In this interview during theCUBE's coverage of AWS re:Invent, Han Xiao, vice president of AI at Elastic and former chief executive officer of Jina AI, joins theCUBE’s Rob Strechay to unpack how Jina AI’s technology is reshaping the Elastic ecosystem. Xiao explains how Jina’s search foundation models – specifically embeddings, rerankers and small language models – serve as the "brain" behind Elastic’s orchestration framework. This integration aims to solidify Elastic as the essential computational layer for search, enabling developers to build highly accurate, multimodal and multilingual systems that are critical for powering the next generation of agentic AI.
The conversation delves into the nuances of "context engineering," which Xiao describes as the art of optimizing the information fed to Large Language Models (LLMs). He details how small language models are increasingly utilized to compress context and rerank passages within massive token windows, ensuring LLMs receive the most relevant data without unnecessary noise. Xiao also highlights that Jina AI will become the default model provider for the Elastic Inference Service (ELSER), streamlining the developer experience by providing immediate access to state-of-the-art tools for building robust search and retrieval workflows.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
AWS re:Invent 2025. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for AWS re:Invent 2025
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for AWS re:Invent 2025.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
AWS re:Invent 2025. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to AWS re:Invent 2025
Please sign in with LinkedIn to continue to AWS re:Invent 2025. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Milin Desai, Sentry.io
In this interview during theCUBE's coverage of AWS re:Invent, Han Xiao, vice president of AI at Elastic and former chief executive officer of Jina AI, joins theCUBE’s Rob Strechay to unpack how Jina AI’s technology is reshaping the Elastic ecosystem. Xiao explains how Jina’s search foundation models – specifically embeddings, rerankers and small language models – serve as the "brain" behind Elastic’s orchestration framework. This integration aims to solidify Elastic as the essential computational layer for search, enabling developers to build highly accurate, multimodal and multilingual systems that are critical for powering the next generation of agentic AI.
The conversation delves into the nuances of "context engineering," which Xiao describes as the art of optimizing the information fed to Large Language Models (LLMs). He details how small language models are increasingly utilized to compress context and rerank passages within massive token windows, ensuring LLMs receive the most relevant data without unnecessary noise. Xiao also highlights that Jina AI will become the default model provider for the Elastic Inference Service (ELSER), streamlining the developer experience by providing immediate access to state-of-the-art tools for building robust search and retrieval workflows.
In this interview during theCUBE's coverage of AWS re:Invent, Milin Desai, chief executive officer of Sentry.io, joins theCUBE’s Christophe Bertrand to discuss how the company has evolved from simple error monitoring to a comprehensive code monitoring and reasoning platform. Desai highlights Sentry’s massive growth to over 150,000 customers and explains how their focus remains squarely on the developer. He details how Sentry is leveraging AI to transform the debugging process, moving beyond detection to "fixing it faster" with a 95% accuracy rate in identifyi...Read more
exploreKeep Exploring
What updates can you provide about the current status and focus of Sentry as a company?add
What kind of services does Sentry offer for debugging and monitoring applications?add
What advancements has Sentry made in improving the accuracy of identifying root causes of issues?add
What is the primary function of Sentry and how does it support developers in fixing software issues?add
>> Hello, everyone. Welcome to this CUBE conversation, where we're going to be talking a lot about code and developers. And I'm very, very pleased to be joined today by Milin Desai, who is the CEO of Sentry, sentry.io. So it's been a few years since we've spoken here at the Cube with you. I think 2020 was the last time. You are now a very successful, well-established player in the space, 150,000 users, or companies, I should say, over $200 million in funding, which is always nice, and some great investors. And you are present geographically around the world. So can you catch us up on where you're at with the business?
Milin Desai
>> Great to be here. It has been a while, and a lot has changed. At the same time, not much has changed. So what's not changed about Sentry? Our focus. Our focus continues to be on the developer. That is the only and primary persona that we continue to serve. And the area that we operate in, which is code monitoring. So we focus on when things break in code versus, say infrastructure. And that's what makes us unique, our focus on developers and code. So that's something that has stayed constant over the course of these past five years. What has changed, as you said, is the scale. Over 150,000 teams are now using Sentry. And we started in this area called error monitoring. When your code errors, Sentry would tell you everything about it, the deep context, the environment variables, the stack trace, everything to help you fix it. And we've expanded from there now to collect all kinds of telemetry that helps you debug. So when code breaks, can you debug very easily? So we've gone from error monitoring to interpretation of APM, application performance management, that is user-centric. So we collect spans, we do logs, we have session replays, all kinds of telemetry that helps you debug the user, the experience they have with your app across any platform, across any framework, web, mobile, gaming consoles, agents now. It doesn't matter. Front-end, back-end, pick your language and framework. So pretty exciting on that front. Code breaks, that mission continues to stay. And now with AI, especially in the past year and a half, fix it faster. So the ability to take this deep context that Sentry has around what's broken, to then apply it with AI gives you 95% accuracy in root cause. So that's kind of the closed-loop that our customers have wanted, and it's been really exciting. So code breaks, fix it faster, at scale now, doing a lot of things in terms of collecting telemetry, fixing things. That's the new Sentry. Super excited, and the momentum is real.
Christophe Bertrand
>> Well, so let's talk about that, because you mentioned that now you can essentially get 95% there, which means the job of the developer is kind of changing a little bit in a good way, right? And we'll talk about AI, and in this case, I guess AI in support of Sentry. And then we'll talk about the other side of the coin. So I guess the high value that you expect from your developers can now be fully maximized because you've essentially taken away some of the easy stuff from their daily operation, because AI is helping you operationalize some of that stuff. That's maybe not as high value, right?
Milin Desai
>> Yes. I mean, I think if you take the use of AI tooling for developers, as you said, it is definitely helping the individual developer be more productive, right? There are tasks that you can just send off now with background agents come in, do the work, and you come back, and you see the output. There are small micro-tasks like fixing a bug, as I mentioned, with Sentry context, that goes extremely fast that you can delegate to an agent. So I definitely think it is helping kind of make, I would say a developer more productive, more efficient. They can go faster, but it doesn't change the role of what I call the developer in terms of being a builder, designing, architecting, figuring out what the flow is of the information or the code flow. It does do a bit of coding now, but the core element of the developer, which is architecting, designing, implementing system-level software, doesn't change much.
Christophe Bertrand
>> Right. So it feels like essentially it's solving some productivity issues, which is great. It's maybe doing some coding, which is, I guess okay, but more importantly it's allowing the developer to really maintain their core function. So in this context though, isn't the developer of the future, the developer that really knows how to maximize tools like yours that actually leverage AI and be smart about it?
Milin Desai
>> I mean, this is something, I think this past year, early spring we asked not just the developers, but all of Sentry saying you need to become AI-native. So developer, an accountant, our legal officer, everyone has been experimenting, figuring out how AI could help them get better, be more productive. And this is not just about Sentry. I think irrespective of what organization you're in, you do need to become AI-native. And so to answer that specific point around developers, 100% there was this hype around developers will no longer be needed. And we continue to hire. We continue to build. We're going to probably build more. And with AI, it enables us to go faster, right? It allows us to take the tedious away and focus on the more important stuff. So imagine having... We always had a capacity problem building software. Some of that capacity is brought on with AI, which helps a software team go faster, focus on the most important things. The easy things can be delegated to agents, a Sentry agent that does a root cause and a bug, goes and finds a pull request with the code. You review it. You ship it, right?
Christophe Bertrand
>> It's like having a junior developer work for you, and it's like, okay, agent, go do this, go do that, which is really cool. And I think it's really changing in many ways the development cycle because of that, right?
Milin Desai
>> It's getting faster. At the same time, it brings its own set of challenges because you don't understand the code fully. It probably is not documented well. So those kind of challenges in a completely AI-driven world where understanding code when issues happen become harder, but these are all growing pains, but it is the way forward. I expect every developer to be AI-assisted, and that's going to result in great productivity and great experiences in terms of software that is built and products that are built in general.
Christophe Bertrand
>> Right. And you've mentioned or suggested you're doing this internally, so you started sort of eating your own dog food or following this best practice yourself internally. Can you tell us more about how you are applying AI internally? Because I'm very curious about that.
Milin Desai
>> So at a very fundamental level, the first thing which is really interesting, is pretty much all the leading model companies out there or the best names that you can think of as AI companies use Sentry today for the fundamental thing, which is what we do, which is find issues in a software with context. And so what we ended up doing at Sentry is for our developers of course, using that context, can we help close that... Okay, you found the issue. Here's the stack trace. Here's all that information. Can we help fix that issue? And that's where we put our focus initially on, is closing that loop with Sentry at the helm of it, that extra context that we have from production, using that to fix that issue as a closed loop. We call it Seer, which is our reasoning platform, which helps you provide, here's the root cause, here's the pull request with the code. That pull request can also be delegated to your favorite coding agent. We're starting with Cursor, but more to follow.
Christophe Bertrand
>> That's very interesting. So you mentioned reasoning. So this is a very different conversation than just observability or error fixing or detection. So it's a very different world. I remember, well, a few years ago when people were retooling for Y2K, so it's been a minute. I remember the software labs, as we used to call them, were really struggling with analyzing thousands and millions of lines of code. And there was really no AI, and you had to fix it by hand, and you had to hope. Yet, you caught all of those data fields in there so nothing would blow up at Y2K. Very different world, but this world now is the world of AI. So I'd like to maybe switch and pivot the conversation to how you are going to support your existing customers as they are embarking upon this, well, maybe perilous journey with AI. And one of the first questions I want to ask you is around data and around resiliency. So what are you doing in your solution to really support resiliency, not only at the code level, but at the data level, making sure that your own solution does not get attacked, does not let anybody in. And I'm not talking just access control. I imagine that's well-covered. I'm really talking about some very sophisticated attacks that could happen, maybe AI-driven themselves. What do you do to help? Do you have any tools or detection or best practices you put in place? Because I know that it's already started. People are already poisoning models. They're already attacking inference models and trying to derail the efforts. So what are you doing at the code level?
Milin Desai
>> So without getting into the specifics of the tooling, at a very simplistic level, our strategy is that the context that we collect belongs to the customer. And when you start with that premise, everything is about privacy and trust. And so with that context, the system was always at Sentry built with this idea that the data that comes in belongs to the customer. It's scrubbed. It's taken care of in terms of all the PII and all that information. So that's step number one. When it comes in, we don't train on our customer's data in that regard. So even the conversation where we are taking that context for a specific customer and saying, here's a slowdown that Sentry has detected, we take just that relevant context, work with the LLMs in the background, and essentially it's a very binary, two-way communication, not something that is happening across customers and where we are doing things like learning across them and other aspects. That's something very specific around us, where we take privacy very seriously. We take data considerations very seriously. And so every customer's data is isolated in that regard. And when we do essentially, even for AI purposes, leverage that data, it's contextual within that realm with the provider in that regard. So those are the guardrails that we have had in place from the very beginning, and we've just expanded those controls beyond the basics of code breaks or data observability to now even AI.
Christophe Bertrand
>> So yeah, I think there's going to be a ton of potential there as a market, as unfortunately those attacks continue, and those best practices become more critical. So it feels like the persona of the developer, while maybe not fundamentally changing, now may in the context of AI and because of the consequences of what could happen, they may have to be more security-conscious than they ever were before. You mentioned agents. So in your solution, you have built agents that you delegate to. Obviously, it's all going to be about agentic AI, maybe with hundreds of agents out there making decisions, talking to each other in those customer environments. How does your platform evolve to support agentic AI? Or does it have to change at all?
Milin Desai
>> Agents are not going to work in a vacuum. They're going to be part of the broader software stack, where even when a chat interface eventually has to talk to a backend to get data in that regard. So we see this as just another interface that we will work through. And because of our architecture, which is we focus on code, we are agent-less. And so essentially agent-less in the sense that Sentry doesn't need an agent to instrument you, we essentially see all the traffic that the agent is generating in terms of communication within itself or with other aspects of your software. So an agent interfacing with a web or a mobile interface to go to a backend or a middleware is all traced, it's all logged, and it's all understood within Sentry. And specifically as it relates to AI stuff, prompt and what went in and out, in terms of the prompts that themselves are available for monitoring within Sentry. So basic, I would say agent observability is just a natural extension for the core Sentry platform. We had to align with the SDKs that are out there, the software development kits, that are out there from the open AI's of the world, and we are off to the races. So it's been a very natural extension. And the beautiful thing about this we are finding out is that they don't operate in a vacuum. They're part of a bigger software stack. We are already on that part of the world. So that interaction is a closed-loop interaction across everything.
Christophe Bertrand
>> Yeah, I think is a very interesting point you're bringing up. And for our viewers, I really want to double-click on what you just said, which is that you can and already are monitoring all of the interactions between inputs, prompts, outputs, what agents do, what they talk about with each other, and you log it all. And that's going to be the most critical piece in establishing compliance moving forward. Or if something goes wrong, what went wrong? Why it was a bad decision taken? It could be bad data. It could be lots of things, but you can prove and rerun essentially the interactions just as if you were providing some sort of video surveillance of actual employees and what they talked about, what they did.
Milin Desai
>> So it's interesting, the early conversation around an agent's observability was it needed to be something separate. So I want to separate two parts of that conversation. You brought up a very good point. One is what you would call what's happening within the agent, which we would call observability or understanding the interactions. I think that is something that we provide out of the box at Sentry. Then there's the evaluation of the prompts, whether it's through that interaction or at a later point as models change, which is an evaluation platform. And I think the industry right now is in debate whether that is a product or every company will have to build its own evaluation platform themselves. So I think where we have drawn the line, saying we will help you with all your interactions as it relates to agents and monitoring, what's happening in the system level, when it comes to evals, that's not something we do. And we believe that we'll figure out what the next step over there is? Is that a product? Or every company's going to have to figure out how to eval their agents, whether it's across models or the prompts and the behaviors associated.
Christophe Bertrand
>> Well, one thing is for sure, and it could be called a causability platform, you will have to be able to demonstrate causability and what happened in a bigger sort of scheme of things. So there is definitely going to be a very important part for you to play. Again, we'll see what the implementation of that is. You are going to be at AWS in the next few weeks. So is there anything we should know about what you might be announcing or any big highlights for us there?
Milin Desai
>> I mean, just like everyone else, we will be at AWS. And I think what's happened over a course of the past year and a half is, as I said, Sentry has essentially expanded its value play in terms of finding when things are broken, especially with code, the breadth of things we are covering over there, so coming out and letting folks know that. And I think the fix it faster part is a newer part of Sentry. And so we are pretty excited to talk about, not only Seer, which helps you root cause your issues and software much, much faster, at 95% accuracy, but we are also in the business now of preventing bugs. So because of the production context we have, we will now tell you in your pull request if you're about to ship a bug. And so that's been in beta now. We have thousands of organizations using it. We are catching hundreds of thousands of bugs right now, preventing them, not catching them, preventing them from getting shipped, which is a whole different value play. So the whole reasoning platform that is driving the fix it faster is what we are really excited about, and we will unleash it or talk about it more going forward, starting with re:Invent.
Christophe Bertrand
>> Right. So reasoning, not observability, may be a broader spectrum here type of solution and, of course, a very big role to play in building the AI infrastructure for your clients as you yourself leverage AI. Well, this was a great conversation. I'd like to thank you so much for joining us today.
Milin Desai
>> Thank you for having me.
Christophe Bertrand
>> And to our viewers, thank you so much for joining us. My name is Christophe Bertrand, principal analyst here at theCUBE Research. Thank you.
