Name: Introducing Unify AI Assistant: First Look
Uploaded: 2026-04-08T17:51:07.519Z
Duration: 31 min
Description: Introducing Unify AI Assistant: First Look

Transcript for "Introducing Unify AI Assistant: First Look": Alright. Hello, everyone. Welcome to our AI assistant webinar. We're gonna give it a few minutes while everyone logs on. In the meantime, feel free to let us know where you're logging in from today. My name is Shelly Eisen Liebene. I'm part of the product marketing team, and I will, be one of your hosts today. So, as we wait for more folks to join, let's go over a few housekeeping rules. Please feel free to use the chat during this session and send us any questions you might have. At the end of the walk through and presentation, we will have time for q and a with our lead engineer for the AI assistant, Jeffrey Mone. And if we're not able to answer all of these questions today, we will make sure to follow-up right after. The last thing I'll say about that is that, please note there's the chat and there's the q and a section. So the questions registered in q and a will be answered first. So make sure to put them in there. Alright. I see New Jersey, Canada, Texas. I myself from North Carolina, so we have some good coverage. Alright. Hey, Jeffrey. Thanks for joining. Alright. I think, we can go ahead and get started. So during the first part of the session today, we'll have a walk through, by Jordan, our lead PM for the feature, and then we will see you back shortly for more q and a afterwards. And with that, we can go ahead and start the session. Everyone, thanks for joining me here today on this call. If you're here, you're probably somebody that does the heavy lifting in your software delivery processes, helping with, the triage build failures, managing pipelines, onboarding new users, making sure releases are moving forward. And the session today is just for you. We're gonna do something a little bit different. We're diving in and showing you the first beta of a Unify AI assistant that our team is working on, and we're really excited to show it to you as we've been working on this for a little bit. We think it's really going to change the way you work with the platform. So what we're gonna do is we're gonna dive in here. We've got some information for you if you find this interesting later on to be able to sign up for the beta and give it a try and, share your feedback. So real quick, my name is Jordan. I'm a product manager on the Cloudbee's AI tooling team. I've been in the software world for, like, over twenty five years, working with developers as a developer myself for most of that time. And the past few years, I've really focused around AI experiences, AI tooling, and excited to share some of that with you here today. With me is also my tag team partner, Shelley, and our marketing crew. She's going to jump in with the live q and a later on as well. So save your questions as we go along. So we're going to, today focus around first, what's the problem we're looking to address here? And we're gonna get into our approach with Unify AI and where we're going. We're going to then spend our time around what is this AI assistant, really get into a demo and show you some scenarios around how you can work with it. And then we will wrap up going through how you can get access to the beta and then on that live q and a that I mentioned. So the problem we're looking to solve. Well, this is our old friend, contact switching. So you may be familiar with this. A bill breaks, critical security issue comes up. Somebody's even just got a question about your release system. And you're spending a whole bunch of time really having to switch through tabs, like in going through Jira issues or in Confluence, look helping you look for answers. You're going through documentation. You may find that one. They're like, there. That doc looks like exactly what I need. And then you find out that it hasn't been updated for, like, over a year. Then there's always also that one person. We've all worked with them. They've been in the company for ages. They know everything. But when they're out on a long vacation, when they are not around, suddenly you're stuck with, yeah, how do we solve this thing? And this can have a real impact on your teams. Not only is it difficult to possibly have an issue come up in the middle of the night and suddenly, it turns into an all hands at 3AM diving into an issue, or you're even having to pull some of your top engineers off of other big important roadmap items that your team is facing. And diving into an issue often to get started, this can even take, like, 20 to 40% of your time to just get context. If it's a build that failed, you're spending that time out of trying to figure out, okay, where's the problem? Where do I get started here? You may be switching through a bunch of different tools to figure out the problem or working with AI that doesn't quite have all the context to help you figure out where to go. And way back, I don't even maybe, what should be a simpler part of the process, even just onboarding into the system for somebody can take weeks rather than days. And that's just valuable time that's being eaten up that could be used for, other more important things. So we really wanna help you, say, if you're the release manager, be able to answer that question. Is it safe to ship? Get to an answer in, like, thirty seconds rather than having to take a long meeting. And a lot of these things, they aren't, you know, hypothetical situations. I think we've all been through this at some point. I hear it from customers regularly. And this is really what we're after looking to help out with here. So our approach with Unify AI is is built in the same strong basis we have with Unify where it acts as a single observability layer across the different tools that your systems use, your teams use, whether it's Jenkins or GitHub Actions or GitLab, working across delivery quality and security. With Unify AI, our goal is to also meet you where you're working. So we're very intentional on this top line here. If you like to work in an IDE or, you know, you're trying to just, you know, spend a lot of time in Slack communicating, organizing, coordinating or in the Unify UI itself directly. And we wanna be able to meet you there with AI experiences. And what enables this is with our, kind of central we call the context and action layer. And this is a part of the system that provides a shared context like information about your different builds and, the status of your security policies across your organization. This has, APIs that you work with through tools, whether if it's a AI assistant. You might be using MCP servers to get communicate back through Unify. And the system also defines guardrails to help protect these actions. So especially when you're working with AI, we wanna make sure that your interactions with the system are safe, protected, and finally, also auditable. So you can trace back any action toward who did it and when. So now the bottom layer of the stack, the goal is to help build out a system that can help you plan your intent in tackling an action, respond then to that either in triaging or answering problems, and eventually getting down to even executing, being able to deliver action straight in your application. So AI is moving into the delivery pipeline fast to the pressure to adapt. It's definitely very real. And there's a meaningful difference between AI that knows your environment and AI that doesn't. So without the context of your system, your policies, your history, your stack, AI isn't really assistance. It's more of a guess. And in a regular regulated environment, we know that a wrong guess is an audit finding. In a production pipeline, it could be an outage. And this is why we're building the Unify AI assistant a bit differently. So it's not a standalone tool that you separately switch to, but something that lives and integrates inside the platform where your work is already getting done. It's got full context of your pipelines, your policies, your delivery history before it takes even any action. So in practice, the goal is that engineers get unblocked without leaving their screen. New team members can stop waiting on the person who knows the system to start finding answers themselves. When something goes wrong, investigations happen right where the problem's at, not in a separate tool that several context switches away. And we're going for faster teams helping with onboarding and answers quickly all without adding yet another thing for you to have to maintain. So let's switch now over into demo mode, and I'll show you three demos here. One, kind of simply what the assistant is, how it helps with in context navigation for tasks onboarding. We'll take a look at how we could set up a new feature flag with the help of it in just a few seconds. And then finally, we triage a failed build in the system to help us get started while we go down the path of diagnosing in real time. So here we are in the UI of the Unify application. We're running in our preproduction environment. So, you know, these are the real early bits that we have under development. You may see some debug, menus or so. Just know that's, you know, further prove this is the real deal. So imagine if you will, I'm an engineer on a team and I've just been asked to add a new repo to the system. Now I could go ahead and go through the steps through the UI myself or look in docs if this is my first time and figure that out. But I'm going to use the assistant to just opened up here. It's available on every screen in the Unify application. I'm just going to ask it to help me with this task. Help me add the scheduling service repo to the system. Now the assistant understands the Unify application and our setup, and it was able to navigate me to the right point of the application to start configuring my repo to, add it in through GitHub. And that will guide me through the steps here. I can go with, you know, click on the next step to configure the repo and, again, be given instructions on how to set this up. But I actually just now got a new request for something else the team needs immediately, which is a new feature flag. So I'm gonna pause for a moment on adding this repo. Now, again, this is a live environment. There was a little bit of a surprise, so we're gonna go ahead and switch to the right organization for this project. Okay. And I'm going to ask the assistant to help me set up this feature flag. So help me create a scheduled transaction feature flag. The assistant understands that I'm actually working with a feature flag related question. It's going to the feature management section of the UI, and now it can help me. From here, it give me some instructions on how to take that next step to create my feature flag. I'm going to go ahead and use the suggestion and actually say, create a new flag. And you'll see that it opens up the form, and it can fill in the name for me with what I typed before, and then away we go. So I do happen to know that this environment has a little bit of a glitch right now. And if I click save, we're about to have an error. So we're going to cancel out of this before we move on to the next request that just came to me. So the final thing, suddenly, I was just asked about was a build failure that's going on, and can I help investigate what is happening there? So we'll go back to the assistant. Help me look at the latest runs for the web portal project. And the assistant again is following along with me. It's not bugged by all this context switching. It knows I'm looking for something build related, takes me to the runs, and sure enough, I can see the list of failed builds that are going on right now. So something bad is definitely happening. I'll click into this run, and I see there's some errors in here that I could go ahead and read through. But, oh, as a final piece, let's just ask the assistant to help us out. Help me troubleshoot why this run is failing. Now the assistant has knowledge of my pipeline, my setup, the configuration here, and it's able to take a look at the log files it's seeing on the screen and give me some details about what may be going on. Again, the development environment, so ignore the error that just came up. But it did spot here that there appears to be some issues with our feature flag setup. And it will give some suggestions for next steps on things I could do to fix those. So now in the span of just a few minutes, we've actually been able to bounce through three different tasks of onboarding, the feature flag creation, to troubleshooting a build. I didn't have to jump around to other applications or other sources. We had a few little surprises like that error from the dev environment, but that's okay. These are live bits. And, yeah, I was able to stay in context within the Unify application without having to go to external tools to accomplish these tasks. So one final thing I'll call out in the UI here for any of you who are interested in joining our beta and starting to try this out. We have this little section down here that says, was this helpful? The thumbs up and thumbs down. These tools, if you report your, findings to us through these, these go right into our systems to help us tune and tailor the responses from the assistant to work better for you. So please make sure to help us out and provide feedback using these. Okay. So now you've seen the assistant and what it can do. This is a little different from many other AI tools that are destinations you have to leave your work to go to and say copy an error message and ask questions without specific system context that then can give you a real generic answer. The AI assistant in Unify works the other way around, meeting where you're already at and already knowing what you're looking at. With focus around it, having the, context of your system, understanding the pipeline state, your test results, and help provide specific answers that have specific knowledge. It's not just a generic web search or a search over static docs. And the system gets more tailored to your environment over time as you are putting more info into the system, like, you know, more build, further defining your release processes, and using the system itself. And finally, it works within the guardrails of your existing system, providing guidance that is both, auditable as well as governed and secure. So now I wanna give you a bit of a sneak peek on the road map. What you're seeing in the demo today was our phase one beta where we focus on accelerating with guidance, recommendations, and navigation, not yet taking autonomous actions. And then we're starting here on purpose instead of jumping to autonomous agents because in a delivery system, you wanna make sure that you get it right. A wrong action code made an outage or a broken release or maybe even a security breach. So we don't wanna skip this trust building phase. We want you to see the assistant make good recommendations consistently before we ask you to let it act on its own. Looking ahead, we move on towards these governed agents with trusted agentic delivery on things such as triaging bill, the test security issues, helping you with executing delivery health reports, or even setting up a self healing system. So with that, we'd love to invite you to join the beta, be a part of our customer feedback program where you can get a direct line to myself and others in our product team to help give us feedback, and that helps shape the road map to work well for you and your organization. And we'll also give you some dedicated onboarding session time here to get going with the system, answer your initial questions. And all we ask in return is for your feedback and using it on real workflows, not just our demos, but giving us structured feedback in the way you want to work with it and how it's working for you. We'll ask for a couple of sessions to check-in with you, get some, directed questions with you, and, dive into your feedback a bit, and may ask for some permission to cite some learnings that we can share with others along the way. So if you're interested, here's the link to sign up now. And with that, we will move on to the live q and a portion where I will hand this off to my friend Shelley to help take your questions as you drop them into the chat below. Thanks so much. Alright. That was a great walk through by Jordan. Thank you so much for that. With me here is Jeffrey Won, like I said, lead engineer for the AI assistant, and we are excited to take some questions. If you did not have a chance to type in your question in the q and a section, go ahead and do it now. While Jeffrey is answering some of the questions we already got, I will take a look at the chat and make sure we did not miss anything. And if there's anything that's left unanswered or you have more questions popping up later, please reach out to either your account manager if you're a Cloud based customer, or contact our sales team, and we will be happy to assist. With that, let's take a look at the list at the list we have coming in. Alright. I I see the first question coming in is about which AI model is being used to to power the assistant. Great question. So, we we are currently using, Cloud SONNET for our base model for all our input prompts, but, it is also possible that we would be using Haiku, which is the lighter model, to handle lighter request, downstream as well. So it it is gonna be, most likely gonna be SONET, but it might become hybrid, handling between multiple different models. Yeah. Thank you. And and, you know, with all of the conversations, going around in the market these days about the cost of running AI products, the the the impact of token economy on our CapEx and OPEX. When you think about your infrastructure, that kind of optimization around which kind of actions can be done with, you you know, certain capabilities versus other can come in handy, not only in speed but also in cost. So this is definitely something that we're seeing a lot of practitioners and thought leaders talking about. The next question coming in, there's a question about if a future version will support on prem AI models, if if you have any visibility, into that today, Jeffrey. So that is not in discussion yet. But, if we were to make decision to make AI assistant available on prem, then, yes, on prem model would be supported in that case, but, that hasn't been discussed technically yet. Alright. So sounds like, again, with the beta, we're starting with the UniFi SaaS interface. And, based on the learnings and progress, we'll take it from there. That is correct. is helpful. We have another similar questions around, whether this is compatible with, all of the cloud is all of the cloud based products, CI, CDRO, feature management. I think a good way to look at it, and Jeffrey, I I will hand it over to you in just a sec, is, is within within Unify today, within, the SaaS interface today, where and and how is the assistant available could be a good context to start with. Yeah. Great question. So AI assistant, as Shelley has mentioned, is available on every single screen on Unify, and it is able to scan for the screen to understand what is going on on the screen. But for, for security purpose and some other, technical limitations, we have hidden away some of the elements, from the assistant from the unified front end. But you can assume that assistant is available on most of the pages, and it is able to understand the contextual, information present on Unify page. I think if we were to go a step further, can can Unify Assistant touch external applications like GitHub or Jira. It is unable to do that, currently, but that is a planned feature for, for q two and q three. Yep. Thank you. Yeah. Thank you for that. We have another question again with the complete compatibility of existing tools, our prospects and customers have to date. And so, can the assistant work with Jenkins information, GitHub actions? So currently, even even with the GitHub Actions Jenkins, if there's a if there's a Unify workflow file included in the repo, that generates a workflow within Unify, which can can which can directly interact with, which which AI assistant can directly interact with. And as you see on the demo, it can directly read in the logs to triage what is going on with errors, any issues during the build process, things like that. So, yes. So if if the unified workflow is integrated to the repo, then, yes, GitHub actions and Jenkins would be supported through, AI assistant. Yeah. Thank you. This is very helpful. I wanna move on to questions around, security, trust, amount of access or level of access that the AI assistant has in into, into accounts. So I want to ask you first, how are we handling any prompt injection or security threats when it comes to the design of the AI assistant? Great question. There are two there are two layers to it. Currently, from the from the, from the back end layer, we do the first level scan to see if the input prompt is injecting, is has any malicious intention included in the prompt. But there's also a second layer that we've leveraged from AWS guardrail, which is a AWS provided, catch all kind of default security layer that we build, right before LLM to make sure no, malicious prompt gets injected to, to the, to the LOM at the end, to cause any trouble or or expose any unintended data from our back end. So, yeah. So there are two layers that are there, to to prevent such security concerns currently. Thank you so much for this. I see we have someone struggling with the application form. So what I'm gonna go ahead and do is put in the chat another link to reach our teams and maybe start the conversation from there. One second. I wanna make sure that we answer it in a timely manner, and then we'll move on to the next. Alright. Gonna put it in chat. Another way to reach us here. Perfect. Okay. Moving on. Let's see what else. Can you tell us more in terms of, you know, level of access the AI system has into the account's information, to the system. That's a that's a good question. So, currently, if the user so AI assistant runs with the user, it's essentially a partner in crime within Unify. So, whatever data that the user who's who's using AI assistant, whatever data that user has access to, theoretically, AI assistant would have the access to the same information that user is exposed to. There are currently an internal discussion to see if we can limit the AI assistant visibility if the user has too much access to the application. So there that is a bit of a security gray area that we're trying to, discuss and and see if we can come up with a better solution, currently. Yeah. So, again, the the role based permissions that we have on the Unify level now travels into the AI assistant Yes. operation. Now, Yep. I think there is a similar question around, again, access to data or using data. That's by Joan. Can I feed my Unify Assistant with external data before incorporating in my internal systems? Depends on the format of the data. So if if the if the data current currently, AI assistant only only, accepts text inputs, It might accept image or video input or or some other reference data for multimodality. But, if if you want to feed in external data in tabular format into current AI assistant, it'll be able to analyze on the data and answer your question if the data is limited in size. But if it if it's a huge CSV file that you want to you want to introduce a assistant to, there might be a bit of challenge, but that is a that is a technical gap that we can easily resolve down the line. Yeah. Thank you so much. Just taking a quick look at the chat and q and a. I think we have time for one more question. Let's talk a little bit about, what would the automation behind everything that the AI assistant is designed to help do would look like in terms of the audit trail that it may leave, explainability, provenance, that aspect of of the design. Great question. The visibility and audit trailing was actually the first thing that we had in mind when we started architecting the Assistant. We have a separate service called Trajectory that sits together with Assistant to essentially track every session and every, every session and every, I guess, every work session with the user to make sure that we trace we we make an audit trail of every request that comes from the user to check whether, the request was handled successfully, whether there was an issue, whether there was a feedback as as it was shown in the demo. So we we we have built in, as much visibility as possible when it comes to any kind of, any kind of interaction that user has with AI assistant through the usage of trajectory and, and the session logs and session monitoring tools that we've built into that service. Thank you so much for going through this. It's just, really mind blowing for me to see how, you know, after AI has impacted cogeneration so quickly, now it's starting to reinvent or disrupt or reimagine what the DevSecOps, you know, practice is. is becoming. And so when you think about governance, when you think about, auditability, there are different flows or different mechanisms or even different features, capabilities that are being built especially to manage that. So thank you for shedding more light on that. We are at time. I want to thank you all for joining us today, for your engagements, for the questions. I wanna thank Jordan for the walk through and, of course, Jeffrey for being here with us and answering questions. Please feel free to reach out to our team to get access to the beta, to get access to the program in which you can have more time with Jeffrey and Jordan, to maybe talk through how you can ingest, some mock data before you start with, you know, real data or anything else that is on your mind. So thank you very much. Have a great day, and we'll see you soon. Bye bye. Thank you.