. We put in streaming, shimmer states, and thinking strings naturally into the platform, but we've done it in ways that allow you to develop very quickly.

I want to bring over and run a demo here. What I want to show you is how you can use some tools you might use today when you're developing with AI within Slack. This is Slack, if you haven't seen it. What we have here is a payments channel. This is a group of developers working on a payments app. GitHub is in here. In your Slack, you probably have many different apps and bots sending messages, so this may seem familiar to you. In addition to GitHub, what we also have is a Copilot agent. Launching agents in Slack is essentially native. We've had the capability for a while and we've matured it, but it just comes out of the box functionality.

So in this example, let's say there's an issue, and I want to invoke an agent that's in this channel. I'm going to call up the GitHub Copilot agent. From this action fail, and just call it like I'm calling another human, right? There are actually 6 different agents in this channel, but I'm just going to call GitHub Copilot.

What it's going to do is take off and say, all right, I have a job to do. I'm going to do it. You might see some emojis being used. So I think it'll say, yup, I have some eyes, I'm looking at this. I know there are 8 people in this channel. Your channels may have 1000, you may have 5. But typically, you want everyone in the channel to understand that something's happening and I don't have to do anything. And then the one person being able to do that thing, right? That's probably the most efficient way to work.

So what Copilot will do is automatically let your team know, yup, this is being looked at, being taken care of. We've created a PR, and you can look at the web as well. Then it'll use some of our new UI elements to just update the progress along the way, which is pretty cool. So if you use Copilot before or you know, Cursor or Codex, this will look very familiar to you. You can see your running agents. We always try to give you the ability to delink exactly where you need to go, in case you want to go do some deep work outside of Slack, right? Because sometimes that's the best place to do it.

So if you really want to see what's happening, if there are different processes running, you can go ahead and delink there. And then this PR essentially will just be updated over time. I think overall, it's going to do maybe 3 commits. And then Copilot will just do his thing. Again, we'll delink for the people that need to go out.

The best part of bringing AI development into Slack is it's becoming more and more asynchronous, right? No one wants to sit there and stare at their computer and watch it work. So when you're in Slack, it'll just give you a notification when it's done, right? If you want to go check, you can see what it's doing, great. Or you can go about your day and your little activity monitor will pop up when it's done. I want to point out a couple of other things. So in this one channel, we also have 6 agents total. We have some other agents here. We have Claude, Codex, and Cursor. Does anyone use Cursor? That's right. Anyone? Oh, that guy does. Anyone use Codex or Claude? OK, OK. All right, so me and Mosa and Claude. So all these agents are in that channel as well, and you can call them in the exact same way.

That's an interesting distribution though. In addition to that, we have some out of the box agents. So oftentimes, if you don't want to build, you just want to use Salesforce agents. We have those as well, and those are going to be under the agents tab. And what those are really useful for is saving your team time. So Channel Expert is one that we came out with earlier this year. I'm sure everyone in this room's had channels where people drop in and ask questions. Maybe they're good questions, maybe they're bad questions, but they're definitely distracting questions, right? So what the Channel Expert does is it automatically looks back for the channel, also looks at any resources that you choose to give it access to. You can just click and add to it, and it'll go ahead and try to answer those questions.

So it's great for people that aren't on your team. Maybe it's great for you if you came back from vacation and you're like, all right, what happened? Or maybe there's like, oh, does this issue happen before? It's just a little value add. I use it in all my channels. It's great. If someone asks something, there's a very good chance that they can get that answer through those resources or the past 500,000 messages or pings that came into that channel. So that's cool. And the employee agent's a fun one too. Because this channel is called Payment PR. Every once in a while, you might get someone pop in and be like, oh, my reinvent expense was rejected, which obviously the development team would not support that. So this agent would kick in and be like, all right, no, no, here are the company resources. Like you want to go here to ask accounting about this. So yeah, having all those within, I mean, maybe you wouldn't have 6 within one channel. This is obviously for demonstration purposes, but you can, and they can live there in harmony.

So looking over here, we have the plan. The agent is saying, you know, this is what I'm going to do, this is how I'm doing it, and I'll update you along the way. At the end, it'll clean all this up actually, so it'll just have a nice message. There's it is in progress. I think it's, OK, so it's done. And what you can do, you can also open up a VS Code. So really, the whole goal is to let the people find what they need to find, let everyone else know things are happening, cut down the communication. OK, so we're back here. Obviously, this is an extremely simple solve. This isn't a really complex genetic solution, but it did solve it. It created a PR, and I went there. So if I close here and I go over to the actual PR, the other thing that happens is I might be over here, I might be reviewing it, maybe it takes five minutes or two weeks.

Because that agent is over in the channel, whatever I do here automatically reflects so my other teammates can see. What's interesting about that is it also shows up in search. For incidents that pop up when you're using Slack Search or searching in Slackbot, all of this data will now appear. For example, you can see that an incident actually resolved. That happens automatically for everyone in the company, which I think is really interesting. So while merging without best practices or comments is obviously not the best thing to do, you can see it actually keeps the information updated. You can see the history of the issue and then the actual resolution.

All right, I think that's done. Let me move over here. I showed you a couple of examples. We launched probably hundreds of apps this year. The Slack platform is always free. You can just install apps without worrying about it. If you're paying these companies, we want to make sure you get the most value out of that investment. So if you go over to the Slack marketplace, you can click to install and add them to Slack. I encourage you to check that out after this presentation.

Building Slack Agents: Technical Architecture and Rapid Development with AI-Powered Tools

Now, shifting more from AI-assisted development to actually building agents in Slack. This is good because I think most of you haven't built on Slack before. So what is a Slack app? Like I said before, Slack is an event-based platform. Essentially, an app is something that sits here, receives events, and sends events back. We're language agnostic. I do want to call out one really cool thing though, which helps getting from zero to working in hours to minutes. That's the ability to use WebSockets. When you launch a local Slack app, you can use our CLI. Many of you probably use VS Code or AI-assisted code, which works well too. You can actually run it in socket mode, so you don't have to worry about setting up an endpoint or getting approval to set up a static endpoint for events to go to. You can actually run it right away within seconds and have that work in Slack.

In that scenario, hopefully you're starting in a sandbox, but then when you go to production, you have the option to keep it in socket mode or set up a static endpoint. So you always have that option, and when you're building for a company, it's a really nice option to have. I don't know if anyone's ever been in those arguments trying to open up an endpoint before, but it can kill a project extremely quickly. There are a couple of concepts in Slack with OAuth. They're not that unique, but when you install a Slack app, we have the bot or the agent. We also have the ability to authenticate each individual user, which is increasingly popular these days. I mentioned earlier how you can get the context for that user. You can take actions on behalf of that user. You can search, you can understand who they talk to, what they talk about, and how things relate in their life within Slack. You can pull all that in through our APIs, and it's just a simple OAuth. You can get that through the user. That's a really important topic as well.

Let me come back over here. I thought about showing you some complex ways to build and getting deeper, but frankly, if I was going to build on Slack and launch an agent, this is what I would probably do, so I just thought I'd show you the simplest possible way. This is Cursor. If you use Cursor, you'd recognize it. In the past ten years at Slack, we've made a lot of hard choices. It's incredibly simple to make a complex platform. You just say yes to everything—yes, yes, yes, yes, yes—and you have this incredibly complex platform that takes eighteen months to learn to configure. You need a PhD just to enter this space. We've always taken the opposite approach. We want to be simple, straightforward, and easy to use. We want you to be able to build an app in less than one hour. We want to be quick. It's paying dividends now because it lends itself very well to AI-assisted or AI-powered coding. When you plug in your tools against Slack, this is using a simple one-paragraph, maybe two-sentence prompt saying, "Build me an assistant-like agent using public Slack docs and go nuts." It's searching the web, there are a couple of tools, but I'm just letting it run. It's obviously running in fast forward, but when I recorded it, it took sixty-seven minutes. At the end, you get a fully functioning Slack app.

That's probably the biggest advantage of our platform—just the simplicity and the ability to try it out. When you're building, I don't know how many of you in this room have built something that is in POC purgatory. Probably everyone. Most of my stuff is in POC purgatory, I would say. The real advantage to this quick approach is getting it in the user's hands. Something might seem incredible in your mind, and you might think you're carrying everything, but when you get it in the user's hands, it's like, yeah, that's not that sweet.

Being able to actually launch and test quickly is critical. You might fail 99 times out of 100 when you're on the frontier of building agents and trying new solutions. Being able to do it this quickly and launch is critical. I would always caution against spending 6 months with a plan and building big architecture because everything changes since then. Just get it in someone's hands and start testing. I gave it that one prompt, and that is an installable Slack app. You could run that by socket mode, and it's in your Slack workspace. You don't have to do anything else.

I want to open up to some questions. What do you have? Questions about Slack, about AI, about agents, about tools you use, anything I've talked about. Does anyone have anything? No one? Well, check us out over at booth 840. If you see Red Hat, it's past that. If you want to come by, we can show you how to build, talk through solutions, and work with you on whatever you need. Thank you.

---

; This article is entirely auto-generated using Amazon Bedrock.