GitLab UX Team, 26 Apr 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: UX AI Integrations sync up 2023 04 26

Description

No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).

A

Foreign welcome to the AI ux sync today is April 26th um and Sarah I see that you're typing, but you have the first item. If you want to go.

B

uh Yes, okay, um I'm gonna, link to this in a second, we had an engineer on the source code team create a really cool demo of continuous chat, which is eventually where we want to get to with the interactive common AI chat component.

B

um There was one concern, though, is that engineering is moving faster than ux, and so what we ended up with this morning is two different chat experiences which we have reconciled and we're going to wait to release it and align them. But I just wanted to share that with this group, so we can keep an eye out for it.

B

um It came out of it was. It was surprising and I think that's what happens when you put very strong engineers and tell them to go build stuff with no guard rails, so that was the just the one coin I wanted to make there I also wanted to ask um Michael is going to be working on designs for this, but he does not overlap with us a ton, Justin and Katie. Are you guys able to talk to him at any point? You have any overlap with him.

C

Well, Katie does normally um okay well, I mean I, guess I do to some degree too. Normally this there's two calls. Oh.

B

Yeah doing the other one yeah.

C

Okay, so that that later one is a good opportunity to count direct access to Michael.

B

Okay cool, um so this is a long-winded way of saying. I would love your support in making sure that all designers are angling towards the same experience and helping Engineers on their team? Do the same? I know that verify is working on a conversational or had been working on a conversational chat experience and now they know that we are building the common element so.

B

This was a public service. Mountain slash ask for help, slash just making everyone aware.

A

Well and sorry, it's just hard to keep up with all of the things, but the the one that you're building this kind of continuous check. Can you describe the use case? I know that you're building the unified thing, but are you building for a specific use case or how will that appear in the product? Yes,.

B

Great question: so um the source code team is building the explain this code block they added. They continue to iterate on their MVC because they are moving very quickly and the demo which I'm having a hard time finding, which will explain this to you, um allows you to ask follow-on questions so explain this code block. What does this function mean? Tell me about this variable.

B

That is the use case. I would like the continuous chat to be a function. We can apply to other use cases that should be a common solution, I think for things that need it. But if you want, if it's appropriate or natural for a user, to ask follow-up questions um set.

A

Yeah that totally answers the question so I believe Michael was working on that and the context in which you trigger this is when you're in the Fall view. Viewing a code follows that right.

B

um That's another great question, so I have put the slack spread there, so the MVC was the MDC was the interaction where you're looking at a file and then you've got the question mark. That appears on this side that you click and then in context. It explains it for you. They turned it into a chat they iterated on it.

B

So it became a chat experience. I asked them to not ship that to production today, so it's not going to reduction so that we can unify the user experience between the get lab chat. Where you can ask questions about our documentation and the chat where you can ask questions about a block of code should be the same experience. We should not have two experiences, um so they are going to wait until later this week, maybe early next week, so that Dennis can um make the uis look the same.

A

Cool. Thank you. No.

B

C

Yeah, this is in line with the framework that Pedro is putting together as well. There are sort of three different levels of interaction: okay and the one that the question mark is is in is a is a lower level of more context, oriented and then once they need more depth, then you would go to the higher level, which is chat or discussion or whatever. We want to call it. Yeah.

B

C

um So there was just a disconnect in that I think I mean like like we know, everything's going so fast.

B

uh Okay, you just brought up a concern for me. Justin though Pedro is working on that.

C

He's working on the mental model, not anything design like a special design. Okay,.

B

I, just when I hear things like that, I want to clarify, because I've got Michael working on what the chat on her face should look like.

C

Luckily, is the designer and and I think, all of us all ux designers who are working on a stuff are aware of this concept. So what you're? It's all good. It's all good! Okay,.

B

um Okay, that's all I had thank you.

A

Foreign I just had a kind of two quick fyi's uh open AI terms require us to always disclose from the UI that the customer is interacting with an AI system, so I'm working on an MR uh which hopefully will get merged today around the exact language and copy guidance that we should use in the UI.

D

C

Last I saw is, we were making a little more generic, so we didn't have to specifically say open AI right.

A

Okay, yeah, that's right.

A

um um And my next point is also just a quick FYI that we will release code suggestions, which is kind of our answer to GitHub co-pilot into the web IDE as a beta feature on the 22nd of May I, just linked the designs in the Epic in case. Anyone wants to see this I'm just posting this here. For those who may not have visibility of this feature to have a look.

C

Do you want to walk through you got.

D

A

Oh yeah sure um yeah. Let me just maybe open up the Epic share. My screen.

A

So basically, uh what this feature is is when you're inside the web IDE. As an engineer and your writing code, uh we use a model that we created to be able to recommend or suggest uh like an autocomplete Style lines of code.

A

um We have an open beta for visual studio, desktop out at the moment and we're fully focused on the web. Ide and the reason is um gitlab has a company. The Strategic direction is to really push for remote development, so we're putting a lot of investment into the web IDE um but yeah. Basically, it's pretty cool I think the best uh one that I can show is um different kind of ways that uh it can complete code for you um so single lines. It can also complete multiple lines like an entire function.

A

um It can complete in the middle of the line, um and it can also uh complete, if you just give it some instructions in plain language. It can actually generate code for you and I think for the open, Beta we're supporting like six seven, eight nine languages with a high level of confidence, but uh with the intention to expand to more- and there is also an intention in the future to expand this functionality into other Ides, such as like IntelliJ Etc, but yeah. That's the that's the gist of it. um I'll stop sharing.

C

That's very cool: do you know if the uh source code team who's working on the explainers code? Are they working on the web or the standalone.

A

The explain this code that um Sarah was just mentioning or.

D

A

um Isn't that in the gitlab product itself, not in an ID or sorry I, don't want to send your question. Maybe I.

C

Don't know yeah exactly I, guess I'm, not sure where it is I've.

A

C

uh At the things, I haven't actually explored it. The.

A

Mvc is when you're in the file view in gitlab itself as you're like uh selecting lines of code. You get this little question mark iclone on the side that you can click on. So it's in the product itself um and as Sarah was mentioning like uh you know. Maybe we're gonna expand that to other other parts of the gitlab product.

C

I guess I always thought that was part of some IDE and not just a file reader.

A

Yeah right but I, think the um you know, GitHub co-pilot actually does offer explain this code as part of your IDE. It's a feature that's currently in beta, um but I think it makes sense to also extend that functionality into the IDE. Eventually.

C

Yeah, when I was talking to moan at the beginning of the week, that's that's what I pinged you on asking. If you had designs for that, because she thought you did for some reason: um yeah, it's I think like I shouldn't live there too. We've.

A

Got a lot there's a lot of things that competitive products do like that can generate entire missing test files. They can you know this. um Some competitors have a bit more of that conversational. Like tell me what's wrong with my code and then it's like. Oh, there might be a security vulnerability, so you know the stuff that Becca's working on the stuff that Michael's working on I think all of that functionality could be integrated also into the IDE eventually.

A

But I would assume that my team is going to focus on improving the models so that we can cover more languages with a high degree of accuracy before we would kind of go into those additional features.

C

Yeah that makes sense, but uh I have a feeling we might have to coordinate with those other teams to get them to integrate. Somehow.

A

D

C

Yeah, that's cool though, and that's the only app that's that's in beta that that is using our own. Yes,.

A

uh Suggested reviewers as well, which is now ga, oh okay, yeah, but the yeah. Otherwise, uh oh yeah, we're not using third parties. It was I, did have a customer call yesterday and I asked him about how he would feel about using third parties and he's like I just want the best feature: I, don't care! How he's like I, don't expect gitlab to build these models.

A

You know stand on the shoulders of giants kind of thing, so that was interesting, but he he was working at a startup, so it might be a different perspective from an Enterprise customer.

C

Yeah, especially once all these legal fights and stuff the people, I guess, depending what the result of those legal fights are and who owns the code and.

D

C

Totally different story, when it's a huge company.

C

Okay, cool stuff.

C

um Did you oh your second point to read only yeah.

E

Oh yeah I could just share what I'm working on quickly uh still just just for awareness. Let me just share my screen here. That'd be easier.

D

E

Right, uh yes, so originally I took on this issue here, or this is the Epic. So this is another example where engineering is kind of moving faster than ux here, and uh a lot of this has already been done, and there was already um uh uh it was kind of like merged already or just kind of it's still under feature flag. But basically the idea is to summarize issue uh discussions and uh in this UI or in the first iteration of this. What happens? Let me just uh I, don't know if you've all seen this.

E

Are you all familiar with this? Have you seen this Katie or Justin.

A

um Assume no okay.

D

So this is I'm just gonna.

E

Show this so basically, you have a an issue and there's a lot of activity in the issue and uh we want to have the ability to summarize quickly. You know larger issues, what the issue is about or summarize the activity of.

E

What's what the discussion's about so uh in this first iteration, let's see here, it was at there was a there, was a little new icon added into the toolbar uh or into the new like if you're, creating a new uh discussion or thread here, there's a new little tandooki icon here to New Key icon that lets. You uh summarize the comments and once you click that it just kind of puts a summary in line. Is it as like another item within the activity uh thread?

E

uh So after thinking about this, it doesn't seem like I'm thinking about the pro that we're trying to solve. It doesn't seem like we want to just have another summary inside of the um yeah see. This is the the summary right here. So it gets summarized Within your input window and then you have the option to like copy and paste it somewhere or just again submit it as a comment. So I thought that interaction wasn't kind of what we wanted or what it's not really solving the problem of summarizing.

E

It doesn't seem like you want to like post that as a comment, your summary of uh of the activity thread so uh yeah long story, short uh after lots of discussion uh I think it was a game that created and we we were talking kind of back and forth about this and then gave it created a new issue here that is, uh to move the summarize summarization workflow out of the activity feed.

E

So this is the newer uh design here and uh just quick note. I just got feedback from Michael on this I haven't uh kind of looked through this yet uh because I know he is kind of working on the kind of unified experience of the drawer.

E

So the idea here is you'd have some kind of button here around the activity once you click that it would pull out the drawer to give you a summary of what's uh what's happening in the thread, and the idea here is, uh you know, you'd maybe want to compare whether the summary is correct. Still with the actual thread. You know you might want to look at both, so you could still kind of scroll through the thread and see the summary uh and also I think there's some issue.

E

That I saw that I came across that any kind of text, Heavy uh AI experiences, would should probably leverage the drawer just for consistency's sake.

E

um So uh I just have to look through this, but just wanted to make you it's more of making you both aware of kind of what what I've been working on here and I'm, going to work with Michael, Lee and uh Alex on kind of unifying this experience, making sure that it's it's a similar experience when the drawer comes out, so the drawer doesn't always look different depending on you know where you are so uh you know thinking about this more it might it might be that you know we always have the drawer.

E

You know contextually kind of uh have options in there, depending on what you could do on that page. For example, on the merge request page, you might want different options inside this drawer versus you know. If you're on the uh issue or if you're at like on a planning board, maybe or a or if you're, on, like to optimize or value stream analytics, you might have like different options that AI can assist you with uh or some um some options there.

E

uh So yeah, it's just kind of thinking a little bit more ahead or about how this tour can maybe be leveraged but uh uh yeah. That's all I have for this uh for now and then uh I'll have more updates later, hopefully by the end of today of how this will change. But uh let's see here so again, there's also an MR for this. But uh again the Mr does this here, where it summarizes the activity here. So this is.

E

This is going to be released at this point and then there's going to be another follow-up issue, then to get in the drawer. So it seems like uh again, like uh uh Engineers are just working faster than uh that.

E

I could keep up even with uh with the ux on some of this stuff, so so I've been having to kind of uh put the brakes on things, but I also don't want to stifle uh you know or momentum here and I want to make sure that you know we're actually testing even what this is kind of coming up with. So uh so that's why we decided to just like let's, uh let's get this merged, let's get this out again. This is still under feature flag, um but yeah.

E

So that's kind of the the balance that I've been kind of having to uh to do here said enough. Any thoughts on that it's like with uh not releasing something versus kind of trying to release and get feedback on it.

C

I think you're taking the right approach, we're in an experimental phase, it's okay for engineering to be working this fast. As long as it still makes experimental phase. The guidance is that uh in general, ux isn't really expected to do much while in this experimental phase, okay.

C

But if you, if, if your team has the time to have a have that group's designer support, it I, don't see a problem with it, especially since this use case seems fairly relevant. A lot of people are like it's one that popped into my mind when I was thinking about this a few weeks ago and I think everyone's kind of had these names scenario. So it seems like a problem that could use solving so I, don't think it'll go away. In other words,.

E

A

Since this feature is.

E

A

In production in some in some way, I would be really interested if you could run some user testing with people's actual their own git lab and actually have them interact with it and give commentary on the um usefulness of the content that it generates um to understand. Is the content actually correct and useful?

A

Do they still need to go through and read all the comments to double check it uh and, like you know, what are they looking for when they get a summary I think that's the benefit of having something out in the world is we can actually test it with customers, real data.

D

E

Yeah we do have a there is a feedback issue uh on this feature right now. There is some useful feedback. That's come through already. uh I did, as I was playing around with it in that Mr. uh It's kind of interesting, because I tried creating just like a quick, uh really simple issue about you know. Should we make a mobile, app and I kind of made up some fake names about put a discussion there?

E

It was only like, maybe like five threads I put in there, and uh it was interesting because every time you kind of regenerated the summary it would slightly change it wouldn't always be the summary. Even though none of the comments changed so I thought that was a little interesting. Yes, I'd be curious to see. Also, you know what this looks like or what it comes back with with larger uh issues or issues with a lot more activity.

E

There's uh currently uh a limitation there with you know the token size of how much you could uh you could, how much data you could send at once so I there's another issue: that's being worked on right now at the moment to be able to send like a larger chunk of or Break It break it up somehow and send a send larger chunks back to uh uh open AI at the moment, yeah, so that that's interesting to be able to test that, because currently and there's also another issue that they're working on with error handling so right now we don't really handle errors really well.

E

So if there is an issue with a lot of you know, activity and you click generate the summary- nothing happens, you you get no feedback, and uh so uh that's all some other things I'm working on, because right now that you're handling around and all this stuff.

A

Makes sense, I was just thinking about um the user's degree of trust, especially like. If the. If we take your sample thing, should we make a mobile app.

A

uh If it were me using that and there's like a million threads I'd, almost rather ask a question which is like hey AI: should we make a mobile app and uh it takes it, summarizes it, but also maybe links to the comment where the final decision was made, because if it's a high uh risk, you know like if I'm looking for specific information and the AI gets it wrong, I would not solely trust the AI I might use the AI to double to have the inclination to double check. What the AI said so I wonder.

A

I can see a value of maybe a bit more of a conversational interface in the future and also uh citations like where did the AI get this information so that I can quickly jump to that.

E

Yeah, that's a really good uh good feedback. There I think someone brought up something similar that they would like to kind of somehow link to the threads from the drawer than into the actual uh activity, but yeah I love the idea of citations and uh yeah and just being able to ask it, like you, said, I think that's really an interesting way. Instead of you having again read another summary and then kind of try to map them. You know back together somehow.

C

Okay, so that that actually brings up one of my questions that I wrote next is: should this maybe be part of the chat UI.

D

That uh Sarah was talking about you.

C

Know maybe maybe yeah that's a good question. That's the drawer. You know.

D

That's the drawer that you basically.

E

Opened up right and that's kind of what I was kind of thinking more and more too is like.

E

It seems like a lot of these experience, because the chat is getting so good that you might just want to ask it certain things or maybe again, maybe depending on the context of what page you're on you might just it might hint at a question or like an example, question that you might want to ask like we were thinking also a little bit with this when the value stream analytics, instead of looking at all these charts and stuff and graphs and trying to see what's going on, you might want to just ask the interface or have the chat, interface open and ask it what metric changed.

E

You know. What's what's the most significant change, you know, why did it change and then- and it would just kind of give you links to what to look at or just give me the highlights you know what's what's going on, instead of you having to kind of look through it? So that's just again one case in maybe Valley Stream analytics where this might work, but uh yeah it seems like yeah.

E

That's a good question though overall is, should it should everything just be almost a chat now I was kind of against it at first, but it seems like it's. It's getting so good that it's uh it's kind of almost nice to have like a real person to kind of ask. You know what should I do here. What's going on, you know, yeah.

C

And I think that helps get around the quality issue, because uh you might think that you're saying something or expecting something to come up in that thread and it doesn't so, then you can ask hey what I thought this was about this and then it can search it and be like oh yeah yeah here it is.

E

Yeah, that's a good point, yeah yeah and we might like want to surface something first, the information, but give you the option to to pop open the chat. For example, in some of these cases, like to kind of you.

D

Can even do things like that? Doesn't.

C

Look right, I'm, pretty sure this thread was about this and then it can go back and reflect.

E

Yeah I could definitely see that that work in there yeah.

C

I mean when I've been playing with uh chat a lot while doing uh a side project. Just so I can get acquainted with the model of conversational, and if you, if you let yourself relax a little bit, it comes, it starts to feel very um sci-fi.

C

D

Can you can really just like.

C

No, that's not right, or is there an easier way to do that and because I thought you could do it some other way and it goes all the other you can and it starts telling it it's it's crazy, but that's I think that's pretty standard.

E

Oh yeah, no definitely I was trying. uh When was it last week, I was looking into it as well, and playing I'm trying to just develop a quick Django, app and Django View kind of app and I was just I just started. You know, I'm, like can I. Do this with uh open it or chat, GPT and I started, went down the same route and I was also having that kind of like Star Trek sci-fi moment almost where it's just like. Oh well, you should try this and this I'm like well. What about? What?

E

If I, did this and like? Oh it just you just started having like a real conversation almost, and that was that kind of blew my mind a little bit yeah.

C

E

I think if we lead in with.

C

The open being open to the chat experience it does help to assuage that concern around. If it's just text and you're like well, I guess, that's all I can get I can regenerate and see if it does something different. Otherwise, I have to go back and compare, and then at that point instead of really saving you time.

E

Yeah, that's a really good point. Yeah.

E

Yeah I'll uh I'll Circle back with uh Mike, Lee and uh and Alex, then on some of these experiences and and see how we could unify that, but yeah I love the idea of. Maybe you know here's some information that we think is relevant.

E

If you want to you know, if you want more, you could kind of open up or have the chat right there to kind of uh ask it more.

D

E

We're gonna get more details on cool yeah yeah thanks. Thank you both for typing.

A

All good um you just reminded me before uh You, released that um predictive uh chart for deployment frequency I think it was.

E

um Oh yeah, it was the forecast like uh yeah. We.

A

E

I think we just ended up using there's no AI usage there. We just use linear regression kind of some kind of linear regression model at the moment. I think there is some uh uh yeah I kind of need to Circle back on that issue to see uh what kind of feedback we're getting on that cool.

A

E

That's it um for.

A

It's out like customers can use it. It's not just internal.

E

I think so I'll have to double check on that yeah yeah. Let me make a to-do for myself here.

C

So yeah and uh just to bring it up, I know that uh Pedro made a comment in the slack Channel and I alluded to it earlier. um He is putting together a conceptual mental model framework thing that he's going to do a walk through on a quick video walkthrough and then share it with all of you.

D

All the designers in the ux AI Integrations Channel for feedback.

C

I think it's pretty awesome very simple yeah, but you know to the points. Succinct, wise, but I think it'll be pretty helpful to help us uh get consistent and wrap our heads around the patterns that we want to use so I think he was shooting for the end of this week to get a video out or something.

A

Also and yeah speaking of Pedro I mentioned it before we started the recording, but I might actually push this meeting to be like one hour earlier, uh so that Pedro can also attend this. So that means I, guess labor.

E

A

To be sleeping, yeah.

E

I might watch the recording then after.

A

Yeah, you can go to the Apec one and say you Justin and yeah.

E

A

We have low attendance for this one then I'll cancel it as well. It's just an idea for like Becca and um there's a couple of other people in America, Europe, I think so.

C

You know the timing an hour early won't be bad for me. It'll just be bad, because I usually have a lot of my other meetings in the mornings. If I may not be able to attend because of that. Well,.

A

Well, yeah: we can just play around with it, but yeah I was thinking some kind of good. If you can come to at least one Justin and Pedro could go to the other one, and then we have kind of the full coverage.

C

All right is there anything else going on.

D

C

Right cool thanks.

D

C

Then, thanks so good to see you again have a good rest of your day.

D

C

Thanks everyone.