AWS re:Invent 2025 - Build AI Agents with Island: Customer Story Featuring Amazon (AIM125)

🦄 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 - Build AI Agents with Island: Customer Story Featuring Amazon (AIM125)

In this video, Island presents their Enterprise Browser technology and AI agents for secure enterprise automation. The presentation explains how Island's browser provides data governance and control that consumer browsers lack, enabling safe deployment of agentic AI in business environments. A key highlight is Amazon.com's implementation, where they automated purchase order processing and social media workflows, saving 30 minutes daily per user across 1,000 employees—totaling 80 hours weekly. The solution allows teams to record workflows, add human oversight steps, and deploy AI agents with full audit trails while maintaining role-based access controls and credential security through the browser's trust layer integrated with AWS Bedrock.

; 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

Introducing Island's Enterprise Browser: A New Category in Data Governance

Hey, everybody. Thanks for joining today. This is a presentation from a company I'm with called Island, and I can't wait to show you. This is a game-changing technology. We're doing some really great stuff with something called the Enterprise Browser. If you've never heard of what that is, you're going to learn a bit about it. You're also going to understand how we helped Amazon on the .com side do some revolutionary things, specifically with AI. Unfortunately, my co-presenter, Adam, could not make it today, so it's just going to be me as we go through this.

We will have some time for questions if anybody would like to understand a little bit more about what this is. Obviously, our booth is back there, but we're going to have a roaming mic as well for questions, and there will be time for questions. So, what is the big deal? What is the challenge here today? Well, the challenge is that we have these AI agents that can do a lot of stuff, but we need to make sure they're governed in a good way. Today, we are thinking about the data governance that goes along with that. You can do basically whatever you want. The sensitive data goes right in. There's really not a lot of governance to that. Data comes back out.

Then we start talking about agentic AI, right, when it's doing things. It's really nice when it does things for you from a personal consumer type way. Maybe you give it your OpenTable login and it does a reservation. That's cute. But when we bring that into the business, we need different controls. We need different governance around it. What we're going to talk about today is that control and that security and making sure it's reliable, and some use cases that Amazon found that really helped them drive some true value and some cost savings within their enterprise.

First, if you haven't heard about what this category is, it's probably the most exciting thing I've seen in tech for a long time. I've been in tech for a long time. This is super exciting. I've been with the company for five years. We started about five years ago. In order to bring in data, think about when your data gets brought to you. It's brought in usually through a web browser, some sort of an interface. This is where it gets decrypted. Today it happens in a consumer browser that you use, right, Chrome and Edge and others, and that's where that thing called SSL kicks in and it does the big two prime numbers thing and magically brings you your data.

About four years ago, we launched the Enterprise Browser and now we have controls over that. This brought us data protection controls. It brought us delivery controls. It brought us things we could never do before without trying to break open SSL and inspect the wire and see what the data was doing inside there. That's getting harder to do as you guys know. When we brought this to market, we put all the controls that are needed to also gain network access. Think about this as really not just a browser, but really an enterprise workspace that gets you access to the data that you need to get to and governs and controls the data as it goes in and out of the organization. That's kind of the basis of what we've done.

Since we have brought this to market, we have some of the world's largest banks. I think we have most all of them at this point, at least in the top ten. We have some of the largest travel companies in the world, a lot of adoption in healthcare and BPOs, right, because that's where the data leaks out. Unfortunately, this is where data, the operators can potentially accidentally leak data or intentionally, deliberately leak data. We have some fantastic customers that have proven that this technology is real. They've invested with us and obviously they're big companies that we have working with all around the world.

Enterprise AI Agents: Secure and Governed Agentic AI in the Browser

Now, we have AI, and this is kind of where we are. You guys are probably going to be flooded with all kinds of stuff around AI today and tomorrow and for the rest of the week. It's cool. It's super cool technology. We started with the chatbots, thanks to ChatGPT and others, and then Anthropic created the protocol. We can do our modeling protocols. We had connections we brought in those AI agents, right. We're starting to also see consumer-based AI browsers hit the market as well. Those are really focused on the consumer.

The challenge though is what do we do with trusting the data? We need to make sure it's useful so that builders and developers can access backend data. But they need to have governed access to the apps. They need to have governed access to the data. It all needs to be audited in a secure and safe way. What I mean by audited is we need to see what they did. What did they click on? What data did they go to? Where did they move the data? Did they move it from this system to a different system?

Unfortunately, in the world of AI agents, when that access is granted, it doesn't know much about the role of the user. You might have somebody that has access to the HR system and they could query that. They could say, "Hey, what's the CEO's social security number?" Of course it's got access to that data. It could deliver that data, but the context of the role was missing, right. That's when it starts to get complicated. The one thing between you and those roles is today an Enterprise Browser.

Today, an Enterprise Browser governs activity and controls access to resources. Passwords are a good example of this. As I mentioned before, I don't mind giving somebody my OpenTable account password. Maybe I don't give them my bank account password. I certainly don't give them the keys to the corporate accounts. But that's where we are today. We need those credentials in order for AI agents to work.

So how do we do this without giving those credentials to AI? The answer is the Enterprise Browser. That's the way to do it because that's where credentials can be injected. That's where data can be secured so it's never exposed visually and never exposed on the wire. It can be sent securely. That's a big advantage for everything we need this to do.

This is the future. When we start talking about agentic AI and you start seeing demonstrations and future plans for this, this is where the rubber meets the road. It has to be done in the browser. We've already built this technology and enabled it in a way that's safe and secure. Of course, we need to have collaboration, and this is very different than consumer AI agents. Again, as I mentioned before, these are useful and great. These are great for the individual, not for the role within the corporation.

When we think about the enterprise-grade requirements that we have, this is where we need to step it up. We need to ask what role is that user responsible for, what do they do within the company, and can they have access to that data. That's where those Enterprise AI agents really come into play. Today, we are talking about the Island AI agents as something that we have introduced. This lets teams create, share, and run secure, reliable agents in the browser. That's the place where you need them. This is where we can control the data protection. This is where we can automate work across the businesses and keep the data protected. This is agentic AI in an Enterprise Browser. Today we're calling it AI agents.

Just to show you how we've built this, there is a trust layer, and some of these are components within Island. The Island components of private access, getting you to those network resources, are already built in. The components of data protection and those layers are already built in. When you create a model, it's very easy to create that model with a set of skills that are repeatable. In order to do this, it's really easy. You just do it inside of a browser. You hit the record button and go through a series of steps.

Some of those steps might be manual intervention. You might have somebody that needs to do something or approve something. After all, they've got access to the keys to the kingdom and everything else. So we need to make sure we have that audit trail and that visibility. Having those actions and that human oversight has to be done where humans work, and again, that's inside of that browser.

We've built this with control policy into that hardened Enterprise Browser. It can access storage, secure storage. What's secure storage? It could be your cloud storage. Can you get there with Chrome and Edge? No, you used to, but now you have the Enterprise Browser. That's your only way to get there. That's also the agents' only way to get there. We've created that closed-loop system for that agentic AI to take over and be able to do the functions that are needed for the company, and of course, that full audit trail. This includes screenshots, mouse clicks, where they typed, what they put in, what came back out, and what the screen looked like. Full audit, full capability for that.

When we have this governance and control, you have to think about who's going to use it. We've got the mode set up so that a builder can build what they need to build for an agentic AI within the browser. Again, do the recorded replay. We have the access to the applications. Those are running in a browser, whether they're PWAs or just basic web apps or maybe WebAssembly apps.

Then of course, what can they do? Which button can they click on? Which button can they never click on? That's an important thing. If you don't want an agentic AI to accidentally click the big red button that deletes twenty instances of your EC2, that's going to be a challenge. That happens in a browser. That can be an agentic AI configuration. So that configuration is important. Can they just read the data or can they delete it? Now we have that ability to give that oversight and deploy with confidence.

We can deploy this in a number of ways. Obviously, the Island browser can be run in a remote browser isolated browser environment. This is a locally running browser as well. Most of the activity that you do is going to run on the endpoint. It's going to run locally with your system on your network, with your data, with your credentials. That's the productivity that we really wanted to see, and we can integrate with Bedrock for that as well.

Full integration with AWS Bedrock means that you can keep your data stored within the Bedrock system, and it's obviously built for scale there.

Amazon.com's Success Story: Saving 80 Hours Weekly with Automated Workflows

So what did Amazon.com do? Again, I mentioned that we have a customer testimonial, and I was hoping Adam could be here to talk about it. What they were able to do is they found a way to use Island's AI agents to build real productivity within some of their key systems. They did this in a few ways, but the whole point of this was that it was easy, and they put the power in the hands of the folks that actually knew what they were doing. These are folks that know it's a repetitive task. If they're doing the same thing over and over again, they know that they could save time. So let's put the power of that automation into the hands of the folks that could use it, and let's make sure it's governed and safe.

This is how it's done. It's pretty simple when you think about a recorded replay. If anybody has been in IT as long as I have, you probably heard of Mercury Interactive. We used to record and replay scripts like Selenium scripts, COM, CORBA, DCOM, and all these different protocols. It's kind of like that, but super easy because it's built into the browser. All we do is prompt the user with what workflow to build and include a set of skills along the way. Those skills could be click on this, search for this, do this button, take this PO number and copy it and paste it over here. That's productivity that you can quickly achieve across applications. Again, this all happens within that governance of the browser. You validate it, you edit the flows if you need to, you add human interaction steps where needed, and then of course you share that with your department and increase productivity.

These are the three use cases, and I do want to have questions here in just a minute, but these are the three use cases that they actually put to work. Amazon.com and their marketing teams and a few others are doing some really cool automations already as Island customers. What that means is they're automating purchase orders and the updating of a lot of these manual tasks that are happening. The social media team and the marketing team are doing things across different platforms, moving data, and adding tags to certain areas so that they can be quickly monitored a little bit better with those tags. Again, all things that they would just be naturally doing in the browser, we can automate that and make sure that they're now governed and safe. That's that tag creation there at the bottom.

In particular, I like this solution. Adam really wanted to showcase this himself, but the challenge here on this purchase order use case was that they were spending 30 minutes per day for 1,000 users. So 30 minutes per day for 1,000 users today at Amazon right now are being saved because they've implemented this agentic AI solution with the Island Enterprise Browser. Basically what it is, it's updating permissions and copying and pasting PO numbers. Before it actually gets executed, before they actually pay that bill, they get a little prompt that says, "Okay, we did all this work. Is this the bill you want to pay?" They hit yes in the browser, and then it executes that transaction and actually pays the bill. That's 80 cumulative hours per week, which is an amazing impact over 1,000 users. That's a big deal.

That's what I wanted to show today. Obviously you can see the power when you think about agentic AI. If you're not thinking about what it can do within a browser context, you're not considering it thoroughly because this is where the data is. This is where you have access to the data. Now with the Island Enterprise Browser, you've got the governance, the control, the audit, and you have that capability just naturally built into a single ecosystem for that app delivery mechanism. We'd love for you to come see a demo. This is something that I could talk about for three days and you still wouldn't get it. If you come see it, I guarantee you it's not just a browser. It's a true enterprise delivery workspace. We'd love to give you a demo and walk you through it.