Django and AI: A Community Conversation

This video features Laura Gates and Thibaud Colas at DjangoCon Europe 2026 in Athens, Greece.

Django and AI: A Community Conversation
1:27:32
Published October 26, 2026
93 views

WORKSHOP: Django and AI: A Community Conversation by Laura Gates and Thibaud Colas

This workshop proposes a facilitated Open Space-style discussion that brings the Django community together to surface, explore, and document the key themes, concerns and opportunities around Django and AI. Rather than prescribing answers, we want to harness the room's collective expertise to map the landscape as practitioners experience it. The session, hosted by Thibaud Colas and Laura Gates, will use a participatory format drawn from Open Space Technology, where attendees propose and lead breakout discussions around the sub-topics that matter most to them – whether that's AI-assisted development, ethical frameworks for Django projects integrating LLMs, the Django Software Foundation's approach to AI policy, or something we haven't anticipated.

More information: https://2026.djangocon.eu/speakers/JUFWZE/

Follow DjangoCon Europe on social media:
*X - https://x.com/DjangoConEurope/
*LinkedIn - https://www.linkedin.com/company/djangocon-europe/
*Mastodon - https://fosstodon.org/@djangoconeurope

#django #djangocon #djangoconeurope #djangoconeurope2026

Transcript

8,999 words · auto-generated · en Show

Automatically transcribed, so expect mistakes in names and technical terms.

0:03

Speaker 1: Now is the workshop Django and AI Community Conversation. The decision is led by Laura, an expert in human-centered AI, and Thibaut, a Django Current team member, and the product lead for Wagtail. They have designed this community to explore this topic together. So please welcome Laura and Thibault.

0:35

Speaker 2: Thank you. Well put please welcome the whole room. We're gonna have you do a whole lot here today. This is very interactive. Very briefly, I'm Thibault and last year I was president of the Django Software Foundation. Woo! And um I do lots of open source, both at my day job at Torchbox, working on Wagtail, as well as in my free time. AI, I think, is a super important topic. It's a topic I'm feeling very burnt out by. It's very stressful. I also do think it has the opportunity to be very impactful in both good and bad ways. For all of us. So this is kind of why Laura and I thought it would be a great topic to bring to you all today. The topic of Django and AI, there's lots of aspects to it.

1:23

Speaker 2: There is the value that people are getting out of Django as a product. People building things with Django and AI. There is the impact on our community, the impact on the Django Software Foundation. It's a very profound paradigm change across lots of areas. We've asked you all for help to divide this up into a few discussion themes for today, which you see around the room. Should I be going through them now?

1:55

Speaker 3: do the introduction then you go through them.

1:56

Speaker 2: Okay. So I'll hand over to Laura who'll explain the form introduce herself and present the format.

2:01

Speaker 3: Yeah hi I'm Laura Gates. Um I live in the UK and I work at Bathspow University. I'm an academic in theater and performance. I'm also a software engineer. Um and I f I focus on um methods, creative methods for feedback and impact data collection. So one of the things we'll be using in today's session is an app that I've um a platform I've built. This is its early user testing, so expect some bugs and let me know if you find any. But it will help us collect hopefully um everyone's voice in the room, which is uh what it is intended to do. Because sometimes large groups for a discussion don't tend to lend themselves to every voice being heard.

2:49

Speaker 3: So we want to find ways to hear every voice. Uh this is structured as uh using open space technology. Has anyone encountered an open space event before? A couple people, a few people, great. So I'm just going to quickly read the rules. So that the rules of open space is number one: whoever comes are the right people. Number two, whatever happens is the only thing that could have. Number three, whenever it starts is the right time. And number four, when it's over, it's over. It also operates on the law of two feet, which means that in a moment you're going to split into smaller groups You're not obligated to stay in

3:35

Speaker 3: any one particular group. Um you can operate in one of three types or in your own type. We have three types of engagement. One is worker bees, one is bumblebees, and one is butterflies. Worker bees you can think of as um being in the hive, really working at building up the honeycomb, the home of the hive. So that would be people who like to stick with one group, stick with one topic, dive deep into it. Then we have bumblebees who move from flower to flower and they cross-pollinate. So if you're the kind of person who might want to sit in one group for a bit. get up, go to another group, and share some of the ideas. That's cross-pollinating. And then there's butterflies, people who just flit around and they don't have to pollinate. They don't have to contribute in any conventional way.

4:21

Speaker 3: It is a contribution itself just to be in the space and moving through it in whatever way you are most comfortable with. This is a very truncated version of open space. Usually open space events are run over several hours, usually at least a day. So we um have truncated it. But They are very good at generating themes and ideas that people bring into the space with them Um and to that end, I would like you all to, if you haven't already, can you go to obwob. app? So the sites up there, obwob. app, and the join code is 591072, or you can scan this QR code.

5:14

Speaker 3: All right, everyone in? Great. And I'm going to ask you Before I push the first question, to turn to the person next to you and take one minute to talk about what thought about Django and AI are you bringing with you right now? to the person next to you, take one minute, I'll time it, to talk about what thought about Django and AI are you bringing with you right now?

5:46

Speaker 4: So after I do this question, then I'd recommend the topics because I didn't want to narrow the thoughts. Yeah.

6:43

Speaker 4: That's a line.

6:49

Speaker 3: Okay, start to wrap up what you're talking about. Great, that's just the beginning. There'll be lots of chances to unpack all of this more. If everyone can get the ABWA app up on their phone, and I'm going to now push that question and if you could individually answer it.

7:55

Speaker 3: What do you call me that you want to do? All right, if you're still writing, that's absolutely fine. Um Tipo 's going to say something for the people online about how they can participate as well.

8:22

Speaker 2: Yes. We very much want to capture what people here in the room are discussing and also the people online joining us remotely. For those people joining remotely, you'll already have access to the same survey um app via um the Zulip chat we've put it in there already and uh please follow along as we discuss things here in person follow the same prompts, the same questions in the app, and you'll get to benefit from our recap at the end of our session and we'll then integrate your input into that recap. after the workshop.

9:01

Speaker 3: Great. If you're still wrapping up the question, that's still absolutely fine. um just to frame what this uh the space is intended to be a space of experimentation. Think of it a bit like a lab, like we're playing with these ideas and experimenting with them. So in that spirit of openness and and act kind of a we're we're here to work through these topics together. You'll see that there are little mini whiteboards set up around the room. When you came in you grabbed a chair that was in a circle around one of those whiteboards. In a moment. So Tiva 's going to read what each of these whiteboards has a different topic on it, generated from what we've gathered from your responses over the last couple

9:50

Speaker 3: days. We've collated them into key topics. If you did submit a response into the app, we have that. It will be part of the output. So if you don't see the detail of what you put, we still have that. that'll uh be put into the community conversation. But these are the collated themes. So Tipa's going to read what the themes are. Have a think about where you want to start. And then walk around the room, find your whiteboard, and create a circle around it, and then we'll give more instructions.

10:21

Speaker 2: Thanks, Laura. Um yes, so the themes, we've picked eight of them that are primarily focused on people using Django, our community, and uh open source projects. So I'll start describing them around the room and start over here. So with building with AI and Django. So practitioners on Django projects, developers using AI. I think it's worth to say even if the topic doesn't say AI explicitly, it's an it's an implicit theme in all those discussions. So up next, learning and preserving skills. How do we encourage skill development in juniors and seniors? How do we prevent atrophy of skills over there? Open source contributions, so people contributing to Django

11:07

Speaker 2: or the wider ecosystem with AI. core maintenance, so kind of this other maintainer side of open source. And in the back of the room here, optimizing Django for AI. Should we? Could we? How do we? Who gets left behind? So this idea of potential inequalities in who has access to AI? How we navigate the hype? I would love to know the answer. And finally, Django's continued relevance, big topic to unpack. So there's eight of them. And do you want to say a bit more about? Yeah.

11:43

Speaker 3: Um yeah, just whatever struck you just now is the one that you want to join first or join and stick with, uh head over to that whiteboard, form a circle of chairs around it. There will be pens. and a whiteboard eraser there. And uh and I will give further instructions once you're clustered in your groups.

12:15

Speaker 2: What should I do during the discussion?

12:19

Speaker 4: Yeah, bumblebee.

12:21

Speaker 2: Okay. The bumblebee is cross-pollinating the areas. It

12:26

Speaker 4: has well, yeah. No

12:28

Speaker 2: one, yeah.

12:30

Speaker 4: Seems to make sense.

12:52

Speaker 2: Do you two know Are you the only person who cares about it? If if you wanted to have a discussion about this with just you two, that's fine. If you feel like you'd rather go somewhere else, that's also fine. So just see how it feels.

13:14

Speaker 3: Okay

13:18

Speaker 2: I might come I might come say hi.

13:22

Speaker 3: In each group , Can you nominate a scribe? This is the person who will draw on the whiteboard or write on the whiteboard as you're having your discussion. And at the end, we're going to come back into a big group, and somebody from each group is going to present On what you wrote on the whiteboard. So it could be the scribe, or if you want to nominate a separate presenter, that's fine. Just know who's going to present at the end. Know who's scribing, know who's presenting. In terms of the scribing, it can be in any form you like. It can be done as a mind map, it can be done as bullet points. a haiku, a picture, whatever form

14:09

Speaker 3: you want to express the discussion in is great. All right, now I'd like everybody in the groups to turn to the person next to them. And Take a minute to discuss this question. Why have you chosen this topic group to start with? So turn to the person next to you and take one minute to talk about why did you cho choose this topic to start with.

14:45

Speaker 5: It's like everything that they can do out of it and it's just really kids like, you know, no one really feels about auto people. Right.

15:14

Speaker 3: Yeah. Exactly. Yeah, yeah. Thank you, 'cause they're trialing up the app and I want pictures of people using it. Yeah, I'll put that for Okay, wrap up what you're talking about. Wrap up what you're talking about. And if everyone could pull out their phones again, this is just, I'm gonna give you a moment to independently put in an answer to that question in the app. And then we'll move on to the group discussions.

16:01

Speaker 3: So take a moment to enter your response to why have you chosen this topic group? Join the other things. What do you join the summer food?

16:59

Speaker 3: Okay, as people finish submitting their response to that question Just go ahead and start the discussion in the small groups. I'll let you guys do that on your own timing as people finish up. All this up

17:59

Speaker 3: I want to share that Yeah, there's a clean uh based on that comfortable. I don't think it's a

20:04

Speaker 3: Yeah, we'll use that one.

20:53

Speaker 3: If any group needs more than one whiteboard paper, we have more at the table over here, so feel free to come grab some. Yeah. Yeah.

22:05

Speaker 3: I won't do it. Uh close like lending code. Yeah. Uh there's no shape that they probably Uh oh, I just don't like that.

22:54

Speaker 3: So we can look at the time as well. We were going to have an automation Uh hard And just like that you can't return to your point.

23:41

Speaker 3: And then we're saying it's a company. Do you accept the company because any of the church would have spent all do you expect it? So there's some stuff

24:35

Speaker 3: Yeah, you might use the windows in the window with the wire. So I can find out the button and then after that

25:39

Speaker 3: I think it's hardly because Yeah, that's good.

26:41

Speaker 3: But I would also do that. Uh you can do it on the two.

27:33

Speaker 3: So I 'm going to Maybe often that we can have to do that.

28:33

Speaker 3: Um then maybe how we can optimize uh documentation for this or something. Did you get a simple workshop yesterday where we talked about declining traffic in docs? Whether that's declined or that 's I uh suspect that donations uh from people

29:01

Speaker 2: actually go back. That's where we showcase our sponsors so our sponsors do pay us so people see that they sponsor Django. So is that a success metric or a failure metric? I think it it depends uh from what is functional system. Yeah. Do we have metrics to start with? Do we need metrics? I think so, but just like I don't have a phone, I don't have a phone. Everybody's just searching where's the react iteration.

29:52

Speaker 2: How to upgrade to join those six total? We have server logs for seven days, I think. I just don't know who has access to the logs. I think they could be useful to know if you want to. Oh yeah. So lots of this are about documentation, it looks like, and better meeting the formats that people want these days. I don't know.

30:37

Speaker 2: Delivering AI features by uh I see yeah. How does that fits with the package ecosystem then should the package ecosystem in Django also do those things? Should those things Yeah, so the skews for example, if you make a package, you make skews that go with that package? Yeah, it's interesting. I

31:20

Speaker 5: it's always gonna be up to the back of the new dinner, right? Some people are still hating the eye up there, so

31:25

Speaker 2: yeah. It feels like when we say optimizing Django for AI, what we mean is like for people who use AI rather than AI itself, right? So at least as long as we are clear about that. That I lots of those things they feel like they would be quite helpful even for people who don't use AI Okay, well I mean well going back to docs and then everything is docs. But depends on yeah, if if the docs became organized more logically as a side effect

32:11

Speaker 2: of reorganizing them to be a file and marked in Do you think that's as high of payoff? Well structured and also readable. Yeah. I think

32:37

Speaker 6: we used to have well I think we still have really good documentation relative to other projects, but it's the docs are run quite organically all the time. And then maybe Can find some long pages for example.

32:50

Speaker 2: Yeah. Long pages is a big one that could be great for everyone. At least for for Wagtail, my other projects, it feels like the reference docs are helpful but people really like to have how to documentation for the specific thing that trying to achieve. So you really need multiple types of docs to make it more accessible. It's not just like look at the reference for that.

33:21

Speaker 6: Yeah.

33:39

Speaker 2: Did you talk about um changing Django's APIs? Like Like for example typing. Uh yeah, things things are uh prototyping of like Things that are in Docks are pretty good. Um yeah, okay, yeah.

34:16

Speaker 6: I do think um Um yeah, type annotations will also help for sure. Um just the more type annotations will help I think because the more the more things that are keeping the AI in check the more likely it is to not go off and do waste.

34:36

Speaker 2: So the underlying thing here I think is QA capabilities if you could verify that a Django program is correct to some degree, then the LM can do that and it's much easier for you to mostly write code So for example for Wagtail we would encourage people to have something that goes through all the models on your projects and tries to instantiate them and just check that that works So I don't know if QA things like that have a place in Django. Or just like a better Django linter, for example. Like you know there is um the checks framework in Jesus.

35:23

Speaker 6: I feel like the system checks are very slow though at times. System checks can be quite slow to run and asking your AI to run the system checks all the time is quite hard. If you run like a fast Lintali Ruffa or something, then you can just tell it, oh yeah, it's gonna have that.

35:42

Speaker 2: So the check yeah, the check framework, is it to check the correctness? of the code or is it essential that it's at runtime? Yeah. Okay.

35:53

Speaker 6: So

35:54

Speaker 2: 'cause it feels like you could replace that with a skill and the skill could Tell the AI like oh here is a check that you should do. It doesn't necessarily have to be at runtime to be helpful always, or actually it does sometimes, yeah, still. There's a static version of it because

36:12

Speaker 6: it's gotta build all the models out and stuff.

36:15

Speaker 2: Yeah, okay. So wait, how slow is too slow? What what use case are we talking about?

36:32

Speaker 6: after every uh change of things. And for some things that's really fast. And for a few things I can't really I can't really have really tests that we're going to do.

36:46

Speaker 3: Okay, in about ten minutes, we're going to be wrapping up the small groups and coming into a big group to share. the group discussions. So start thinking about how you will present what's on the whiteboard so that you can convey your small group discussion. Just maybe some small

37:25

Speaker 2: type locations. Um the possibility direction, maybe it could be worth it to look at the doctrine so the classes uh method structures etc but um maybe it's uh Just a thought. Uh and don't know if great suggestions, how to improve it, but again say um if you look from the perspective to improve it from the standpoint and try to make it more readable for the users themselves that it would affect what's more

38:24

Speaker 2: Yeah, I think that is I'll head off in about ten minutes. Think of how you want to present that, if it's one of you, if it's multiple of you. Sounds really good so far. Mr. Harbin down. The bottom button as a lot of response. I know, I know. What are you expecting?

39:35

Speaker 2: It's still closed up and they are open up uh ref stations.

40:05

Speaker 3: Whatever setting of the microphone goes into the recording, right?

40:10

Speaker 2: I'm gonna start moving the ones that are over here and just display it over here. And I think in terms of people we can have a small

40:23

Speaker 3: Yeah, yeah, we'd be good. I think you're right you have. Um so in seven minutes I'm gonna have them do the final question. Okay. And then we yeah, then we need to get those whiteboards over here. So

40:37

Speaker 5: do you want to do six

40:39

Speaker 3: minutes? Semicir. Um circle. Okay. Oh, you mean 'cause to present.

40:46

Speaker 5: Some people want to be able to see then.

40:48

Speaker 3: Okay. So you're thinking like the all of the blackboards will not

40:52

Speaker 2: well. That's great, okay. Yeah it's gonna be a junk or something and pump on the back end. And now I can apply a pump inside the junk and uh say right uh model and then Can you run inside the Django

41:37

Speaker 2: or do you find it? Uh , and actually like the I don't really wait for the like smaller

42:29

Speaker 2: If you cannot find yes, I mean I think that's a bigger two.

43:39

Speaker 2: And that is a couple Do you remember actually used to use Ring for

44:09

Speaker 6: the service? I think you would make the topic. I think I would have uh had time to still something uh for hearing just because we've covered so many uh the wide variety I think one call for the theme that it might be nice to focus on a bit. Um there's a lot of good things that we could do for Okay , yeah, yeah.

45:08

Speaker 3: Okay, in about three minutes, we're going to start transitioning into the next phase. In about three minutes, so think about wrapping up this part of the discussion, getting your presentation ready. Not a formal presentation, just presenting the ideas. I think I think so. But then you also then do something very

46:07

Speaker 3: Um white three I'm gonna ask them to question a lot of service And I think very soon But if you come from a background because they can have a back of the

46:54

Speaker 3: One more minute. One more minute till we move on to the next phase. Actually, do these uh So yeah, first of all. And uh

47:22

Speaker 2: when you're done with the first Then we move all the time and we can make it useful together.

47:35

Speaker 3: Okay, if everyone could wrap up what you're talking about right now. There's more opportunity to unspool these ideas, but just wrap up what you're talking about right now. And if I could get everyone's attention. Okay, so stay in your seats, stay where you are. Tom in your group. And pull and turn to the person next to you and take one minute. To answer the following question, discussing it with the person sitting next to you, what's one new thought you've had about

48:22

Speaker 3: this topic? What's one new thought you've had about this topic? Take one minute to discuss this with the person next to you. It

48:36

Speaker 6: occurred to me before that all this stuff can be up to the Um they can see it.

49:28

Speaker 3: All right, wrap up what you're talking about with the person next to you and please pull out your phones and get the ObWOB app back up on your screen So get the app back up on your screen. I'm pushing that question. Take a moment to independently answer it. What's one new thought you've had about this topic? I think it's a time to do it.

50:33

Speaker 3: So if you wrap up what you're saying with the person next to you and if you can answer the question into the app. We'll make sure we gather everyone's responses together.

50:43

Speaker 5: Are you okay to do seconds in the group

50:46

Speaker 2: presentations? presentation we just read off here. Yes, okay.

51:04

Speaker 3: Okay, we'll take one more minute on answering the question in Obwab , and then we'll come into a large group. All right, if you're still answering the question, that's absolutely fine. If you have finished We need a big group effort to create a large semicircle facing this wall. So it would start about here, arc around this way.

51:50

Speaker 3: And end over here. We'll put that against that one. Thank you. Maybe stop Yeah, right.

52:45

Speaker 3: Yeah that 's what you 're doing. Like clouds, yeah. Yeah, but you can jump into Oh yeah, but what can we do? Could we

53:27

Speaker 2: want to

53:27

Speaker 3: Could we move back a bit over on this side and get some more chairs in so we can fit everyone into the circle? Thank you. We should s we should um sit. Maybe on the side. Where do you want to stand? Do you want to side of standard? Sorry. I have some experiment and not get side framed or the point. Someone say the other five.

54:12

Speaker 3: Okay So, based on that last question, um was one new thought. Hold that in your head. And what we're going to do now is move down the lovely whiteboards that you've generated in your groups. with whoever your nominated presenter is presenting for us. So starting with optimizing Django for AI, who's presenting on that? Jacob.

54:43

Speaker 5: Thanks. Hi. So our group was talking about optimizing Django for AI. Immediately we had to clarify, oh, we're really talking about users of AI? Is that what we mean by four? AI, and that led to, I guess, a couple of big themes. One was that could mean delivering AI features via Django for users who want to build AI features. And then there are users that just want to build features. features. So there's where that led to the other, I guess, major theme we discussed, changes in Django or Django's pack, you know, Django's package ecosystem to help LOMs suggest better Django patterns. or just understand Django

55:29

Speaker 5: better. So we spent relatively more time on that topic. Um does that make sense the difference I'm drawing there between building AI features versus just building Django. Okay. So we can go quickly over the topic that we discussed last delivering AI features with Django. The comparison with Laravel came up first. Larivel has Laravel Boost. So we tried to imagine what could we build in Django. We don't want to build Clippy. Hey, it looks like you're building an API framework. But you might want to be able to use, you know, REST framework endpoints. Um on an MTP server. And so we're not sure if there should be tooling to help help you get help you bootstrap that. Um but like I said, we spent most of our effort thinking about changes in Django. We spent a lot of effort thinking about documentation, maybe after The influence of Thibaut's workshop yesterday.

56:14

Speaker 5: That immediately got us thinking about should we and how. Then we got sh we realized should we and how both kind of revolved around how much effort is this gonna take. And what are the opportunity costs? What are we not building because we're changing the structure of our docs to compile the markdown for an agent to consume? But zooming out and thinking about our motivations, it was pointed out that Should we? Well, you have to think about we're helping users if we do, and Django has a good track record of doing that, so it seems logical. Flip side of that would be, well, isn't it the tool's job? If the tool can't do something Isn't it the tool's responsibility to improve? That led to a discussion about what improvements can we make that will benefit everyone.

57:00

Speaker 5: An analogy with accessibility improvements often benefit everyone, the so-called curb cut effect. So we started enumerating a bunch of potential problems. We have potential long pages in the docs, missing how-to 's, no doc strings, and then a discussion about whether those things matter. Um yeah. And I guess ultimately like the idea that we'd need a prototype one area to have a self-contained small problem to work on. similar to what we might do with type the usefulness of type annotations in Django, right? Demonstrate value and feasibility by focusing on a small area instead of focusing on the whole thing. That led to some more ideation about, you know, linters for Django, the check framework built into Django.

57:46

Speaker 5: Would it be more useful if it was faster? Uh and in and one last interesting piece, should we use site analytics to drive some of these decisions? What are people searching in the docs? And if there are you know improvements we can make to let the LLMs answer those questions, will that cause a decline in traffic to our docs? And is that a success or a failure situation?

58:11

Speaker 3: Thank you. All right. Who gets Who gets left behind? Who gets left behind? Who was presenting on that?

58:30

Speaker 7: So this is all of these effects pretty much all open source packages, but obviously specific to Django. Sorry. One thing is that AI tools generally are targeted at higher GTP economies. So if we're aiming to create a package that's ease of use for all of our users and we're spending time and effort building features which is only going to be accessed via developers in higher GDP economies we're sort of targeting the West over anywhere else. The problem is likely to be increased as companies sort of profit from the hype around their tools.

59:20

Speaker 7: And bringing back to Thibaut's talk yesterday of companies leading and uh leading like their LLMs to go for certain tools over the others we're sort of incentivize to make our tools uh our tooling in general more ai accessible um another one is OS open source maintainers so you you're receiving, reviewing, too much, too much, well, slot basically. So there's an opportunity cost to be acknowledged here if we lean more as a platform slash org towards AI, you're encouraging adoption for other open source packages as well

1:00:07

Speaker 7: that rely on Django. Yeah. There's plenty of conversations online in terms of like whether or not packages or well various formats should be AI slash SEO. optimized and the impact that may have and that's a similar reason to like companies pushing their own tooling. Whether or not that could affect Django long term is a different question. in terms of will it rank it higher if it is well uh well optimized for LLMs or could it just simply be blocked if it's not? The biggest one I think in the room is like junior engineers. There's a lower barrier to entry now because of LLMs. They're cheap. If we make Django better and easier to pick up, that's great.

1:00:55

Speaker 7: But now we've got tons and tons of engineers who may or may not be you know reliant on these paid for tools full stop. That obviously has knock-on effects of simple slight supply and demand. There's more opportunities. Well, if there's more opportunities to learn then the there's more competition, salary perhaps Pressure, etc. Uh more to the DSF sort of level. Um funding often comes through donations and that comes through various channels but the main one being Django docs bringing people first to the platform

1:01:41

Speaker 7: and then on to you know the community in so in itself, but if we are leaning away from that and we're leaning towards LLMs, there's not really an incentive or doorway into that. side of things. So we are sort of cutting off a path of monetization that we wouldn't have had that we did have before. And then I guess the last one is if we do make these changes, then we leave behind the people who have been either skeptic or negatively impacted by AI in general. Um I think we Whichever route we take is going to alienate people. Some people will be super pro, some people will be against. It's just a choice that's going to be made

1:02:26

Speaker 7: with or without our choice. But yeah, that's it.

1:02:37

Speaker 3: Okay, and now we're on to building with AI and Django. Who's presenting on that? I'm really sorry, just for time to make sure we have time for the final discussion. I'll be people have been keeping relatively in this time. I'll be cutting you off at three minutes.

1:02:56

Speaker 7: Okay, I'll try to be fast. Uh we were discussing about building with AI and Django, and we figured out there's two ways to look at it. One is to use AI for development, and the other one is uh integrating AI into our Django projects, right? So we were focusing mostly on this first branch, the uh use uh AI as an assistant for our development. So first we were discussing how non-tech or zero dev experience people personnel can now write code and implement functions, function functionalities. Uh then we were also talking regarding that about losing our jobs, which I think even though most of our group was like pro-AI and we are building with AI, we are still some of us uh scared of losing our jobs and I we think that the landscape is going to change.

1:03:45

Speaker 7: Uh then we were talking about planning and refactoring. So how much time you should And how much time you should try to spend on the refactoring phase. Uh in addition to that, we were talking about the code review and what are the changes if you review code that you know. So basically if you're an experienced Django developer, you look at the code that the AI has written differently than if you review code that you actually don't know. I don't know. You don't know JS and the A AI has written some code. You don't even know how to review EV Even if you wanted to, right? But for the Django, if you know it, you can like scream skim through it and see a lot of stuff if the AI has missed it. So we were talking about those differences. Um what else? Oh oh we were talking

1:04:30

Speaker 7: also talking about the defaults that that uh AI goes to. So basically if you start a project, what framework will the AI choose to build your back end, to build your front end if you don't prompt it Like from our experience it goes to React for the front at mo um mostly. This is probably to due to statistics because it's the most problem the most likely outcome, right? Um okay, well tells. So we were talking about self-validation, so we need to provide an AI or agents with ways to self-validate. and I don't know spawn separate agents that can check the code that the other agent has written. We 're talking also about the increase in demand for productivity either from your bosses or from your clients

1:05:15

Speaker 7: or from whoever you serve your code to. Uh we were talking also about the security concerns. that we're all aware of, like uh your agents reading your. nv files. Uh and uh a member of our team suggested there's something called clot condom, right? which is a thin layer around your cloud that uh is supposed to mitigate these issues. Um so yeah this is the things that we discussed

1:05:49

Speaker 3: Thank you. And we're on to core maintenance of Django. Who's doing core? Emma.

1:06:00

Speaker 8: Thank you. Uh so to stick to to the team we soon realized in the discussion that there was two sides of this, contributors and maintainers. Um so let's let's start with the the contributions. Um this uh AI uh opens the door to less experienced contributors to try to contribute to Django, which is something that historically we've been welcoming. and asking for. Um there's uh this can lower the barriers so without uh going to uh AI generated good If I don't speak English, today I can contribute to Django

1:06:48

Speaker 8: and I can just use an LLM as a translator. It works much better than Google Translate Um but this is this also means a lot of growth. Um no that's the other topic. This also leads to question about growth of the developers. So one of the reasons that we uh welcome new developers and that we encourage contributions by uh more junior people in Django is that in the hope that they are going to grow as a developer and can become experienced uh contributors and maybe even maintainers of Django But is AI helping with with that growth or is it hindering it instead?

1:07:33

Speaker 8: So that's for the that's for the contributor side. Now for the maintainer side, uh there is um uh uh a recurring thing which is uh the overwhelming effect of AI on maintainers. For many reasons. One is the amount of PRRs that are coming in Uh two is also the um the quality of of the peers and the time that has to be spent on a peer to know if it's a good peer or not a good peer. Um but there's also plenty of bad peers that come in that

1:08:19

Speaker 8: would That are made by AIs, but that we don't care if they're made by AIs or not. They would be bad pure from the get-go if they were made by humans, something like a 2000 like this. Um one so one of the solutions that we've been talking about uh uh to to fix that that overwhelm and and burnout of the maintainers is what would be the ways to also amplify the work of the maintainers And there's been different solutions given between f AI and non-AI solutions to try to amplify uh the work of the maintainer as much as

1:09:07

Speaker 8: the other side of things is being amplified so the number of peers, the qual the the bad quality of peers and that thing. And uh one of the things that came Okay, the last is that um this shone lights on the fact that um AI in itself is a tool for bad actors to generate burnouts on mass generating burnout on maintainer. if there is a bad actor that wants to burn out maintainers to be able to take over a project or try to uh push a malicious peer onto the project or something like that, it's easy to do now.

1:09:54

Speaker 8: And the the main thing is that we we we realize it's just the start of the conversation and there's a lot more things that need to be said.

1:10:07

Speaker 3: Thank you. Um open source contributions.

1:10:20

Speaker 6: Thank you. Um this is going to sound like a rehash of the film. things you've heard so far. I feel like we're already converging as a group. Because one of the first things we brought up is whether or not open source contributions are sustainable. In the LLM and A agent era because there's more fixes coming in, more code, but also more slop

1:10:44

Speaker 7: and uh it might lead to maintainer burnout. Um so uh there's also the case that different people with uh different values or agent files uh will provide different types of PRs that may or may not match the project that they're trying to contribute to. And uh there's the risk that people will not develop their own skill set uh because they're using agents uh as their primary mode of operation. uh and thereby it's like they're robbing themselves of a learning opportunity. Um so far it's all uh uh risks uh by the way the opportunities will come soon. Um there is some positivity here. So there's also issues with trust.

1:11:31

Speaker 7: So it's already difficult to trust people like other contributors in general. That's why. Why we have reviewers, right? But now we have like a second layer of indirection in trust, so we don't just have to trust the developer, but also their agent And there's the inherent asymmetry. So it's it's becomes easier to contribute but harder to review because there's more to do. Uh and uh that might lead to a situation where it's easier for the maintainer to

1:12:02

Speaker 6: redo the PR from scratch with their own agent. than it is to review the PR from the contributor, which can then lead to like a negative feedback cycle where people are disincentivized to contribute in the first place because the maintainer will just redo the work anyways. Um so then that that leads to like ecosystem uh collapse. Uh and now for the positive side, um it could lower the barrier to entry for for people, so it might democratize

1:12:32

Speaker 7: code more, which is good, right? Um and then uh let's say that we have a a a near future where AI are are better. We could have automated code review. It could reduce the feedback time for new contributors. So once you're contributing a PR, you you want like feedback. And the longer that takes, and that to be frank is sometimes measured in months, um

1:12:59

Speaker 6: it it doesn't really help you with your learning cycle uh if you're a new contributor. Uh so if you could have automated code review that could make it faster. Uh but the inherent risk is that it could reinforce bad bad practices uh if the AI will give you basically poor feedback and you're not experienced enough to recognize that. Um or it might like reinforce the wrong style. So

1:13:25

Speaker 7: it realized in the contributor knowing enough to recognize The good versus the bad feedback. Another opportunity is in double reviewing, which is that we could have like human and LLM review at the same time, and then the LLM might catch things that the human missed. And um I think that's just about it

1:13:52

Speaker 3: Perfect timing. And now we're on to how to navigate AI hype.

1:13:58

Speaker 2: Thank you. We had a very brief discussion about it, about this with a small but mighty group. I think we just first of all shared that it was a problem and a source of fatigue for us to navigate the hype. and we're definitely hoping that eventually the underlying value will emerge. The signal will be there, not the noise. We reflected on the fact that it was quite aggressive pushing. And the fact that it was in some occasions from companies, but also some pressure from the people around us to some degree, whether it's our boss, whether it's our customers, sometimes our colleagues as well, so that there's two elements there. And as far as how to respond to it,

1:14:44

Speaker 2: it was basically a matter, in our opinion, of uh playing the playing the long game. being aware of how things are heading eventually, investing in your own skills so you're better able to benefit afterwards after a while. And in the meantime also two things in the meantime , trying things for yourself, so make your own opinion. And also keeping in mind being aware of the negative externalities of AI that are completely absent from the dialogue of the hyperbole. Yeah, short and sweet discussion.

1:15:18

Speaker 3: Thank you. Django's continued relevance. Who's speaking to that?

1:15:28

Speaker 7: So I was over here. Uh it was the first group I chose and there was nobody there. Um which Is that a sign of Django's relevance that no one wants to talk about Django's relevance? You know, you decide. I jumped over to how to navigate AI hype, but then I had to come back to Django's continued uh relevance because for me there was a discussion at work you know this thing is super old you know do we still want to use it and there's a case to be made that old technology is safe technology is technology that you can use, um, you know, because it's reliable. So just because Django is old doesn't mean it's bad. Um everyone in this room is using it. There's hundreds other users that are not in this room, you know, that still use it.

1:16:13

Speaker 7: The second thing uh which is very important is kept up with the times, both core and third-party ecosystems When I first started with Django is Django 1. 6, it didn't have migrations. They added migrations in 1. 7, and for me it really like opened it up. Since then we've had uh async stuff in there, channels. Um you know, we just got tasks. Uh so it's really not something that's dead. It's um it's still alive. Um the third thing we discussed is do we want Django to explicitly support AI or to focus on humans? These things aren't necessarily exclusive of one another, but um you know for me as a bit of an AI skeptic I would rather just focus on humans but it's unavoidable that at some point

1:17:01

Speaker 7: um you know AI enters into the discussion which is the reason I'm I I'm here. Um and along with that, you know, I wanted um to write will uh will AI still be here in five years in three years, but we decided on five years because three years is a bit uh too short notice. If it's if the bubble pops and everything like explodes, like what do we do with the pieces? We get to keep the shiny pieces. And how does that affect Django in the long term and the short term? The last one was uh Django's great docs give AI something to work with. For me, the docs are sort of like fodder for the machine. Um Because

1:17:46

Speaker 7: a lot of the things are extensively documented, AI is probably particularly great at using them and producing Django code. So that's the last thing to consider. Which angle's relevance.

1:18:04

Speaker 3: Thank you. And finally, learning and preserving skills.

1:18:16

Speaker 4: So um instead of just going over what we chronologically talked about, I think it's probably a better summary if we go section by section. So essentially we came up with a bunch of issues with learning, but we also tried to spend some time talking about what are some of the opportunities and what are some of the solutions that come to the problems that we've talked about. So some of them you're probably going to be familiar with. How do we judge AI output AI output if we are not sure of our own skills, if AI is degraded. our skills, how do we teach core coding to new juniors in the future? And there's of course that tension between productivity and learning. Do we, you know, there's a push towards focusing more on productivity. productivity, but then you also need to spend time learning where

1:19:02

Speaker 4: where within the process do you find time to do that. Then we got a little bit meta. We of course talked about do we even need to preserve certain Are certain skills necessary for us to keep in mind, or is that something that we should just move to the agents and trust that they will do it? Um, which obviously got us back to the definition of what program programming itself is. Is it just about writing the code? Is it just about creating software? Is it or is it more about creating a more holistic product? And one of the opportunities actually that we came up with besides the opportunity obvious one is a learning aid is it presents the opportunity to perhaps become more full stack, be more involved not not only in the back end or front end but also in pr in design, also

1:19:48

Speaker 4: in uh conception. If we're spending less time on code, that means we hopefully have more time to have a more integrated understanding of what we're trying to achieve together with the rest of our team. outside of the engineering team. One of the things that came up which was unexpected was there was a quite a bit of pushback about accessibility accessibility. We heard a lot about how AI can remove barriers from a lot of the other groups and make things more democratic, but we actually found quite a few ways in which which uh AI is reducing accessibility. So things like how now if you really want to learn how to code well you need to have an expensive

1:20:33

Speaker 4: subscription in order to uh code at the level that is expected of you as a new coder. How do you know what you don't know if you're only asking the AI, there's less access to experienced mentorship now. That used to be a priority before, and now all if all of the seniors are spending their time on coding. and expecting juniors to use AI to teach themselves, then we're losing, even juniors have less experience now to good quality mentorship And there was definitely something a bit of uh resistance that came up. Some of the juniors uh felt that uh essentially what there there seems to be an arbitrary cutoff date where if you happened to become a computer programmer before AI, you were very lucky because you got a lot of the core fundamentals

1:21:26

Speaker 4: You had the ability to practice those core fundamentals, which are now useful to you in using agents well. And for those same people who are telling juniors, well, use agents AI to uh teach yourself how to code, use these agents. Um essentially there's a sense of are you maybe pulling the ladder up after you uh where you got the skills and now us we seem to be hitting a ceiling of skills without having access to good quality experience mentorship. So that was something interesting that came up. In terms of solutions, it was uh difficult to come up with um solutions that all of us could agree on. Most of them were just things we should begin to start thinking about. One thing was maybe we should ban learning.

1:22:13

Speaker 4: Sorry. Then AI tools during learning. Another one was maybe we should integrate AI tools only at a lower level. So instead of using AI agents, uh just asking questions to AI and making sure that they have access, juniors have access to AI for for asking questions and like learning from them rather than actually coding. That will come at a later level. But really it seems like the tension was not really about what to do as a solution, more about present versus future. There's a lot of um like maybe seniors need to start thinking beyond the present. It's very easy right now to focus on productivity. Uh coding feels easier now but more people

1:22:59

Speaker 4: need to be thinking about about what's gonna happen in the future. What is the future gonna look like, not just for the tools, but for the people using the tools.

1:23:13

Speaker 3: Thank you so much. I'm gonna hand over to Thibaut for a moment.

1:23:17

Speaker 2: Thank you everyone. I'll try and summarize my my feelings about this. Very briefly. It's quite clear we we have overwhelmed maintainers. Most of us are overwhelmed ourselves in our own jobs. We have a strained foundation. But more than anything, what strikes me is um the resilience of our community as a whole, how able we are to have open discussions. about this and the fact that our AI adoption isn't something we do top down like oh you use Django you should be using AI. It's something we're able to navigate together which I think is a is a big big strength and uh those discussions is not just the good good things it's also the negatives we're able to navigate. I really like this principle from our our open space format of having the right people in the room.

1:24:07

Speaker 2: I really think is what it comes down to with Django that we're able to navigate navigate this together. And um yeah really those negatives they are externalities lots of us are expecting to have to face head on. But we have lots of room it seems around here to experiment, make those things happen. So what I want to emphasize just to end is that those this conversation will keep keep going outside of this space. Some of you might be aware that the foundation we have an AI research working group that I'm a part of. I've been trying to start and struggling. We'll definitely get that going more. We also have the sprints to contribute to Django and the community together on Saturday. And I would love to unpack that. Like all of this more. We're gonna

1:24:54

Speaker 2: I think we've just lost uh The people. So yes, during the sprints we'll unpack all of this, we'll publish this, but I also want all of you to feel empowered to contribute to to improvements throughout those those themes.

1:25:12

Speaker 3: Thank you. All right, thank you everybody. In an ideal world, we would open it up to the room and have a nice long discussion. We're running out of time. So this as as Tebau said, the discussions will carry on outside the space. This is a launching point. Could everyone just pull out their phone for one final question? This will be really helpful for us, both in uh instigating your own thinking on this and us. Knowing what the community is thinking about. So the question is, what's a next step you'd like to see? We're framing this as experimentation as a lab. action, what's the next step you would like to see?

1:26:26

Speaker 2: shout out for people who have already finished answering the question the app you're using right now the web app is built with Django by LoRa it's one of the first times it's Tested live by so many people. If you have any feedback about the AWAP app, it should be very appreciative.

1:26:46

Speaker 3: I would. Thank you. All right, thank you everybody. Um I hope this uh felt like a generative um experience. We really wanted to know what every person. in the room is thinking and and give everybody an opportunity in some way to feed into the discussion. Um and uh yes the discussion's going to carry on, come up to either of us um again at the event. Tipo said the sprints would be a great place for some of these next steps. And this is a launching point, not an end point. So thank you so much.

Conference

DjangoCon Europe

April 15 - 17, 2026

Athens, Greece

Note: We understand that names change, people change, and bodies change. We respect each individual's journey and privacy. If you have any concerns about a video or need us to remove content, please don't hesitate to contact us. We will handle your request with care and promptly address any issues.

More videos by Laura Gates and Thibaud Colas

More videos from DjangoCon Europe