The SDLC You Build Yourself — Andy Wong, Andrei Bocan & Shane Wolf, Atlassian

Welcome, everyone. Welcome, everyone. Welcome to the AI ​​Engineer World Fair Welcome to the AI ​​Engineer World Fair Welcome to the AI ​​Engineer World Fair . . . Thank you for participating in today's workshop Thank you for participating in today's workshop Thank you for participating in today's workshop . Let me briefly introduce myself . Let me briefly introduce myself . Let me briefly introduce myself . I am Andy Wong . I am Andy Wong . I am Andy Wong . I have been working at Atlas for four . I have been working at Atlas for four . I have been working at Atlas for four years, and I have been working years, and I have been working years, and I have been working in the AI ​​field for two and a half years in the AI ​​field for two and a half years in the AI ​​field for two and a half years . I have . I have . I have two excellent team members with me today two excellent team members with me today two excellent team members with me today . We are . We are . We are Shane Wolf and Andrew Shane Wolf and Andrew Shane Wolf and Andrew Bellicon, Dev AI Division Managers. Today, we will look at how agent Bellicon, Dev AI Division Managers. Today, we will look at how agent Bellicon, Dev AI Division Managers. Today, we will look at how agent systems are changing the software development systems are changing the software development systems are changing the software development lifecycle lifecycle lifecycle and and how orchestration is changing everything how orchestration is changing everything . . . While Andy is talking, While Andy is talking, While Andy is talking, I will go around handing out numbered papers I will go around handing out numbered papers . It might sound a bit strange, but you will understand it better once the workshop starts understand it better once the workshop starts . Could you please . Could you please repeat that? repeat that? Yes, Andy will go around Yes, Andy will go around Yes, Andy will go around handing out papers with numbers on them handing out papers with numbers on them handing out papers with numbers on them . It might feel a little awkward at first, but you will . It might feel a little awkward at first, but you will understand it much better once the introductory presentation is over . Well then, let's . Well then, let's . Well then, let's quickly start with the first introduction slide quickly start with the first introduction slide quickly start with the first introduction slide .

. . Since this is the AI ​​Engineers World Fair, we need to Since this is the AI ​​Engineers World Fair, we need to Since this is the AI ​​Engineers World Fair, we need to focus on implementation, so let's focus on implementation, so let's quickly skim through these slides quickly skim through these slides quickly skim through these slides . I will . I will briefly briefly introduce everything that Atlassian has developed over the past year and a half, and provide an introduce everything that Atlassian has developed over the past year and a half, and provide an introduce everything that Atlassian has developed over the past year and a half, and provide an overview of all the tools that Atlassian will offer overview of all the tools that Atlassian will offer . As . As you all know, AI you all know, AI agents have revolutionized the software development agents have revolutionized the software development agents have revolutionized the software development lifecycle lifecycle lifecycle . . . Over time, as the number of AI tools increases, the Over time, as the number of AI tools increases, the Over time, as the number of AI tools increases, the effect of productivity improvement is actually effect of productivity improvement is actually effect of productivity improvement is actually showing a decreasing trend showing a decreasing trend showing a decreasing trend . So, as I . So, as I . So, as I mentioned last time, we believe we can capture the greatest opportunities across mentioned last time, we believe we can capture the greatest opportunities across all areas on both the left and right sides of the code all areas on both the left and right sides of the code . In other words, it includes planning , orchestration, the code review , orchestration, the code review , orchestration, the code review process, and even the operations part process, and even the operations part . . Therefore, we have focused on developing aspects that are not Therefore, we have focused on developing aspects that are not clearly apparent in the software development life cycle clearly apparent in the software development life cycle . We have developed . We have developed tools such as teamwork graphs, as well as tools such as teamwork graphs, as well as various agents that can be utilized on both sides of the code various agents that can be utilized on both sides of the code . Examples include AI Planner, . Examples include AI Planner, . Examples include AI Planner, AI Auto Dev, Reviewer, and SRE AI Auto Dev, Reviewer, and SRE AI Auto Dev, Reviewer, and SRE . At . At . At the center of all this is something called an AI context machine the center of all this is something called an AI context machine the center of all this is something called an AI context machine . After . After . After briefly looking at these tools briefly looking at these tools briefly looking at these tools , what we will mainly be , what we will mainly be , what we will mainly be testing today is the Jira Coding testing today is the Jira Coding testing today is the Jira Coding Agent. This is an Agent. This is an Atlassian proprietary coding agent with in-depth organizational knowledge of the Atlassian ecosystem Atlassian proprietary coding agent with in-depth organizational knowledge of the Atlassian ecosystem . So, this agent

. So, this agent is an enterprise context agent that deeply understands everything within the Atlassian ecosystem. You can read Jira tickets, is an enterprise context agent that deeply understands everything within the Atlassian ecosystem. You can read Jira tickets, Confluence documents, as well as emails and Confluence documents, as well as emails and Confluence documents, as well as emails and Slack messages to gather all the context needed to Slack messages to gather all the context needed to Slack messages to gather all the context needed to plan and implement a specific Jira ticket plan and implement a specific Jira ticket . You can interact with pull requests . You can interact with pull requests within the Jira ecosystem directly within the Jira coding agent within the Jira ecosystem directly within the Jira coding agent . In other words, you can maintain and edit everything within the cloud. You can also assign tasks directly to specific agents. If you do not want to use the Jira coding agent, you can also assign Jira tickets to your preferred agent, such as Claude, Cursor, or CodeX to your preferred agent, such as Claude, Cursor, or CodeX . Finally, . Finally, there is also an SDLC automation feature.

there is also an SDLC automation feature. there is also an SDLC automation feature. As you know, Jira itself has As you know, Jira itself has As you know, Jira itself has built-in automation processes built-in automation processes built-in automation processes . If specific conditions are met, . If specific conditions are met, code code can be generated using the Jira coding agent or another agent selected by the user. can be generated using the Jira coding agent or another agent selected by the user. can be generated using the Jira coding agent or another agent selected by the user. In terms of code, we utilize AI reviewers In terms of code, we utilize AI reviewers In terms of code, we utilize AI reviewers . This provides functions similar . This provides functions similar . This provides functions similar to tools already widely used in the industry to tools already widely used in the industry to tools already widely used in the industry . I intend to . I intend to implement a review viewer agent that reviews pull requests to find bugs, ensures compliance with standards, and validates requirements. Therefore, implement a review viewer agent that reviews pull requests to find bugs, ensures compliance with standards, and validates requirements. Therefore, the review is not limited to the code itself the review is not limited to the code itself the review is not limited to the code itself . We are building a system that collects all contact information from Jira, Confluence, . We are building a system that collects all contact information from Jira, Confluence, . We are building a system that collects all contact information from Jira, Confluence, other SCM tools, as well as Slack, to other SCM tools, as well as Slack, to comprehensively understand the implementation details required for code pull requests and provide reviews . In addition, it provides standard features that are centrally managed. Every team and organization has its own standards that it wishes to apply to each repository and organization has its own standards that it wishes to apply to each repository . This is a . This is a somewhat cumbersome task, but somewhat cumbersome task, but somewhat cumbersome task, but we aim to support administrators and we aim to support administrators and we aim to support administrators and software engineers in applying these standards to software engineers in applying these standards to software engineers in applying these standards to all pull requests within a specific repository all pull requests within a specific repository much more easily through standards much more easily through standards . In addition, this

. In addition, this . In addition, this standard works standard works standard works on all SCM platforms, such as GitHub, GitLab, and Bitbucket Cloud, and on all SCM platforms, such as GitHub, GitLab, and Bitbucket Cloud, and can be applied to pull requests. Another goal that Atlassian pursues through AI reviewers is to reduce unnecessary information generated throughout the code review process. The existing code reduce unnecessary information generated throughout the code review process. The existing code review process tends to make overly review process tends to make overly review process tends to make overly detailed criticisms or provide detailed criticisms or provide too many low-value comments too many low-value comments too many low-value comments . . . At DVI, we have been At DVI, we have been At DVI, we have been training and filtering several models in-house training and filtering several models in-house . We utilized these models to filter out low-value or unnecessary comments filter out low-value or unnecessary comments , and to filter out unnecessary comments from the pull request comments themselves , and to filter out unnecessary comments from the pull request comments themselves . . As a result, we confirmed that the overall comment As a result, we confirmed that the overall comment As a result, we confirmed that the overall comment resolution rate improved resolution rate improved resolution rate improved . As . As . As you can see in the table below, you can see in the table below, you can see in the table below, using this model, we achieved a code resolution rate of about 40% using this model, we achieved a code resolution rate of about 40% using this model, we achieved a code resolution rate of about 40% , which is , which is very close to the human baseline of 45% very close to the human baseline of 45% very close to the human baseline of 45% . And as . And as . And as always, you always, you always, you can measure the effectiveness of these tools can measure the effectiveness of these tools can measure the effectiveness of these tools . At Lasting, we have . At Lasting, we have built dashboards for various tools to check the number of reviews, the built dashboards for various tools to check the number of reviews, the number of agents executed, and to number of agents executed, and to identify the value of each tool within DVI identify the value of each tool within DVI . Finally, I would like to . Finally, I would like to . Finally, I would like to introduce a few products that I have recently implemented

introduce a few products that I have recently implemented introduce a few products that I have recently implemented . One of them . One of them . One of them is a tool called AI Auto Dev. You will is a tool called AI Auto Dev. You will easily be able to tell what function it performs just by looking at the name easily be able to tell what function it performs just by looking at the name . This feature analyzes the Jira board to . This feature analyzes the Jira board to . This feature analyzes the Jira board to autonomously autonomously autonomously identify actionable tickets. identify actionable tickets. After applying all standards and initiatives, After applying all standards and initiatives, create a pull request that is ready for review create a pull request that is ready for review create a pull request that is ready for review . All of these processes . All of these processes . All of these processes can be executed within the Jira ecosystem can be executed within the Jira ecosystem can be executed within the Jira ecosystem . You can configure various settings, such as maximum monthly credit . You can configure various settings, such as maximum monthly credit . You can configure various settings, such as maximum monthly credit usage and the number of automatic development run tickets per user, and you usage and the number of automatic development run tickets per user, and you can also specify whether to run only in specific repositories or in all repositories within the Jira organization. In addition, it provides a dashboard showing the history of all automated development executions, how they work, the number of automated development pull requests requiring review, the number of pull requests meeting requirements, and more. We are striving to provide as much metadata as possible so that engineers and teams can understand the overall performance of automated development striving to provide as much metadata as possible so that engineers and teams can understand the overall performance of automated development . Automated development is also . Automated development is also . Automated development is also integrated with Slack integrated with Slack . Once the . Once the . Once the automatic development feature is integrated into the repository itself, it automatic development feature is integrated into the repository itself, it starts automatically processing specific work items depending on whether there are available work items or inputs required to complete them. A more detailed analysis of ROI is

more detailed analysis of ROI is available at any time. We will always available at any time. We will always strive to determine the ROI of the entire automated development function by measuring as many aspects as possible. Finally, I will talk about AI talk about AI talk about AI . Oh, I'm sorry . Oh, I'm sorry . Oh, I'm sorry . We have all . We have all . We have all experienced difficulties, and we know very well how difficult it is to diagnose and solve problems, especially at 2 or 3 in the morning experienced difficulties, and we know very well how difficult it is to diagnose and solve problems, especially at 2 or 3 in the morning . AI aims to analyze the root cause of an issue by re-identifying the full context from previous records in systems such as Jira and Confluence when such failures occur, and to provide suggestions on how to resolve them provide suggestions on how to resolve them . This . This . This feature is also integrated with Slack feature is also integrated with Slack feature is also integrated with Slack . So you can . So you can . So you can quickly summarize and quickly summarize and quickly summarize and understand what is happening, and determine whether it is understand what is happening, and determine whether it is understand what is happening, and determine whether it is affecting many external customers affecting many external customers affecting many external customers or only a few or only a few . In addition, we will explain the cause of the problem . In addition, we will explain the cause of the problem and the solution in detail and the solution in detail and the solution in detail . As you can see . As you can see . As you can see , we provide root cause analysis , we provide root cause analysis , we provide root cause analysis , predict the likelihood of problems occurring, and , predict the likelihood of problems occurring, and suggest various solutions.

I have talked a lot about the importance of context so far , and I will explain how helpful context is to agents how helpful context is to agents how helpful context is to agents . What we are . What we are . What we are developing is an AI developing is an AI developing is an AI context agent. context agent. context agent. Well, many of us Well, many of us Well, many of us deal with vast amounts of repositories deal with vast amounts of repositories deal with vast amounts of repositories , and often all that context is , and often all that context is , and often all that context is scattered either within the repository itself or across scattered either within the repository itself or across fragments of documents scattered throughout the workspace . The goal of the AI ​​context engine is to . The goal of the AI ​​context engine is to collect all these contexts and collect all these contexts and generate standardized contexts so that all agents can run more efficiently generate standardized contexts so that all agents can run more efficiently . I have . I have used the word "context" a lot, and this is the part we have been used the word "context" a lot, and this is the part we have been considering most important recently considering most important recently considering most important recently . We intend to . We intend to . We intend to collect all possible contexts to collect all possible contexts to build an engine that can drive agents more powerfully in the future. As you can see build an engine that can drive agents more powerfully in the future. As you can see , through the AI ​​Context Engine, , through the AI ​​Context Engine, , through the AI ​​Context Engine, all tools such as Jira and Confluence, as well as all tools such as Jira and Confluence, as well as all tools such as Jira and Confluence, as well as existing instances, will existing instances, will run using semantic code search, and will be able to provide the necessary context for all agents that will be used in the future provide the necessary context for all agents that will be used in the future . What . What we intend to do is make various tools, such as plugins, various tools, such as plugins, various tools, such as plugins, accessible accessible accessible . The goal is to enable engineers to . The goal is to enable engineers to simply add tools to browsers, IDEs, or CLIs and use them immediately .

. As I mentioned, since this is the AI ​​Engineer World As I mentioned, since this is the AI ​​Engineer World As I mentioned, since this is the AI ​​Engineer World Expo, I tried to keep the explanation Expo, I tried to keep the explanation Expo, I tried to keep the explanation short. We short. We want to focus on building. So, I will provide you with a simple workshop where you can try out the Jira Coding Agent firsthand provide you with a simple workshop where you can try out the Jira Coding Agent firsthand . You will see Shane . You will see Shane . You will see Shane carrying a specific number carrying a specific number carrying a specific number . I will share that number . I will share that number with you with you with you . That number . That number is your account number. is your account number. To put it simply, this number is To put it simply, this number is To put it simply, this number is linked to your login information. The linked to your login information. The login information is as follows login information is as follows . I'm sorry, it . I'm sorry, it . I'm sorry, it was a Wi-Fi issue. I guess I was was a Wi-Fi issue. I guess I was n't the only one. all right. I will right. I will return to this screen so that everyone can check the QR code.

Yes, yes, yes. Let's see. Yes, let's Yes, yes, yes. Let's see. Yes, let's see. I will stop screen sharing see. I will stop screen sharing see. I will stop screen sharing . Yes, yes, yes, I will do it . Yes, yes, yes, I will do it . Yes, yes, yes, I will do it this way. great this way. great . yes. It's not working yet. . yes. It's not working yet. The internet is too slow. Let's see The internet is too slow. Let's see . Ah, the Wi-Fi . Ah, the Wi-Fi . Ah, the Wi-Fi is quite slow. is quite slow. is quite slow. Did you connect here? yes Did you connect here? yes Did you connect here? yes ? It ? It ? It didn't connect. You were on the other side of the Wi-Fi didn't connect. You were on the other side of the Wi-Fi . . Ah, it works now. Here you go. I will do it this way . It . It will change now.

will change now. Uh, is Cloudflare still Uh, is Cloudflare still turned on? Yes, try turning it off. turned on? Yes, try turning it off. Ah, it's turned off. It is turned off. Let's see. How does it work? Yes, the How does it work? Yes, the work I am about to do. yes. Hmm, work I am about to do. yes. Hmm, work I am about to do. yes. Hmm, what is it? yes what is it? yes . Yes, yes, it seems like a Wi-Fi issue . Yes, yes, it seems like a Wi-Fi issue . yes. Is . yes. Is n't that possible too?

n't that possible too? No, it's not working. No, it's not working. Yes, my only suggestion is... Yes, Yes, my only suggestion is... Yes, I understand. I understand. I understand. Now, would you like to test the door Now, would you like to test the door Now, would you like to test the door ? ? ? Yes, I'll test mine as well Yes, I'll test mine as well Yes, I'll test mine as well . yes . yes . yes . All right. yes. If you have any questions while connecting to Wi-Fi, please feel free to . All right. yes. If you have any questions while connecting to Wi-Fi, please feel free to ask at any time ask at any time . Ah, yes. . Ah, yes. great. So, while we are connected to Wi-Fi great. So, while we are connected to Wi-Fi , please feel free to ask any questions you may have regarding the future direction of the product line, , please feel free to ask any questions you may have regarding the future direction of the product line, , please feel free to ask any questions you may have regarding the future direction of the product line, currently currently currently offered products, customer- offered products, customer- offered products, customer- preferred features, workflows, etc. preferred features, workflows, etc. Yes, that's Yes, that's good. Yes, please tell me.

good. Yes, please tell me. Could you explain the context engine in more detail? yes Could you explain the context engine in more detail? yes . The context engine is fundamentally . The context engine is fundamentally connected to various graph databases within Atlassian, enabling you to manage all data that customers connected to various graph databases within Atlassian, enabling you to manage all data that customers input into Atlassian products input into Atlassian products . All products . All products such as Confluence, Jira, and Bitbucket are such as Confluence, Jira, and Bitbucket are interconnected. These connections can be established through inference while the user is using the product, or they be established through inference while the user is using the product, or they can be set to be explicitly connected can be set to be explicitly connected can be set to be explicitly connected . For example, . For example, . For example, when working with Jira tickets, the ticket is created based on the when working with Jira tickets, the ticket is created based on the when working with Jira tickets, the ticket is created based on the specification document in Confluence specification document in Confluence . In addition, these products are . In addition, these products are integrated with third-party products integrated with third-party products integrated with third-party products . We . We . We also plan to integrate with services like Notion or Slack also plan to integrate with services like Notion or Slack also plan to integrate with services like Notion or Slack . All these things . All these things blend into one large context blend into one large context blend into one large context . Ultimately, what we are currently . Ultimately, what we are currently . Ultimately, what we are currently building is a building is a gene search system based on all these contexts . This method is also . This method is also widely used within Atlassian widely used within Atlassian widely used within Atlassian . We proceed with . We proceed with . We proceed with development on one side of the platform ecosystem development on one side of the platform ecosystem , and in doing so, we need to retrieve contextual information such as , and in doing so, we need to retrieve contextual information such as platform documentation provided by other teams platform documentation provided by other teams . Alternatively, projects being carried out on the other side of the organization carried out on the other side of the organization

may affect the part currently under development. All of this information is All of this information is reflected in the current agent coding session through the context engine. reflected in the current agent coding session through the context engine. For example, for Shane, who is currently working on Wi-Fi-related tasks, For example, for Shane, who is currently working on Wi-Fi-related tasks, connectivity connectivity is a key factor. is a key factor. Shane conveys this information to the team members . One . One . One area where we achieved particular success was the area where we achieved particular success was the launch of a new code search product. Thanks to this product, in launch of a new code search product. Thanks to this product, in the past, many teams the past, many teams tried using monorepos so that agents could always utilize context information, but tried using monorepos so that agents could always utilize context information, but now that is now that is now that is no longer the case. That has a few downsides no longer the case. That has a few downsides , and it , and it , and it means you have to build various tools. means you have to build various tools. means you have to build various tools. The speed also slows down. The speed also slows down. The speed also slows down. Internally, the Internally, the Internally, the process of choosing a monorepo was essentially equivalent to accepting the time and effort required to process of choosing a monorepo was essentially equivalent to accepting the time and effort required to process of choosing a monorepo was essentially equivalent to accepting the time and effort required to build the tools for the monorepo build the tools for the monorepo . You end up . You end up spending a lot of time building the tool spending a lot of time building the tool spending a lot of time building the tool . However, thanks to the code context . However, thanks to the code context . However, thanks to the code context engine, the product repository can now be engine, the product repository can now be engine, the product repository can now be used like a monorepo used like a monorepo used like a monorepo . Because you . Because you . Because you can call in the engine immediately when needed can call in the engine immediately when needed can call in the engine immediately when needed . The speed is also incredibly . The speed is also incredibly . The speed is also incredibly fast. It enables code-level access fast. It enables code-level access fast. It enables code-level access . Thanks to this, it becomes possible to . Thanks to this, it becomes possible to access potentially outdated logic. The access potentially outdated logic. The Confluence document is just a Confluence document is just a Confluence document is just a snapshot of an event that happened once

snapshot of an event that happened once snapshot of an event that happened once , but the actual code remains , but the actual code remains , but the actual code remains intact. intact. intact. Accessing code with high throughput Accessing code with high throughput and utilizing the search function is really and utilizing the search function is really and utilizing the search function is really useful. I'm sorry, I useful. I'm sorry, I useful. I'm sorry, I explained it too verbosely. explained it too verbosely. explained it too verbosely. Did that answer your question Did that answer your question ? Yes, that's good. Do ? Yes, that's good. Do ? Yes, that's good. Do you have any other questions? Please tell me. you have any other questions? Please tell me. Well, to rephrase the question, the Well, to rephrase the question, the first question was whether the context could be first question was whether the context could be first question was whether the context could be created in real time created in real time created in real time . Ah, . Ah, . Ah, what was the second question? Ah, what was the second question? Ah, how do I resolve crashes? Yes, that's how do I resolve crashes? Yes, that's how do I resolve crashes? Yes, that's good. It seems there will be good. It seems there will be good. It seems there will be many presentations on memory at the conference many presentations on memory at the conference many presentations on memory at the conference , and it appears that , and it appears that one of the core topics of agent memory is one of the core topics of agent memory is one of the core topics of agent memory is memory itself.

memory itself. memory itself. Agents acquire knowledge from various sources Agents acquire knowledge from various sources Agents acquire knowledge from various sources , but ultimately, a , but ultimately, a , but ultimately, a person says, "This is true. This is person says, "This is true. This is person says, "This is true. This is true by chance." Or "It is true by chance." Or "It is true by chance." Or "It is true in this particular case." You need to true in this particular case." You need to true in this particular case." You need to confirm that. In confirm that. In confirm that. In terms of obtaining information, terms of obtaining information, generally, generally, snapshots are not taken to some extent. Therefore, I snapshots are not taken to some extent. Therefore, I think it is important that information can be utilized in real time. Since it's the latest version, you can think it is important that information can be utilized in real time. Since it's the latest version, you can use it right away. use it right away. use it right away. However, if you cache it, the However, if you cache it, the However, if you cache it, the information will actually become outdated, information will actually become outdated, information will actually become outdated, which can lead to various complex which can lead to various complex which can lead to various complex problems. Well, problems. Well, problems. Well, in terms of authority, the in terms of authority, the way we have historically handled it has been for the person in charge to directly say, "Yes, it way we have historically handled it has been for the person in charge to directly say, "Yes, it has been verified. This content has has been verified. This content has higher authority than other content." It was to enable one to say that higher authority than other content." It was to enable one to say that . . . However, However, However, this method is more difficult in the case of Jira work items this method is more difficult in the case of Jira work items this method is more difficult in the case of Jira work items . Of course, Jira work items are . Of course, Jira work items are possible, but it results in possible, but it results in possible, but it results in much more manual work.

much more manual work. There is a risk that automation will actually lead to the automated system gaining authority, which is a very strange There is a risk that automation will actually lead to the automated system gaining authority, which is a very strange situation. Because the very situation. Because the very situation. Because the very concepts of automation and authority concepts of automation and authority concepts of automation and authority can clash in the same situation can clash in the same situation can clash in the same situation . Is the internet connected . Is the internet connected . Is the internet connected ? ? ? I think so. I assume you are connected to the internet by now I think so. I assume you are connected to the internet by now . yes. It seems the page is . yes. It seems the page is working properly for everyone now working properly for everyone now working properly for everyone now . I would like to reiterate that I . I would like to reiterate that I will upload the Go language and QR code right now. Well, will upload the Go language and QR code right now. Well, to briefly explain it again, everyone will have a to briefly explain it again, everyone will have a to briefly explain it again, everyone will have a number. That number is number. That number is used for login. used for login. used for login. We We We plan to use Atlassian University. plan to use Atlassian University. plan to use Atlassian University. Therefore, Therefore, Therefore, the number you have must include the number previously provided by Student and Shane the number you have must include the number previously provided by Student and Shane . That . That number is the login. number is the login. number is the login. With that login, you can test the Jira coding With that login, you can test the Jira coding With that login, you can test the Jira coding agent and try out various things agent and try out various things agent and try out various things within the Jira ecosystem within the Jira ecosystem within the Jira ecosystem . Now, please . Now, please go to the conference page. You will be able to find the Jira go to the conference page. You will be able to find the Jira test workspace link there test workspace link there test workspace link there . You can . You can . You can write code there as you wish write code there as you wish write code there as you wish . Well, . Well, . Well, there is a starter repository for you there is a starter repository for you . This is a place called Atlassian Swag within Bitbucket . This is a place called Atlassian Swag within Bitbucket . This is a place called Atlassian Swag within Bitbucket . I will . I will . I will show you shortly, but once you all show you shortly, but once you all show you shortly, but once you all log in, please feel free to log in, please feel free to log in, please feel free to use it and try out the agent use it and try out the agent use it and try out the agent yourself.

yourself. yourself. For example, "The web network is For example, "The web network is For example, "The web network is not working yet." not working yet." not working yet." You said that, right? You said that, right? You said that, right? Are you connected to the same network? Yes, we Are you connected to the same network? Yes, we Are you connected to the same network? Yes, we are on the same network. You are on the same network. You must be connected to the same network. It must be connected to the same network. It took some time to load. yes. yes yes took some time to load. yes. yes yes took some time to load. yes. yes yes . Yes, yes, yes. I will speak with the IT manager to . Yes, yes, yes. I will speak with the IT manager to see if there is a solution. We apologize for the inconvenience. Yes, would you like me to apologize for the inconvenience. Yes, would you like me to upload the QR code again upload the QR code again upload the QR code again ? Yes, it's fine. Ah, ? Yes, it's fine. Ah, ? Yes, it's fine. Ah, numbers, numbers, numbers. Yes, I'm sorry numbers, numbers, numbers. Yes, I'm sorry numbers, numbers, numbers. Yes, I'm sorry . Did Shane . Did Shane . Did Shane take the numbers? take the numbers? take the numbers? sorry. sorry. sorry. Where is Shane's number cup? Where is Shane's number cup? Where is Shane's number cup? Shane's Number Cup. That Shane's Number Cup. That is a good question. wait a minute.

More numbers are needed in the number cup More numbers are needed in the number cup . The best number. It . The best number. It . The best number. It seems like a prime number. Probably seems like a prime number. Probably seems like a prime number. Probably . It seems like no one can connect to Wi-Fi . It seems like no one can connect to Wi-Fi . It seems like no one can connect to Wi-Fi . . . Yes, yes, yes. I Yes, yes, yes. I think it was an unsolved problem for 2026, but if there is anything that will think it was an unsolved problem for 2026, but if there is anything that will save us from Artificial General Intelligence (AGI), save us from Artificial General Intelligence (AGI), save us from Artificial General Intelligence (AGI), it would be it would be it would be Wi-Fi. Wi-Fi. Wi-Fi. So, if the Wi-Fi is resolved, So, if the Wi-Fi is resolved, So, if the Wi-Fi is resolved, we we won't need to be saved, so it should be fine. Because we won't need to be saved, so it should be fine. Because we need to process and optimize traffic need to process and optimize traffic need to process and optimize traffic . Yes, do you have any other questions until the Wi-Fi issue is . Yes, do you have any other questions until the Wi-Fi issue is resolved resolved resolved ? Yes, please speak ? Yes, please speak . Regarding the verification requirements. sorry . Regarding the verification requirements. sorry . Regarding the verification requirements. sorry . My question was to ask if we could . My question was to ask if we could . My question was to ask if we could talk a little more about the verification requirements talk a little more about the verification requirements talk a little more about the verification requirements . Well, . Well, . Well, basically, let's assume that, for example, basically, let's assume that, for example, basically, let's assume that, for example, when there is a specific task item, the when there is a specific task item, the when there is a specific task item, the Jira coding Jira coding agent processes that item and has an "approval criteria" checklist . And after the Jira coding

. And after the Jira coding agent generates code for the task agent generates code for the task agent generates code for the task , it , it verifies these specific details once it determines that it is complete. Because the Jira Coding Agent runs in a complete sandbox environment, you can Because the Jira Coding Agent runs in a complete sandbox environment, you can install and install and install and run what is needed in the repository and perform all necessary run what is needed in the repository and perform all necessary tasks. If tasks. If tasks. If the repository the repository the repository only has instructions like "Execute like this," you can only has instructions like "Execute like this," you can only has instructions like "Execute like this," you can open a browser and run the corresponding verification open a browser and run the corresponding verification open a browser and run the corresponding verification step. step. What is a sandbox? Where is the sandbox? Where is the sandbox? Is it hosted? Yes, that is correct. yes Is it hosted? Yes, that is correct. yes Is it hosted? Yes, that is correct. yes . Uh, . Uh, . Uh, or is it something we can bring or is it something we can bring or is it something we can bring ? There is ? There is no such feature yet. no such feature yet. So, it is like Atlassian's So, it is like Atlassian's So, it is like Atlassian's own sandbox. Since it is own sandbox. Since it is own sandbox. Since it is a container, you can a container, you can a container, you can run anything you want. The run anything you want. The run anything you want. The cursor is also activated, and everything else cursor is also activated, and everything else cursor is also activated, and everything else necessary is available. necessary is available. necessary is available. Yes, that's good. Yes, that's good. Yes, that's good. Just regarding payments. yes Just regarding payments. yes Just regarding payments. yes . yes. It . yes. It seems like you have to use your own key. If desired, can you allow all seems like you have to use your own key. If desired, can you allow all agents to run AI using API keys agents to run AI using API keys , or , or is it entirely an internal hosting model?

is it entirely an internal hosting model? is it entirely an internal hosting model? How do I pay now? How do I pay now? How do I pay now? Shane, am I mistaken Shane, am I mistaken Shane, am I mistaken ? ? ? You use an API key. yes You use an API key. yes You use an API key. yes . So, we are using our own API key . So, we are using our own API key . So, we are using our own API key . You . You can connect all agents to Anthropic or elsewhere can connect all agents to Anthropic or elsewhere . All right. . All right. . All right. However, it does not cost anything However, it does not cost anything . . . Yes, the Jira Coding Agent supports several models Yes, the Jira Coding Agent supports several models . We currently support Anthropic . We currently support Anthropic . We currently support Anthropic models, Open AI models, and more models, Open AI models, and more models, Open AI models, and more . I do . I do . I do n't really remember if Gemini supports it n't really remember if Gemini supports it n't really remember if Gemini supports it . . . Claude models and Open AI models Claude models and Open AI models Claude models and Open AI models definitely support it. definitely support it. definitely support it. And fees are charged in tokens based on usage And fees are charged in tokens based on usage . Yes, it . Yes, it . Yes, it is a credit system. It is is a credit system. It is is a credit system. It is slightly different, but basically the slightly different, but basically the slightly different, but basically the same method. You are same method. You are same method. You are charged based on usage. charged based on usage. Is it a per-user payment system? Is it a per-user payment system? In addition to the per-user subscription fee, are In addition to the per-user subscription fee, are In addition to the per-user subscription fee, are charges based on usage charges based on usage charges based on usage ? Yes, that is correct. Are ? Yes, that is correct. Are ? Yes, that is correct. Are all these features all these features all these features available right now? available right now? available right now? Jira Coding Agent Jira Coding Agent Jira Coding Agent has already been released. It has been available for quite a long time has already been released. It has been available for quite a long time has already been released. It has been available for quite a long time . Yes, please try . Yes, please try . Yes, please try using it. You using it. You mentioned credits on the slide mentioned credits on the slide , specifically developer , specifically developer , specifically developer usage credits. I am usage credits. I am wondering if I need to report at the end of the month, from a business and accounting perspective, where and for what tasks these credits were used, or if I can just report them at my own discretion. I'm not sure if I remember correctly, but I not sure if I remember correctly, but I think I need to check again.

think I need to check again. think I need to check again. sorry. The question was about how to report credits sorry. The question was about how to report credits at the end of the month when using AI credits in various situations . . If my memory serves me correctly, it seems that If my memory serves me correctly, it seems that If my memory serves me correctly, it seems that credits are separated by product credits are separated by product credits are separated by product . For example, if a specific . For example, if a specific . For example, if a specific user user user used a specific JCA (Jira Coding Agent) in a specific experience used a specific JCA (Jira Coding Agent) in a specific experience used a specific JCA (Jira Coding Agent) in a specific experience , it seems that level of , it seems that level of , it seems that level of detail is provided detail is provided detail is provided . . . So, So, So, you mean the aggregated method, not the task-based one you mean the aggregated method, not the task-based one you mean the aggregated method, not the task-based one ? yes. ? yes. ? yes. no. Not yet. Well, no. Not yet. Well, no. Not yet. Well, that is a part we are currently developing that is a part we are currently developing that is a part we are currently developing . Ultimately, . Ultimately, as more and more people use Atlassian coding, it seems that it will become important for the return on investment as more and more people use Atlassian coding, it seems that it will become important for the return on investment (ROI) to be clearly evident (ROI) to be clearly evident . For example, if . For example, if someone invests 1 billion tokens in a task someone invests 1 billion tokens in a task someone invests 1 billion tokens in a task , you have to verify whether that task is , you have to verify whether that task is actually necessary actually necessary actually necessary or if it is just or if it is just or if it is just being executed haphazardly being executed haphazardly being executed haphazardly . I'm sorry . I'm sorry . I'm sorry , but what is the latest Wi-Fi technology like , but what is the latest Wi-Fi technology like , but what is the latest Wi-Fi technology like ?

? ? Do you have any other questions regarding the AGI launch? Yes, please Do you have any other questions regarding the AGI launch? Yes, please Do you have any other questions regarding the AGI launch? Yes, please speak. Wow, I'm sorry speak. Wow, I'm sorry . Log in, Andy. Yes, here it is . Log in, Andy. Yes, here it is . Log in, Andy. Yes, here it is . There will be a link . There will be a link . There will be a link . You can enter it directly . You can enter it directly . You can enter it directly . . You can access atlassian.dvao.atlassian.net. The login You can access atlassian.dvao.atlassian.net. The login ID is atlassian.university's ID is atlassian.university's ID is atlassian.university's student@yournumber and the password is student@yournumber and the password is here zxc zxc. Please here zxc zxc. Please use it at the academy. use it at the academy. The account is reset when the session ends The account is reset when the session ends The account is reset when the session ends . So, if your account is . So, if your account is . So, if your account is open, there open, there open, there are a few questions. it's are a few questions. it's are a few questions. it's okay. okay. okay. thank you. it's okay. thank you. it's okay. I will help you. Wait a moment. I will check Wait a moment. I will check . yes. Oh, dear. oh . yes. Oh, dear. oh . yes. Oh, dear. oh my god. Could you please let me know your account number my god. Could you please let me know your account number my god. Could you please let me know your account number ? Is this your account?

? Is this your account? ? Is this your account? Or is it your account? Or is it your account? sorry. This is my account sorry. This is my account . yes. wait a minute. Which number . yes. wait a minute. Which number . yes. wait a minute. Which number is it? What is your account number is it? What is your account number is it? What is your account number ? Is it number 95? ? Is it number 95? ? Is it number 95? wait a minute. yes. yes. That is a good question wait a minute. yes. yes. That is a good question wait a minute. yes. yes. That is a good question . Um, Andre, . Um, Andre, . Um, Andre, could you tell me a little more now? could you tell me a little more now? could you tell me a little more now? Oh, I'm sorry. The question was whether to Oh, I'm sorry. The question was whether to Oh, I'm sorry. The question was whether to use smart features for model routing use smart features for model routing use smart features for model routing . First of all, the . First of all, the current Jera coding agent supports selecting the model to route to supports selecting the model to route to . Yes, that is correct. You can also decide that . Yes, that is correct. You can also decide that "a specific agent must be used for this specific skill "a specific agent must be used for this specific skill " regarding sub-agents and skills " regarding sub-agents and skills " regarding sub-agents and skills .

. However, as of now, there is no smart routing function to specify a particular model. This is a feature currently under development by AutoDev that supports determining, "This Haiku level is suitable for this specific task, and good results were obtained at this level supports determining, "This Haiku level is suitable for this specific task, and good results were obtained at this level ." We will provide you with the indicators necessary to ." We will provide you with the indicators necessary to make these decisions and determine whether to apply the policy make these decisions and determine whether to apply the policy . all right. Skills Skills can basically be selected directly by the user. can basically be selected directly by the user. can basically be selected directly by the user. We possess a variety of skills, but We possess a variety of skills, but We possess a variety of skills, but particularly in particularly in particularly in the case of the Jira coding agent, there is a certain difference between my role and the case of the Jira coding agent, there is a certain difference between my role and the case of the Jira coding agent, there is a certain difference between my role and other roles (formerly other roles (formerly other roles (formerly called developers) called developers) called developers) . The . The Jira coding agent provides a harness by default. You provides a harness by default. You can select a model at the top level of the harness, and specify the skills required for a specific task through skills required for a specific task through skills required for a specific task through skills .

. . For example, you can set this model for specific tasks For example, you can set this model for specific tasks For example, you can set this model for specific tasks , and configure , and configure , and configure which skills sub-agents which skills sub-agents which skills sub-agents should use should use should use . . . Then, skills will be automatically selected as needed to Then, skills will be automatically selected as needed to proceed with the work. proceed with the work. Yes, that is correct. Do you still use Wi-Fi Yes, that is correct. Do you still use Wi-Fi Yes, that is correct. Do you still use Wi-Fi ? all ? all ? all right. I will right. I will right. I will also continue to use Wi-Fi. also continue to use Wi-Fi. Yes, it seems it will proceed that way Yes, it seems it will proceed that way . Oh, I'm sorry. . Oh, I'm sorry. . Oh, I'm sorry. My question was whether I could keep the demo website My question was whether I could keep the demo website My question was whether I could keep the demo website open for a while and if open for a while and if open for a while and if people could try it out once they connected to Wi-Fi people could try it out once they connected to Wi-Fi . Yes, I think it is a perfectly feasible request. thank you. How long is the website open? We can keep it open as much as we want much as we want . It seems likely that it will remain open during the AI ​​or AI . It seems likely that it will remain open during the AI ​​or AI . It seems likely that it will remain open during the AI ​​or AI tool expo tool expo tool expo . Once you are . Once you are connected to Wi-Fi (we apologize for the inconvenience caused by Wi-Fi issues), you can try out the website and experience various tools experience various tools .

. . One thing to note is that your One thing to note is that your login information is very login information is very securely protected when you enter it on this site securely protected when you enter it on this site securely protected when you enter it on this site . Therefore, do not enter information that is not inconvenient . Therefore, do not enter information that is not inconvenient . Therefore, do not enter information that is not inconvenient to upload to GitHub gist or Bitbucket snippet to upload to GitHub gist or Bitbucket snippet . Well, keep this in mind when adding and tinkering with things. all keep this in mind when adding and tinkering with things. all right. Do right. Do right. Do you have any other questions? Yes, I would appreciate it you have any other questions? Yes, I would appreciate it you have any other questions? Yes, I would appreciate it . . Why is it important to clean up old context? I wonder why it is important to clean up old context why it is important to clean up old context . That's interesting. . That's interesting. . That's interesting. Could you explain it in more detail? Could you please Could you explain it in more detail? Could you please Could you explain it in more detail? Could you please tell me what that means tell me what that means tell me what that means ? ? ? I am part of an organization that has been using Atlassian products for 15 to 20 years I am part of an organization that has been using Atlassian products for 15 to 20 years . However, there are a huge number of . However, there are a huge number of product documents that are no longer related to our product product documents that are no longer related to our product product documents that are no longer related to our product . . . How much of that is like that How much of that is like that How much of that is like that ? ? That's a good question. sorry That's a good question. sorry . That is a good question. . That is a good question. . That is a good question. The point is that there are large organizations that have been using Atlassian for 10 to 15 years The point is that there are large organizations that have been using Atlassian for 10 to 15 years . Well, how important is it to remove such unnecessary information so that it . Well, how important is it to remove such unnecessary information so that it does not appear in search results, does not appear in search results, does not appear in search results, or so that agents do not select or so that agents do not select that content during the process of randomly searching for answers to questions ? In our experience, this has generally not been an issue, this has generally not been an issue, because we prioritize the latest results.

I think the concept of authority applies here. For example, if a document created 10 years ago is still up-to-date document created 10 years ago is still up-to-date , we have added a , we have added a , we have added a feature called the 'Verified' badge to Confluence feature called the 'Verified' badge to Confluence feature called the 'Verified' badge to Confluence . This badge . This badge . This badge internally says, 'Yes, that's good.' This internally says, 'Yes, that's good.' This internally says, 'Yes, that's good.' This content is still the latest information content is still the latest information content is still the latest information and and and should be prioritized over other content. It is should be prioritized over other content. It is should be prioritized over other content. It is used to display as. For used to display as. For example, in the U.S., ' Hello.' great.' There is a Hello.' great.' There is a mark that says... It seems there were mark that says... It seems there were mark that says... It seems there were a few questions regarding the standard a few questions regarding the standard a few questions regarding the standard . Unfortunately, . Unfortunately, . Unfortunately, the demo itself the demo itself the demo itself does not include standard features, but I will briefly show you how does not include standard features, but I will briefly show you how does not include standard features, but I will briefly show you how standards work within the Rover Dev product standards work within the Rover Dev product standards work within the Rover Dev product and and and how to apply them to specific repositories how to apply them to specific repositories . Here is the standard . Here is the standard page. Several page. Several page. Several basic standards are already provided, but basic standards are already provided, but basic standards are already provided, but now I will now I will now I will show you how to create custom standards show you how to create custom standards show you how to create custom standards . There are . There are . There are also various templates to choose from also various templates to choose from also various templates to choose from . I . I . I will create a new standard. Select a new standard will create a new standard. Select a new standard will create a new standard. Select a new standard and and and enter a brief description of the standard.

enter a brief description of the standard. enter a brief description of the standard. This standard is 'Token Enforcement' This standard is 'Token Enforcement' This standard is 'Token Enforcement' . Enter a standard name . Enter a standard name . Enter a standard name and set it to 'Force Token' and set it to 'Force Token' and set it to 'Force Token' . Enter a description . Enter a description . Enter a description and add a rule. and add a rule. After adding standards and rules, you can specify the repositories to apply these standards . You can manually select all repositories . You can manually select all repositories . You can manually select all repositories or or or select repositories that match a specific pattern select repositories that match a specific pattern select repositories that match a specific pattern . When such a repository is . When such a repository is loaded, you can decide, 'Okay, loaded, you can decide, 'Okay, loaded, you can decide, 'Okay, let's apply it to the next repository as well.' let's apply it to the next repository as well.' Once all settings are complete, you Once all settings are complete, you can create a standard. can create a standard. Software development Software development has its own standards. Instead of enforcing it manually has its own standards. Instead of enforcing it manually has its own standards. Instead of enforcing it manually or within the PR process itself or within the PR process itself , if you , if you , if you set standards and set standards and set standards and configure the repositories to apply them, Rover configure the repositories to apply them, Rover configure the repositories to apply them, Rover Dev Jira code—that is, the coding Dev Jira code—that is, the coding Dev Jira code—that is, the coding agent— agent— checks for compliance with those standards checks for compliance with those standards and adds comments while running the pull request process.

and adds comments while running the pull request process. Some other features I wanted to demonstrate include the AI ​​Planner and AutoDev. I cannot and AutoDev. I cannot and AutoDev. I cannot show you a perfect demo, but show you a perfect demo, but there is something I would like to show you in the future. We are moving in the direction of moving from left to right in the code from left to right in the code , and AI Planner and , and AI Planner and AutoDev can be said to be closer to the left side of the code AutoDev can be said to be closer to the left side of the code AutoDev can be said to be closer to the left side of the code . Well, it . Well, it . Well, it isn't a perfect demo, but I isn't a perfect demo, but I think I can briefly explain it to you through a slideshow. First, let's start with AutoDev think I can briefly explain it to you through a slideshow. First, let's start with AutoDev . Andre, you . Andre, you have invested a lot of time in the development of AutoDev. Could you briefly explain the have invested a lot of time in the development of AutoDev. Could you briefly explain the main goals you intend to achieve with AutoDev main goals you intend to achieve with AutoDev main goals you intend to achieve with AutoDev and how it works and how it works ? ( ? ( While installing the microphone) I need to While installing the microphone) I need to learn how to use the microphone learn how to use the microphone . You should . You should . You should n't touch it, right? Are you saying I should n't touch it, right? Are you saying I should n't touch it, right? Are you saying I should n't touch it? Ah, n't touch it? Ah, n't touch it? Ah, I understand. thank you. Yes, I'm I understand. thank you. Yes, I'm I understand. thank you. Yes, I'm sorry. please wait for a moment sorry. please wait for a moment sorry. please wait for a moment . AutoDev is our first step in the field of AI-based STLC . AutoDev is our first step in the field of AI-based STLC . AutoDev is our first step in the field of AI-based STLC orchestration orchestration orchestration .

. . Basically, our approach is to Basically, our approach is to empower the engineering team to focus on feature development by freeing them from all narrative-related work empower the engineering team to focus on feature development by freeing them from all narrative-related work . For example, in . For example, in . For example, in the case of AutoDev, the case of AutoDev, the case of AutoDev, in the initial release, we will focus on verifying that we are proceeding with the right work and adhering to the correct standards by pulling work items from the backlog, in the initial release, we will focus on verifying that we are proceeding with the right work and adhering to the correct standards by pulling work items from the backlog, selecting items we are confident the agent can handle autonomously, and uploading PRs for the team to review . In addition, we will improve the codebase to be relatively simple so that agents can use it improve the codebase to be relatively simple so that agents can use it . Oh, I'm sorry. all . Oh, I'm sorry. all . Oh, I'm sorry. all right. This is precisely the right. This is precisely the part where we set the direction we intend to move forward part where we set the direction we intend to move forward . As I mentioned before, for . As I mentioned before, for . As I mentioned before, for now, if you now, if you retrieve items from the backlog and retrieve items from the backlog and let us know specifically which items you would like AutoDev to automate, we will help you process them help you process them help you process them . Well, we are considering a few ways to . Well, we are considering a few ways to develop this feature going forward develop this feature going forward . First of all, . First of all, regarding standard compliance, it regarding standard compliance, it seems Andy will discuss this during the review process. We consider standard compliance to be somewhat similar to linting consider standard compliance to be somewhat similar to linting , but view it , but view it , but view it as much more flexible because it is implemented directly by agents as much more flexible because it is implemented directly by agents as much more flexible because it is implemented directly by agents . We enable the

. We enable the application of standards to be applied to future code in advance through a workflow that adds standards to code reviewers application of standards to be applied to future code in advance through a workflow that adds standards to code reviewers . We are . We are planning to analyze the existing codebase through the AutoDev workflow and gradually deploy PRs to ensure all existing code complies with standards planning to analyze the existing codebase through the AutoDev workflow and gradually deploy PRs to ensure all existing code complies with standards . There are several complex elements in that part . There are several complex elements in that part . For example, no one will approve a PR that requires changing half . For example, no one will approve a PR that requires changing half of the codebase due to a new standard, of the codebase due to a new standard, of the codebase due to a new standard, or a or a or a PR with a large amount of code changes PR with a large amount of code changes . Therefore, at AutoDev, we carefully consider how much workload to assign to the team carefully consider how much workload to assign to the team . No one wants to spend their time reviewing 50 PRs a day . No one wants to spend their time reviewing 50 PRs a day . Therefore, "Our team . Therefore, "Our team wants to be assigned a specific PR wants to be assigned a specific PR ." It provides a feature that allows you to set it as ." It provides a feature that allows you to set it as . For example, . For example, . For example, each team member can be set to handle 3 to 5 PRs per week each team member can be set to handle 3 to 5 PRs per week . This . This promotes incremental development and promotes incremental development and allows team members to focus on developing features or other projects they actually want to work on allows team members to focus on developing features or other projects they actually want to work on . This approach can be . This approach can be . This approach can be applied to both projects applied to both projects applied to both projects . I consider this part to be . I consider this part to be . I consider this part to be setting the direction of the organization setting the direction of the organization setting the direction of the organization . This . This commonly happens in large organizations commonly happens in large organizations commonly happens in large organizations , for example, when all , for example, when all , for example, when all members try to switch from a specific version of Rails to members try to switch from a specific version of Rails to another another another .

. . This is a very big job. We This is a very big job. We also operate over 10,000 services also operate over 10,000 services also operate over 10,000 services , and these services are , and these services are quite similar, as are the experiences, memories, and workflows gained in the process quite similar, as are the experiences, memories, and workflows gained in the process . So, I thought we needed a system where one team . So, I thought we needed a system where one team . So, I thought we needed a system where one team manages everything, manages everything, manages everything, creates PRs, agents creates PRs, agents handle tasks, and the person in charge handle tasks, and the person in charge only coordinates only coordinates only coordinates . And I . And I . And I also wanted to leave a record of that process also wanted to leave a record of that process also wanted to leave a record of that process . To . To . To reiterate, this is our reiterate, this is our reiterate, this is our vision. We believe that this vision. We believe that this vision. We believe that this system is system is system is moving in this direction and will be helpful for the workflow moving in this direction and will be helpful for the workflow . . . Large organizations Large organizations Large organizations operate various services, and a operate various services, and a operate various services, and a significant amount of time is invested in the process significant amount of time is invested in the process significant amount of time is invested in the process . . Since engineers do not want to get too deeply involved in that area, I think this is a good opportunity to solve these problems through an agent. Well, one more thing I think this is a good opportunity to solve these problems through an agent. Well, one more thing I would like to mention is that would like to mention is that when AutoDev performs tasks autonomously on behalf of the team, it is important to understand what AutoDev is doing and be able to when AutoDev performs tasks autonomously on behalf of the team, it is important to understand what AutoDev is doing and be able to intervene when necessary intervene when necessary . For example, like when AutoDev selects the wrong part or assigns the wrong task to the appropriate person. AutoDev will when AutoDev selects the wrong part or assigns the wrong task to the appropriate person. AutoDev will learn through these experiences learn through these experiences learn through these experiences . The . The . The reason you need to add these features to the Jira board is that reason you need to add these features to the Jira board is that most people look at the most people look at the most people look at the Jira board first when looking for what to do next Jira board first when looking for what to do next .

. We talked about this part before, didn't we? Well, as I We talked about this part before, didn't we? Well, as I We talked about this part before, didn't we? Well, as I mentioned regarding the limits on ongoing work mentioned regarding the limits on ongoing work mentioned regarding the limits on ongoing work , it is important to ensure that the team does , it is important to ensure that the team does , it is important to ensure that the team does not get overwhelmed by PRs or other tasks, not get overwhelmed by PRs or other tasks, not get overwhelmed by PRs or other tasks, and that they do and that they do not get swept up in work even though it is not something they really want to pursue not get swept up in work even though it is not something they really want to pursue . . . Another thing being considered Another thing being considered Another thing being considered is cost constraints, which I believe is the trend of the times is cost constraints, which I believe is the trend of the times . . When attempting to process tasks in bulk in Autodev, we do When attempting to process tasks in bulk in Autodev, we do not want the personnel to operate indiscriminately. You must be able to limit the weekly spending per ticket . I consider these matters to be . I consider these matters to be basic conditions basic conditions basic conditions . Yes, basically, when you . Yes, basically, when you start working in Autodev, we will start working in Autodev, we will review the overall scope, including the amount of work required review the overall scope, including the amount of work required . Oh, and . Oh, and . Oh, and I forgot to mention this part. I forgot to mention this part. I forgot to mention this part. Yes, requirements are needed. Yes, requirements are needed. When examining the backlog, it is highly likely that there are work items in the backlog that are work items in the backlog that are work items in the backlog that are not clearly defined not clearly defined . For example, there is an item where someone has written just a single line of summary description saying, "Do this task written just a single line of summary description saying, "Do this task ." These items ." These items ." These items may be valuable, but I am may be valuable, but I am not sure if I can proceed with the work because I do not know what the exact goal is. This is exactly when you not sure if I can proceed with the work because I do not know what the exact goal is. This is exactly when you bring in the person in charge and ask, "We have this work bring in the person in charge and ask, "We have this work bring in the person in charge and ask, "We have this work item, but exactly what item, but exactly what needs to be done? Should the person in charge needs to be done? Should the person in charge needs to be done? Should the person in charge perform this task?

perform this task? Could you help define it clearly?" Could you help define it clearly?" asks. Only when such clear and asks. Only when such clear and asks. Only when such clear and specific details are gathered can the specific details are gathered can the specific details are gathered can the work begin work begin work begin . And at this stage, . And at this stage, . And at this stage, the person in charge will conduct a review the person in charge will conduct a review the person in charge will conduct a review . Well, . Well, . Well, another thing I discovered in this autonomous workflow is that another thing I discovered in this autonomous workflow is that another thing I discovered in this autonomous workflow is that work often does work often does work often does not proceed properly during the review phase and ends up being abandoned not proceed properly during the review phase and ends up being abandoned . It's because people are busy with other things . It's because people are busy with other things . So, . So, . So, we contacted the people in charge at the appropriate time and asked, we contacted the people in charge at the appropriate time and asked, "Could you take a "Could you take a "Could you take a quick look at this part and let us know if it would be better to merge it quick look at this part and let us know if it would be better to merge it quick look at this part and let us know if it would be better to merge it or delete it or delete it or delete it ?" We are building a ?" We are building a ?" We are building a system that allows you to make such requests system that allows you to make such requests system that allows you to make such requests . . . In the initial version of AutoDev, Slack will be In the initial version of AutoDev, Slack will be In the initial version of AutoDev, Slack will be used as the primary tool. used as the primary tool. And I will also talk about the ROI insights mentioned earlier . It is important to ensure that we are not paying . It is important to ensure that we are not paying excessive costs relative to the amount of work we perform excessive costs relative to the amount of work we perform , and to adjust the workload to achieve the desired results adjust the workload to achieve the desired results . I will . I will help you adjust the workload to your liking help you adjust the workload to your liking .

. . great. So, the point I would great. So, the point I would great. So, the point I would like to briefly touch upon once again is the like to briefly touch upon once again is the like to briefly touch upon once again is the planning stage planning stage planning stage . We plan to . We plan to . We plan to launch a new product called AI Planner launch a new product called AI Planner launch a new product called AI Planner . This product is . This product is useful for spike projects, project managers, or administrators who want to check if a specific feature can be implemented in the current repository useful for spike projects, project managers, or administrators who want to check if a specific feature can be implemented in the current repository . When you . When you . When you enter an idea you want to build, the AI enter an idea you want to build, the AI enter an idea you want to build, the AI planner analyzes the planner analyzes the planner analyzes the contents, contents, contents, plans, and feasibility in the repository through our contact engine plans, and feasibility in the repository through our contact engine plans, and feasibility in the repository through our contact engine . If implementation is possible, we will . If implementation is possible, we will . If implementation is possible, we will inform you of the result and inform you of the result and provide the details required for implementation. Well, if you want to implement a specific feature or idea, this tool will create a plan and create a plan and create a plan and suggest a few ways to execute it suggest a few ways to execute it suggest a few ways to execute it . Well, we . Well, we . Well, we 've talked a lot about contact engines. 've talked a lot about contact engines. 've talked a lot about contact engines. Shane is here. Shane is here. Shane is here. He is putting a lot of effort into developing Code Chat He is putting a lot of effort into developing Code Chat He is putting a lot of effort into developing Code Chat .

. . Shane, could you explain the Code Chat feature and Shane, could you explain the Code Chat feature and Shane, could you explain the Code Chat feature and all the features included in the Contact Engine itself in all the features included in the Contact Engine itself in all the features included in the Contact Engine itself in a little more detail a little more detail a little more detail ? yes ? yes ? yes . Do you have a slide? . Do you have a slide? . Do you have a slide? Yes, there are no slides. Yes, there are no slides. Yes, there are no slides. All right. Well, Shane All right. Well, Shane talked about the Contact Engine code chat. We talked about the Contact Engine code chat. We at Atlasy have numerous at Atlasy have numerous at Atlasy have numerous repositories repositories repositories . I understand that there are about 15,000 repositories throughout the company . I understand that there are about 15,000 repositories throughout the company . . So, ' So, ' So, ' Where should this change be implemented?', ' Where should this change be implemented?', ' Where are X, Y, and Z implemented Where are X, Y, and Z implemented Where are X, Y, and Z implemented ?', 'Which service is ?', 'Which service is ?', 'Which service is using the other service using the other service using the other service ?' Even simple questions like that are ?' Even simple questions like that are ?' Even simple questions like that are very complex. very complex. So, the core of a context engine is to narrow the search range narrow the search range to find the location within the code most likely to contain the answer, and then to find the location within the code most likely to contain the answer, and then dig into the code from there to dig into the code from there to dig into the code from there to find the answer to the question find the answer to the question find the answer to the question . And then add . And then add . And then add all the information obtained from the teamwork graph on top of that all the information obtained from the teamwork graph on top of that all the information obtained from the teamwork graph on top of that . Documents are . Documents are often old and inaccurate. If you search using only documents, you will obtain only information that was true at a specific point in time obtain only information that was true at a specific point in time . However, the document . However, the document . However, the document also contains a lot of very valuable information also contains a lot of very valuable information also contains a lot of very valuable information . For example, this . For example, this . For example, this project implemented these features at a specific point in time project implemented these features at a specific point in time . It was true at the time, but it may . It was true at the time, but it may no longer be true now no longer be true now no longer be true now . However, the . However, the . However, the reason for making such a decision at the time is reason for making such a decision at the time is very useful for applying future changes. Therefore, by utilizing the currently existing code and decisions reflected in the code at a specific point in the past, you can

by utilizing the currently existing code and decisions reflected in the code at a specific point in the past, you can plan the implementation path or identify the reasons why the plan the implementation path or identify the reasons why the plan the implementation path or identify the reasons why the code needs to be changed in a specific way code needs to be changed in a specific way . Do you understand? Do . Do you understand? Do you have any questions? Yes, that is you have any questions? Yes, that is you have any questions? Yes, that is correct. correct. correct. Historical information regarding the reasons behind decisions is Historical information regarding the reasons behind decisions is often missing from the code generated by agents, but this historical this historical context is very important to avoid repeating past mistakes. context is very important to avoid repeating past mistakes. context is very important to avoid repeating past mistakes. So, we think of the context engine So, we think of the context engine from that perspective from that perspective from that perspective . That's . That's . That's cool. thank you. I have a cool. thank you. I have a cool. thank you. I have a question. question. Yes, it is widely used in AI planners. I am not sure which of the two of you can explain it in more detail, but it is being used in AI planners. For example used in AI planners. For example , , , when considering how to build something, you when considering how to build something, you can look at the documentation and surrounding context of related projects. can look at the documentation and surrounding context of related projects. So, the context engine So, the context engine is the core of many new features we are developing. It is also is the core of many new features we are developing. It is also used in AutoDev. The used in AutoDev. The used in AutoDev. The AutoDev representative will probably be AutoDev representative will probably be AutoDev representative will probably be able to explain it in more detail able to explain it in more detail able to explain it in more detail . AI Reviewer does not currently . AI Reviewer does not currently . AI Reviewer does not currently use a context engine much use a context engine much use a context engine much , but is , but is experimenting with ways to integrate more information experimenting with ways to integrate more information experimenting with ways to integrate more information . And . And . And another feature revealed today another feature revealed today another feature revealed today can be found in Rover Chat.

can be found in Rover Chat. can be found in Rover Chat. When speaking with Rover, we are currently offering an When speaking with Rover, we are currently offering an When speaking with Rover, we are currently offering an early access early access early access version to several clients. It was version to several clients. It was version to several clients. It was announced a few months ago. announced a few months ago. announced a few months ago. When you ask code-related questions, When you ask code-related questions, answers are primarily provided through Confluence search and other connected tools answers are primarily provided through Confluence search and other connected tools . If you use GitHub or Bitbucket, . If you use GitHub or Bitbucket, use their respective search APIs. use their respective search APIs. However, However, However, by enabling a new feature called Code Context, you can by enabling a new feature called Code Context, you can understand the code more comprehensively based on the current state of the code and past decisions, and answer code-related questions more effectively answer code-related questions more effectively . . Is there a way to use this feature in related public services, for example, through API access or elsewhere? Is there a way to use this feature in related public services, for example, through API access or elsewhere? Yes, Tumour Graph CLI is making these Yes, Tumour Graph CLI is making these Yes, Tumour Graph CLI is making these features available in earnest features available in earnest features available in earnest . Well, I think it would be more appropriate to think of the context . Well, I think it would be more appropriate to think of the context . Well, I think it would be more appropriate to think of the context engine itself as agent-based engine itself as agent-based engine itself as agent-based search search search . We . We have many core basic elements that we use, and we plan to expose these elements to the agent via the CLI plan to expose these elements to the agent via the CLI . However, over time, I am also thinking of . However, over time, I am also thinking of considering integrating more of these elements into the CLI. Yes, considering integrating more of these elements into the CLI. Yes, Tumour Graph CLI Tumour Graph CLI Tumour Graph CLI can access all code context skills can access all code context skills can access all code context skills .

. . thank you. yes thank you. yes thank you. yes . I . I . I have one more question. The first is have one more question. The first is have one more question. The first is about standards. It about standards. It seems like the standard format is similar to a Markdown line, is that correct? seems like the standard format is similar to a Markdown line, is that correct? Yes, that is correct. Yes, that is correct. Is there a way to map these standards in a definitive manner, such as through CI testing? Is there a way to map these standards in a definitive manner, such as through CI testing? For example, a For example, a scenario like 'Checked', similar to a checkbox scenario like 'Checked', similar to a checkbox scenario like 'Checked', similar to a checkbox . I think . I think that feature should be implemented within Markdown itself, and it seems like it would be an interesting use case where specific conditions can be presented. For example, ' Does this satisfy the condition?' Does this satisfy the condition?' Does this satisfy the condition?' Or 'Does this not meet the conditions Or 'Does this not meet the conditions Or 'Does this not meet the conditions ?' Like that ?' Like that ?' Like that . Well, I . Well, I . Well, I think it is impossible at the moment. think it is impossible at the moment. It seems we won't be able to provide this by default, but we will pass it on It seems we won't be able to provide this by default, but we will pass it on to the standards team so that it can be implemented in the future to the standards team so that it can be implemented in the future to the standards team so that it can be implemented in the future . . . Yes, I Yes, I think it would be really good to do that so that we can organize it more systematically. That's think it would be really good to do that so that we can organize it more systematically. That's cool. cool. Then, is there a way to break down a large epic into multiple components and run them separately, like agent waves run them separately, like agent waves ? Usually, it ? Usually, it is a method that allows subsequent work to proceed through foundational work.

is a method that allows subsequent work to proceed through foundational work. For example, a For example, a For example, a method of moving on to the next step after completing the basic work method of moving on to the next step after completing the basic work method of moving on to the next step after completing the basic work . Yes, to reiterate . Yes, to reiterate . Yes, to reiterate , , , I think the approach of breaking down epics into smaller I think the approach of breaking down epics into smaller I think the approach of breaking down epics into smaller issues and executing issues and executing each small part within each issue via agent waves seems similar. Is that right? seems similar. Is that right? After implementing the first step, can we go a step After implementing the first step, can we go a step After implementing the first step, can we go a step further and proceed to the second step further and proceed to the second step further and proceed to the second step ? yes ? yes ? yes . great. . great. . great. Yes, yes, that is a very good question Yes, yes, that is a very good question Yes, yes, that is a very good question . Actually, I . Actually, I . Actually, I looked into this part a bit yesterday. If you use JCA, you looked into this part a bit yesterday. If you use JCA, you looked into this part a bit yesterday. If you use JCA, you take an epic and ask, "Can you take an epic and ask, "Can you take an epic and ask, "Can you break this epic down into smaller issues break this epic down into smaller issues break this epic down into smaller issues ?" You ?" You ?" You can request this. Then, can request this. Then, JCA JCA can break down epics into smaller Jira tasks can break down epics into smaller Jira tasks can break down epics into smaller Jira tasks . After that, you . After that, you . After that, you can set up several automation rules can set up several automation rules can set up several automation rules . For example, . For example, when a Jira epic is disassembled, you can automatically perform tasks such as restarting the when a Jira epic is disassembled, you can automatically perform tasks such as restarting the JCA instance JCA instance JCA instance or starting a JCA session through automation rules or starting a JCA session through automation rules .

. And I think there will be no restrictions on further expanding this function And I think there will be no restrictions on further expanding this function . . Right? For example, I Right? For example, I divide all the stories I want into multiple epics. It involves going through foundational work, that is, stages like V0, V1, and V2, to further subdivide the tickets. I think there won't be It involves going through foundational work, that is, stages like V0, V1, and V2, to further subdivide the tickets. I think there won't be any problem with doing that any problem with doing that any problem with doing that . You may continue . You may continue . You may continue speaking. We are speaking. We are speaking. We are quite flexible. You may quite flexible. You may quite flexible. You may continue. continue. continue. Something like JCA automation seems more suitable Something like JCA automation seems more suitable Something like JCA automation seems more suitable . . . I am not sure if the robot can do it directly, but I I am not sure if the robot can do it directly, but I think it is quite possible if automation is used. think it is quite possible if automation is used. Yes, it seems like Auto Dev is possible to some extent as well Yes, it seems like Auto Dev is possible to some extent as well Yes, it seems like Auto Dev is possible to some extent as well . . Yes, it's fine. Let me Yes, it's fine. Let me ask you a question. I'll check again ask you a question. I'll check again ask you a question. I'll check again . For example, let's . For example, let's . For example, let's assume you are changing code in an IDE assume you are changing code in an IDE assume you are changing code in an IDE . yes. You must consider that JCA . yes. You must consider that JCA . yes. You must consider that JCA operates very independently of agent sessions operates very independently of agent sessions . Therefore, changes made within the agent session are reflected in real time. When you make changes in a VS Code instance in the cloud,

all changes are reflected in real-time and reflected in real-time and reflected in real-time and integrated into the cloud session itself integrated into the cloud session itself integrated into the cloud session itself . Changes made in the IDE are treated as if they were . Changes made in the IDE are treated as if they were processed in a separate instance processed in a separate instance . . . If you use an agent, you can think of it as all operations If you use an agent, you can think of it as all operations If you use an agent, you can think of it as all operations taking place within the cloud taking place within the cloud taking place within the cloud . All . All . All changes made in the agent session within Jira are changes made in the agent session within Jira are reflected in real time. Now, let me reflected in real time. Now, let me ask you one more time. It seems like you intend to run ask you one more time. It seems like you intend to run ask you one more time. It seems like you intend to run multiple agents in an on-premises environment. Is that multiple agents in an on-premises environment. Is that correct? Yes, I correct? Yes, I use Docker. great. I use Docker. great. I use Docker. great. I have one more question. That have one more question. That is a good question. We started supporting Skills a while ago. Yes, that is correct.

started supporting Skills a while ago. Yes, that is correct. Yes, regarding the AI ​​viewer itself, Yes, regarding the AI ​​viewer itself, Yes, regarding the AI ​​viewer itself, well, we provide support skills well, we provide support skills well, we provide support skills . It was released in less than a month . It was released in less than a month . It was released in less than a month . . . great. That's great. That's cool. Yes, thank you all cool. Yes, thank you all . I guess I should go now. . I guess I should go now. . I guess I should go now. If you have any other questions, please feel free to If you have any other questions, please feel free to If you have any other questions, please feel free to ask at any time. ask at any time. ask at any time. Yes, thank you very much. Thank you to Yes, thank you very much. Thank you to everyone who participated in the workshop everyone who participated in the workshop everyone who participated in the workshop . I would also like to sincerely thank the technical staff for helping me resolve the Wi-Fi . I would also like to sincerely thank the technical staff for helping me resolve the Wi-Fi . I would also like to sincerely thank the technical staff for helping me resolve the Wi-Fi issue issue . We give you a big round of applause . . I hope you have a great rest of the week during the conference. I hope you have a great rest of the week during the conference. thank you.