✨ Sherif-sclusive: Atlassian Intelligence Updates (June)

[#1]

👋🏼 G'day Atlassian Community! We’ve been pedal to the metal during the last month and I’m glad to be back sharing the scoop with you. If you missed my last post, I’m Sherif! Head of AI at Atlassian. (Maybe I should combine those two... AtlassAIn? AI-tlassian? I’ll workshop that a bit!)

I’m popping on to highlight the innovations, features, and improvements we made in June and I’ve linked some waitlists for new releases. 

1️⃣ 🔔 Alert grouping & Intelligent PIR (Post-Incident Reviews) Beta

Detecting and resolving incidents is challenging to say the least.

1. Know more, faster: AI Alert Grouping automatically identifies patterns among incoming alerts to correlate and group them based on similarities. Responders can focus on the highest-priority alerts and proactively detect incidents

2. Type less with intelligent PIRs: Creating a Post Incident Review (PIR) takes time. No more scrolling through issues and loads of text to write an incident overview. Atlassian Intelligence provides AI-generated summaries of an incident saving precious time for the ops team, without losing critical insights.

To discover upcoming AIOps features, sign up for the waitlist and stay informed about the latest developments.

2️⃣ 🗒️ AI Summaries for all your Jira Issues (beta)

We started with AI summaries in Jira Service Management, and you told us you want this in all of Jira… 🥁 We’re happy to share that AI summaries are coming to all Jira issues in Premium and Enterprise! Whether you’re doing the work of three people or you're coming back from your summer holiday to a full inbox, Jira issue summaries can help you get up to speed - fast. Summarise that long conversation about a feature you’re building in Jira software or catch up on a conversation about a business project task within seconds.

3️⃣ 📹 Loom AI features to speed up your video editing

Easily record and share AI-powered video messages with your teammates and customers to supercharge productivity with these two new features:

Live Rewind: Kids walked into the room? or your dog started barking while you’re recording? 🤦🏻‍♂️ No need to start again, with Live Rewind you can pause, trim, and resume right where you left off recording. Available on our Google Chrome Extension.

Edit Video by Transcript: “Umm..” “errr…” “like..” “actually…” all those filler words: you didn’t think you said them, but when you play the video back there are a lot more than you thought! Don’t worry! You can now easily remove a specific word or sentence by highlighting it in the transcript and deleting it. Your video will update immediately.

Yes, AI feels a lot more like ✨ when it comes to video!

4️⃣ 📖 Rovo Definitions in Confluence and Jira (BETA)

If you missed the announcement at Team 24, Rovo is Atlassian’s new knowledge discovery offering, an add-on to your Cloud products. We couldn't wait for you all to play with it before we get the Rovo Beta program formally started, so you we’ve included it in your premium or enterprise editions of Confluence and Jira (if you have AI enabled) 🎁 !

Most of you already would have already had it in your Confluence, but today we’re sharing we’ve rolled it out to Jira as well!

Come across a company or team-specific term you don’t understand? Don’t worry, no need to ask someone. Rovo helps you learn in the flow of work by automatically generating a tailored company-specific definition based on your organisations knowledge from Confluence and soon other products.

If you’re interested to learn more about Rovo checkout the website and join the waitlist.

5️⃣ 📄 Content to help your team make the most of AI

We continue to invest in our products as well as our practices. This month we’ve got two new articles to help your team make the most of AI:

1. Writing content in Jira, Confluence, Bitbucket or Trello? Learn how to use Atlassian Intelligence to maximise your productivity. @Ethan Yew, Product Manager for the Atlassian Editor shares how your team can speed up their editing with tips, tricks and suggested prompts.

2. If you’re a team building software with AI one of the most undervalued things is the power of a prompt. Knowing how to use prompts can be the difference between a small improvement and a game-changing result. Julian Connor, Product Manager for Jira AI shares how software teams can use prompts for prototype testing, well before they start any development. Recommend read if you want to learn how to use prompts to speed up your product development lifecycle before writing any code.  

Questions? Comments? Wild applause? Constructive feedback? I’d love to hear it all! We’re using these features ourselves but I’m always looking for ways to understand how you work or how these developments are impacting your workflow. 

I’ll be back with more July updates!

[Dr Valeri Colon (Connect Centric)]

@Sherif Mansour I'm a fan of your head bobbing around in the Loom circle. As I work to get the latest AI features added to the creator's JSM demo site, I'll be adding these videos and gifs to our monthly Atlassian Live Learning AI course. Please keep them coming. The community loves getting an inside look into your work.

[Caity Belta]

@Dr Valeri Colon (Connect Centric) Love that idea! You're crushing it with the Community AI course via Live Learning! Keep up the awesome work :)

[Tapiwa Samkange]

@Sherif Mansour many thanks for this useful update and the personable manner in which you delivered it! Loving the practical demonstration of how Loom can help make updates even more accessible. Looking forward to the next #Sherif-sclusive! 

[Dr Valeri Colon (Connect Centric)]

@Caity Belta I appreciate your support and enthusiasm! We're excited about the next iteration of the course. Instead of role-specific breakout rooms, we're switching to rooms by application to dive deeper into specific tools. 

[Taiwo Ojo - Alluvium]

Excited about Rovo!

[Filip Callewaert]

Panels do not render in excerpt fields?

I tried to include the panels in an excerpt macro but that doesn't work. So that content cannot be shown in lists and reports yet that make use of the excerpt field. Idem with the excerpt-field in Databases. 

Really looking forward to have these AI panels as field types in Confluence Databases - imagine the column "summary" and "action items"!

[Dr Valeri Colon (Connect Centric)]

@Taiwo Ojo - Alluvium a featured guest at the Live Learning AI course on the 18th! We have such cool guests. I've been experimenting with ways to pull our collective knowledge at these free events and form a community of practice around AI. 

@Filip Callewaert Confluence databases came out of beta around June 2024 and the AI features are still being rolled out. You might consider using the MultiExcerpt macro from the Atlassian Marketplace as a workaround.

There is a community discussion about integrating AI panels into Confluence Databases to enhance functionalities like summaries and action items. Stay tuned for updates and please let the community know if you experience any changes in functionality.

[Filip Callewaert]

@Dr Valeri Colon (Connect Centric) , the primary focus of the issue I mention is as such not about databases, but about the excerpt macro in Confluence (and yes, excerpts are also used in Databases, but that is only a second degree focus here). 

At this moment, it is not possible to include an AI Summary panel in an excerpt macro, neither is that possible in a multiple-excerpt-macro you refer to. As such your suggestion is not a workaround for the problem I presented. 

Loom-here-what-I-mean

[Dr Valeri Colon (Connect Centric)]

@Filip Callewaert I love that you just Loomed an explanation! Great way to support the clarification of my understanding. Looks like those features are under development. I'll be facilitating the Confluence breakout room during the session on 18th, you're welcome to demonstrate how you've been using new features.

[Aaron Geister]

@Sherif Mansour  Very nice updates great job for you and the team. Thank you for bringing new features and show casing them to us.

[Haddon Fisher]

Hey @Sherif Mansour - loving this energy and the transparency here! One question for you:

AI technology holds a lot of potential, but beyond any other considerations, just may not be a good fit in all use-cases, even down the road. For example, in our testing of prompts we expect our customer-facing departments would use regularly, the answers were...shall we say dangerously inaccurate...and for the time being we've had to disable all of the AI functionality, or I guess disable it as much as we can. I expect that accuracy will improve over time, but I don't see anyone expecting it to be even 80% any time soon.

However everything I am reading indicates that Atlassian is planning on baking this functionality deeply into your applications. I am sorry to be blunt here, but y'all also have a bit of a history of pushing functionality that your customers are a lot less in love with than Atlassian seems to be.

Is it currently your plan to provide tooling to disable this for organizations who are either uncomfortable, unable, unwilling, or simply do not need to adopt this technology? If not, what is your answer to organizations in this position?

[Filip Callewaert]

Is my interpretation correct that AI does not "read" content in Confluence Databases? 🧐

[Sherif Mansour]

Sorry for the late reply folks was out on leave.

@Haddon Fisher Do you have a specific feature you are referring to where the results were inaccurate far more than you would have expected? That would help me work out which area you're talking about. 

But addressing your general comment on AI quality - yes, I think it's well known that these large language models can and do hallucinate. Each product/feature area have their own response quality guardrail metrics and it differs depending on the feature and the use case it's designed for. In some scenarios, you do want AI to improvise and come up with creative suggestions, but in others you want as precise answer as possible. I can share that each team sets response quality targets and guardrails as well as works towards maintaining those or improving them, but it's different depending on the use case, so any further context would help. 

Regarding: "Is it currently your plan to provide tooling to disable this for organizations who are either uncomfortable, unable, unwilling, or simply do not need to adopt this technology?"

Absolutely, Admins today can enable/disable AI at the product+site level, in some scenarios at the feature level as well although we generally try to minimise feature-level toggles and get to the root of why they want to disable it to see if we can address those matters first. Hope that helps! 

@Filip Callewaert - thank you for the Loom with the context, yes it's not possible to nest the AI panels inside the except macros. I'll pass on that feedback to the team, your use case makes sense to me.

And yes, confirming that AI cannot read the contents of Databases (and Whiteboards) at this stage, and that you can't use these smart panels in a database, but this is something that would be great for us to get to. I've passed on your feedback to the teams working in those areas. Thank you for taking the time! 

[Filip Callewaert]

Great to see the AI Chatbot in JSM responding in Dutch since a short time!! Question asked in Dutch, answer provided in Dutch! Until recently: only answers in English.
Gone are the days of manual translations of documentation (I hope)!

[Haddon Fisher]

Hey Sherif Mansour, thanks for the follow-up!

• We did a lot of unscientific poking-around-testing, but there were two cases we saw a lot of value in and ran some more controlled experiments. In both cases, we hit the same trust-related catch-22:
  • Creating Automation rules
    • Context: Its not feasible for admins to own all of the Automation rules in larger organizations, but there's plenty of cases when an admin has to get involved anyway. Putting aside things like global rules which would require them, even basic business logic can require a pretty complex rule to automate and while it would be great if every Jira user was Automation-capable, it's not realistic.
    • Findings: When prompted for something beyond very basic, the rules we got back often required pretty deep troubleshooting to make work. In one memorable case, it took us a few hours to figure out that the AI had invented a bunch of A4C smart values, but it also regularly created architecture that just would not work.
  • Summarizing long Confluence documents
    • Context: My current organization is a B2B employer services provider, and we have a lot of knowledge generated by the back-of-house (Engineering, Product, BizOps) that could be valuable to the front-of-house (Sales, CS) but is also often long, arcane, and buried.
    • Findings: It made a non-trivial amount of stuff up. For example, our products involve a lot of regulatory-required regionalization, and so we have countries where we can and cannot offer them; the AI was unable to answer this correctly.

In both cases the value proposition sort of hinges on the one thing we all agree we can't have: a high degree of accuracy. If someone doesn't have the context to create an Automation rule or understand a highly technical document, they intrinsically don't have the context to tell when the AI is and is not correct. Kinda sinks the whole boat.

You're correct that it's now possible to turn off these prompts for non-admins but is that going to remain an option? We've been getting a very hard sell on this stuff and anyone whose a site admin still has to put up with all of the ads.

One other sort of off-the-cuff thought from what you said about feature-level disabling: I think the ability to manage this feature-by-feature in addition to product and instance might make admins a bit more willing to try this stuff out. For example, the Confluence content generation functionality has a significantly lower risk potential than some of these other situations; I could see us leaving that on but disabling other things if that option was easy.

[Nikki Zavadska _Appfire_]

Post Incident review report generation is so useful!