A special summer episode on the just-released 2026 Django Developers Survey, from Django 6.1, HTMX, and async to AI, deployment, testing, and Python tooling.
🔗 Links
🎥 YouTube
Will (00:00)
Hi, welcome to another episode of Django Chat, a special summer edition on the Django survey. I'm Will Vincent with Carlton Gibson. Hey Carlton.
Carlton (00:07)
Hello world, what are we doing in summer? That's crazy, we normally have it off.
Will (00:09)
I know,
I know. I was trying I'm trying to become European and it's not working. But we're doing it for good reason.
Carlton (00:14)
Okay, well.
Will (00:15)
So the Django survey it came out in May. The results are here and we're gonna discuss it. So links in the notes, but we're gonna talk about highlights because it's worth saying this is the only objective measure we have of the Django community. maybe you you're on the steering council, like speak to
Carlton (00:33)
Yeah.
Will (00:34)
how important is the survey?
Carlton (00:36)
Well, historically it's been very important. I mean, the Redis backend for the db for the caching, for instance, you know, in the survey, was like, you know, 70 % of people using Redis for caching and we haven't gotten a backend in Django itself. And it was this regular thing was, it Django Redis or is it Django Redis cache and which one was it? And you remember, then you start a new project two years later and you can't remember what you, what you
remembered last time. And so you have to go and do all this research about which one to use. Anyway, that came out. so we had a Google Summer Code project at the Reddish backend. And that came out bit directly. Why? Because, well, the survey said, you know, we're all using Reddish. And then it makes a really good argument for having it in Django. And that's just one example. There's numerous over the years.
Will (01:21)
Yeah, well in in companies
too. MongoDB had official backend and said directly, you know, we had Jib on the podcast, you know, it was due to seeing how well it ranked on the survey. So so yeah,
Carlton (01:34)
Right.
Will (01:34)
so we should say, so we know that there's there's millions of downloads of Django a day. There were thirty five hundred people who filled this out, which is a little less than historic, but not that much less. I think it was about forty six hundred last year. so small sample size, but it's global, like which was
Nice to see, right? There wasn't one big dominant country. I think if I look at the stats, India was 15%, the US 10%, Germany at 5%, and then a very long tail. So that was really good to see.
Carlton (02:01)
Yeah, but also Middle Eastern Africa came
popped up there over 10 % as well. So that was one
Will (02:06)
Yep.
Carlton (02:06)
of the sections higher than the US and I would have liked to the EU as one thing.
Will (02:12)
Well we can
we can update the survey next year. I say that because for better or for worse, I've had a heavy hand on it over the years. but yes, we'll fix that.
Carlton (02:23)
Yeah,
it's funny because the EU isn't really one country, but it kind of, you you think of it like who's from? No, no, alas.
Will (02:28)
I mean does brit does Britain does Britain count? You know, cause I get in trouble of saying DjangoCon
Europe versus DjangoCon EU. You know, Europeans are sensitive about these things. I just don't want to even say either word.
Carlton (02:39)
Yeah, no, I think call it jank on Europe because then we could go to Turkey or we'd go to Britain, we could go wherever. yeah, I don't know. Well, part of Turkey is part of Europe,
Will (02:45)
Is Turk Turkey's part of Europe? It's like part of all right. Okay, all right, let's let's focus. All right, so let's
dive in. So the last thing to say about who filled it out is 43% two years or less of coding, but 41% six plus years of coding. So I think that's
Carlton (03:02)
Yeah, okay.
Will (03:02)
fairly representative. You know, it's not the people that we see at Django Cons who tend to be more experienced, but it's a pretty good mix of people. So we're gonna use these to talk about.
Django
and is this perfect? No. Is it the best measurements we have? Yes. Because otherwise we're just looking at PyPI downloads, which who knows how accurate those are.
Carlton (03:22)
But also to lots of people who are a good proportion of people two years or less of coding. That means there's new people coming into Django, which is
Will (03:30)
Yes.
Carlton (03:31)
a nice thing to see as well.
Will (03:33)
Yes, and again, we don't currently track who visits Django Project.com in any sort of real way. There's efforts around that, but we're really flying blind in terms of where,
Carlton (03:44)
Yeah.
Will (03:45)
who, what. so this is it. This and hallway chats.
Carlton (03:49)
Yeah, I mean, is our only way.
Historically, there was a discussion about adding telemetry to January. Of course, nobody wants to do that. And then in this current climate, I don't think, you know, wouldn't go anywhere near that at the moment. know, there is no there is no way we other than reaching out and saying, come on, fill in the survey. What can we do? don't know. Anyway, let's
get this girl.
Will (04:09)
Yeah.
Let's dive in. Okay. And I should mention there'll be an official blog post I've I've written the link to as well with highlights, but we're just gonna go through so positives, versions. I'll let you start with you. So people report being overwhelmingly on the latest versions. It's 49% on six dot This was filled out, came out in May. So that's very promising. thirty percent on five point two and only a small percentage on four whatever. So that's fantastic.
Carlton (04:38)
Yes, because historically it wasn't the case. Back in the day it was, I'm still running 1.3, I'm still running 1.5, I'm still running 1.7. Okay, I think the success of the schedule release, the stable release schedule every eight months, and the stability guarantees and the deprecation policy, all of that has meant over the last few years it's become very easy to up gauge angle and we've started to bring more of the community with us. think
When we see these kind of numbers, think that's a confirmation that the effort that we put in is worthwhile. So that's super to see, I think.
Will (05:14)
Yeah. I
'cause it's a huge amount of work by the c fellows, by the community, and it's seems to be paying off, right? I mean, I can't remember the last really bad breaking change. I mean and if something is deprecated, it's a long, long, long, long time. You know, there shouldn't be any surprises at this point. So
Carlton (05:31)
Yeah, yeah, no. And
yeah, it's good. we are, yeah, Django is a stable project, but we still managed to make those changes. We still managed to bring in all the, you know, all the new features. We can talk about all the exciting things, you know, every reason.
Will (05:43)
Yeah, I think we're well,
we're gonna have podcasts on that. But I mean, like one thing to
Carlton (05:47)
Yeah.
Will (05:47)
mention is and again you can see in the Django News newsletter, there's double digit pull requests merged into core every single week. Like
Carlton (05:55)
Yeah.
Will (05:55)
it is on top of new features with every feature release, you know. We'll try not to get too much into that, but and async, we're gonna talk about that in a minute, you know. So it's it's
very,
Carlton (06:05)
Yeah.
Will (06:05)
very active despite being mature. So we gotta
Carlton (06:08)
Yeah, and you look at
6.0, is the latest release. Well, 6.1 has just popped out, but the latest release as the survey was taken, we've got Django tasks, we've got template partials mostly in the core. You know, these are big, these are big changes, right? So, you know, we, I think we balance it really well. Anyway. Yeah, yeah, I think we're great. Let's just pat ourselves on the back.
Will (06:22)
Yeah, so it's I think we're doing a great job, Carlton. You know, what do you think? All right, let's let's go down the list. So databases.
Postgres is still dominant, SQLite
Carlton (06:35)
Yeah. Yeah.
Will (06:36)
very popular. I d you know, of course SQLite is popular just because that's the default for local development and getting started. There is stuff. I mean, I did a a talk on deployment at EuroPython, and I'll be doing one at DjangoCon US and
There is a movement to deploy with SQLite. historically there's some issues around multi-connections, but I don't know if that's a thing other than or people are just like, I'm gonna use SQLite or Postgres unless have a reason to use something else.
Carlton (07:05)
Yeah, I mean, I still think choosing to deploy with SQLite is the spicy option, even though the patterns are more established
Will (07:12)
Yeah, yeah, yeah. It's it's come back.
Carlton (07:14)
now. I personally would, so I would use Postgres in almost every circumstance because even if you've got a small BPS, you can run a small Postgres there and it's fine. But yes, I see a lot of people who are excited by not running a separate server, not running a separate process,
Will (07:32)
Yeah, yeah.
Carlton (07:33)
not, you know.
There's a lot of juice in it. And the pattern's now well established. Adam Hill put it all up into a single package where you just pip install your, I can't remember the name of the package exactly, install the apps, and then it configures your SQLite exactly how it needs to be. And for a lot of deployments, that is perfectly fine. But there's still issues with short-lived readers, there's still issues with backups, there's still issues. It's, I still think
it's a spicy choice, personally.
Will (08:00)
Yeah. Yeah, yeah, yeah.
okay, backends, API,
Carlton (08:06)
But yes.
Will (08:07)
API. So this was interesting. There's well, let's say so. Django Rest framework is dominant, 73%. You know,
Carlton (08:13)
Yeah.
Will (08:14)
Django Ninja is there at 17%. It's the same patterns, forms, serializers for validation, you know, it's
Carlton (08:22)
Yeah.
Will (08:22)
it's still being used, it's still popular.
I know some people are working on next generation serializers and updates, maybe.
Carlton (08:29)
Yeah,
there might be something to come at some point in the future, but there's no rush. Like DRF's not going anywhere. It's nice and stable. It's nice and doing its thing. As you said, like 70 % using DRF still. that's phenomenal. you know, for the serializing, the Pydantic, how are you using? Are you using forms? Are you using serializers, validation? Are you using Pydantic, one option? Well, 20 % were using Pydantic. Well, that's almost the same as 17 % who are using Django Ninja. So there's like, you know, if you're using one, you're using the other, but most people just
What I thought was interesting in these figures, well, API first, well, yeah, obviously API first. Well, only about half of people were using Django as an API solution. And out of those half were using it as the backend for a single page application. The other half were API only. But then half of people are building server rendered apps, sending out templates.
Will (09:21)
Well, right. That's the thing. There's a a bifurcation and really it's swung back towards sprinkling in the interactivity with HTMX. jQuery is still there.
Carlton (09:30)
Yeah.
Will (09:30)
You know, React and HTMX are pretty much tied at this point. template partials, thank you, Carlton. You know, there's there's ways to,
Carlton (09:34)
Yeah. Yeah, well.
Will (09:38)
you know, and and that's what I tell people is that if you know you need to go full API, you know, you're a team, fine. But if you're starting out like you don't have to, like just baby iterate your way there because
it is a
Carlton (09:50)
Yeah.
Will (09:50)
big leap.
And if you're an individual or a small team, it's it's more of a maintenance burden than just having it all in the same place.
Carlton (09:59)
I think it's significantly more complicated to build an application with a React front end on top of an API than it is to build a server rendered template based application. So in this economy, you know, it's not zero interest rates anymore. We can't necessarily afford a team with, you know, two front end developers, two backend developers, you know.
and all the coordination tax between them. think, you know, going back to the old school ways of doing it, I think there's a lot of economic sense in that, which I think is what drives it. And HTMX is phenomenal. They've got new version four just being released
Will (10:38)
Mm-hmm.
Carlton (10:39)
as we speak. You know, I was amazed when I saw the numbers and I saw, you know, basically HTMX and React neck and neck. You know, I think React had a couple of points in front, but like...
They're third age.
Will (10:53)
And I think, you know, HTMX was sorry, I don't have the numbers qu I don't want to stare at them, like like five percent in twenty twenty one. So it's really just
Carlton (11:01)
Right.
Will (11:01)
zoomed up. And I I always felt there was an an education and a communication problem, but it seems like people are f people know about it now, you know, 'cause I remember even a couple of years ago, I mean, Carson Gross, the creator of HTMX, gave a talk at DjangoCon US recently, keynote.
And I was talking to someone after at Django Con who'd seen the keynote who said, I never heard of HTMX before his talk, you know? So you
Carlton (11:24)
Right, okay. Yeah.
Will (11:26)
can o you never can assume that people know about a thing. But anyways, yeah.
Carlton (11:30)
No, I mean,
can't, I mean, a lot, was one thing further, like much further down the survey, which I thought was lovely was that 25 % of respondents said, was like, how do you keep up with Django news? Do you read the blogs? You listen to the podcast, you get the newsletter, 25 % of people said, no, none of that. I don't keep up with it at all. there's,
Will (11:45)
Yeah.
Carlton (11:46)
there's a massive chunk of our user base that are just using Django and happily doing their thing. And they're not engaging with the news around what's new in Django
in any way whatsoever.
Will (11:55)
no. I mean
if I ever poke poke my head up and go to a a professional event which is rare, I I still think single digit percentage know what the software Django Software Foundation is, know about fellows, like the it's just like, well, Django just appears like Python, like water, you know. you know, so
Carlton (12:14)
Yeah.
Will (12:20)
But you know, so that's a good thing in the abstract, but it's an important to remember, like it can be insular, you know, but you know, we can always communicate better and you know, but yeah, people are busy.
Carlton (12:30)
Yeah, just on the on
the front and stuff. I wanted to just pick up the CSS frameworks.
Will (12:34)
Hmm.
Carlton (12:35)
There's a question. Bootstrap still number one after, you know, it's still there. Number one bootstrap,
Will (12:38)
Yeah, I mean tailwinds yeah.
Carlton (12:40)
then tailwind close behind and then planes, planes, plane CSS, just only two points back. Right. So then sort of bootstrap, tailwind or, you know, vanilla basically. And then there was others, the long tail of other options.
Will (12:53)
Yeah, there's I mean, Tailwind, you know, we recall Tailwind was this revelation and people really got excited. And I I still use it for a lot of projects, but I use Bootstrap too. I mean, it's it's how much CSS do you want to write? And
Carlton (13:06)
Yeah.
Will (13:06)
okay, now we I'll I'll say the word since we're a few minutes in, you know, AI. I think especially for web developers, AI and CSS, you know, it doesn't certainly doesn't replace a good front end engineer, but you can do more.
than you could before. And so I suspect the pure CSS is partly that because if you really dive into it, you want a framework for your CSS. Like it's it's its own, right? You sp you can spend a career on CSS alone. But
Carlton (13:35)
Yeah, if you're a jobbing jangan or then to use a framework to give you some structure around, you know, your style declarations is not.
Will (13:42)
And it's
easier for someone to come in. So yeah, so I th I think if anything, peop there's a not a pushback against Tailwind, 'cause Tailwind's still very popular, but it's a heavy it's a heavy framework. and
Carlton (13:53)
Well, you've got
to, the big annoyance of it is you still need a build tool, right? So, you know, when I'm using it,
Will (13:58)
Yeah, no, completely.
Carlton (13:59)
I run it through a compressor and compressor runs the tool, but I've still got to have that pre-processor in place.
Will (14:05)
Yeah.
Carlton (14:06)
Whereas if you're just using vanilla CSS, you just write the CSS and, or, you you can get a tool to help you write it if you don't know what's what.
Will (14:15)
Yep, agree. well, to finish up on front end, so Django templating language, still 80%. Ginja is in
Carlton (14:22)
Yes.
Will (14:22)
the teens. This is very stable. maybe it speaks to things like t
Carlton (14:26)
Yeah, year after year after year.
Will (14:28)
template partials, you know, the templating language is mature, but there are innovations.
Yeah, I don't know. I I guess maybe I thought there would be more action here, but it seems like it's the layer on top, it's the H T X that is enabling the stable base to be sufficient for people.
Carlton (14:46)
Well,
also, think there was like when Ginger first came out, it was a lot faster because a lot of the cost is in parsing the template.
Will (14:54)
Mm-hmm.
Carlton (14:55)
Yeah, it might be slightly faster to render, but not significantly. The way it's faster is with the parsing. But for many versions now, we've had the template cache in place.
Will (15:07)
Yes, that
that's a big one. Yep.
Carlton (15:08)
which
it gets rid of the passing step. you know, it's passed once the first time the template's loaded and then it's available for every other time. And so the speed difference is essentially negligible these days. And then it just comes down to this style question about, well, do you want to basically be able to write Python in your templates? And, you know, if you do, then Ging is more powerful.
Will (15:33)
Yeah. And I'm
Carlton (15:34)
and
some people like that, but obviously not enough to drive the community as a whole away from the DTL.
Will (15:40)
I I had a suspicion that you know fast API, which uses Jinja primarily, would lead people who come to Django to prefer the template that they're you know, templating engine that they're used to. But that's something I'm curious about. So if I I I suspect that might happen because look, I know we're Django pilled here, but like fast API is big, the community's growing. I hope there's more
Carlton (16:04)
Jesus.
Will (16:07)
crossovers because it's pretty different lanes at the moment.
Carlton (16:11)
I suspect that the choice of template language is kind of like the choice of cutlery. It doesn't really matter. You're just going to go with whatever's given to you by the framework.
Will (16:22)
We have to fight about tools. Yeah. well
to to to abstract from tools, let's talk about async. so I'll tee it up and then you
Carlton (16:29)
Yep, yep, yep.
Will (16:30)
can go. So this was really encouraging. So 35% said yes, they're using async. 39% said they're planning to, only 26% said no interest in async.
And then of those, fifty-one percent said they were using async views in some capacity, fifty percent channels.
Carlton (16:48)
Yep.
Will (16:48)
this is all heartening and you're doing a lot directly on this, so speak.
Carlton (16:52)
Yeah, yeah, yeah.
So I'm still on the cold face pushing forward to hear the async story as fast as we can. What's nice about 51 % using async views, what that means is that they're using the async story within Django itself rather than say channels. And channels is great, channels is brilliant, but the reason to keep using it really is web socket support because Django...
Will (17:17)
Mm-hmm.
Carlton (17:18)
Django's because of the nature of it, Django is not gonna probably have WebSocket support in core itself. So why would you keep with channels? Assuming everything is rosy in the land of Django, core's async story, why would you use channels where you'd use it for async? So 50 % using it and that tied in with the numbers from are you using WebSockets versus service and events? It's like, So if you're using more or less, if you're using WebSockets, you're using channels and that's why.
but that people are using asyncfuse, I think that's fantastic because for years we had, it's not quite ready. We've just recently updated the topic guide. Look, the API is complete. The remaining issues around async are async issues or Python issues or whatever. I recently chatted to Michael Kennedy on TalkPython about this, but it's...
It's there, it's ready. The other thing that's come along is free threading. We've just got the Django test suite passing with free threading. There's a new version of ASCII ref out which fixes a leak in local
that was there that was stopping that happen. So.
I honestly think over the next couple of years, we're going to see an awful lot of the bet. The long-term bets that Django has made over the last decade, essentially, they're going to come to fruition as free-threading rolls out. We're going to get rid of things like guild contention because that's where people see it is they, because we throw a lot of work into the thread, which you have to do. You're doing CPU bound work, but we throw work into threads. And, you know, if you do that at small scale, no problem. But eventually you hit guild contention and what you, what you need is the
the free threading build of Python to enable those threads to run separately and you get truly parallel work. And I think we're going to see a lot of performance wins for Django over the next, you know, 3.15 is coming, 3.16, 3.17, I think over the next two, three years, it's going to be a very happy time for us in Django land.
Will (19:11)
And I think part of it is I of course I think about the education aspect is just having blog posts, repos, examples of you're doing it this way, now you can do it that way. I mean,
Carlton (19:21)
Yeah.
Will (19:22)
we can I I I guess we don't have a note on it, but you know, that's one of the things that emerges also from the survey in the AI category is the docs are still number one, but AI is number two, followed by YouTube. So,
Carlton (19:33)
Yeah, right. Right.
Will (19:35)
you know, that's good if it helps people.
uncover things in the docs and be become educated that way. But you know, I want to be positive about it. But it's a shift. You know, Stack Overflow has fallen off a cliff. That makes sense.
Carlton (19:49)
Yeah.
Will (19:50)
but you know, as a creator, you know, look, Adam Johnson is still doing multiple posts a week. Like there are people putting out stuff. But, you know, for myself, there's less of a
If it's just gonna get scooped up and put into training data and there's no
Carlton (20:05)
Okay.
Will (20:05)
link, it's harder to justify the time to to create these things.
Carlton (20:09)
Yeah, no, absolutely. think, I think, you know, my, my take here is we have to wait and see how the, you know, the bubble plays out. think there's a, there's clearly there's new technology, clearly it's super, super capable, but it's clearly being subsidized by billions and billions of dollars, which in a non-sustainable way. So how does that play out? And so I think for individual creators or individual developers, I think we need to just keep our heads down and survive to see what happens once, once it's played out, because these tools are useful.
but they're not, if the ability to subsidize them to the extent to the tune of billions of dollars a year disappears, well, how freely available are they going to be? So I mean, one of the topics is, you know, was AI usage. And so AI
Will (20:54)
Yep.
Carlton (20:54)
is clearly a thing. 50 % of people are using it.
Will (20:59)
Well, sorry,
eighty five percent are using it daily or weekly. or fifty eight fift fift fift
Carlton (21:05)
No, it was 50%.
Will (21:06)
fifty-eight percent daily, twenty-seven percent several times a week. So that's eighty-five percent using it at least once a week, and only ten percent said none. Yeah.
Carlton (21:17)
Sure, yeah, I mean, so, but what does that using mean? So 50%, 30 % said
they're using chat, 30 % said they're using code generation. So it's a bit like, you what does it mean? Did somebody ask chat GPT for, how do I do this in Django? That counts as using it once a week, right? Now the interesting...
Will (21:36)
No, no. And this is when we
for the survey next year, we need we want to tease this out. You know, how what is you know, what does using AI mean? Are you, you know, we saw like one of the big things is CLI usage is is pretty high. now is
Carlton (21:50)
Yeah.
Will (21:50)
that because everyone's running agents in the terminal? you know, I s you know, so we want to tease out what models are using, what agents are you using, and then I guess, you know, how much are you using it, and then what are you asking it to do, right? Because it was something like twenty-seven percent said they're doing.
you know, let it run mul multi-files, go crazy.
Carlton (22:09)
Yeah.
Will (22:10)
But most, the majority are still saying nope, or they're saying that like have it run, manually review it. You know, so more of like a managed AI usage than, you know,
Carlton (22:19)
Yeah.
Will (22:19)
just hordes of agents replacing what you're doing. Though some people are doing that, you know, friends we know are doing that.
Carlton (22:23)
And it's, and what's not, yeah,
no, no, no. Yeah. mean, look, there's a whole, there's a whole spectrum here, right? But what's not clear from the figures. So everybody's like, not everybody, but a number of people are clearly engaged in it. Everybody's heard of it, but to what, what plans do they have? Like what access do they have? Have they got, you know, a plan, a paid plan such that they can do significant amounts of agent work? Or is it, you know, actually I've only got a free quota where I can chat to it a bit and then work.
that detail isn't brought out. And my sort of point to go back to your point about creating, what I'd like to see, what I'd like us to get to is the point where the economic bubble aspects that we're currently in has floated away and the long-term, look, you will expect to pay this for that has become kind of clear. And I think at that point, we'll be able to judge much more effectively.
what usage we're going to start to see because, you if it's thousands and thousands and thousands a month for, you know, an agent, I don't know that many people are going to be able to afford that.
Will (23:29)
Yeah, and there's that pricing stuff hasn't come yet. I mean, one thing I noticed being
Carlton (23:34)
Yeah.
Will (23:34)
at the big Python conferences is last year it was are you using AI, what models you're using. This year everyone's using it. It's how do you make the most of it? So agents,
Carlton (23:43)
Yeah.
Will (23:44)
harnesses, but also costs, you know, c and yeah, that hasn't fully played out. I mean, I would love to see more content from the community. You know, there's people like Peter Granstaff is running
He's got a very powerful local computer. He's running local models. you can use Olam or LM Studio to use newer models like Kimi and the cloud. I hope the community can be can talk more about what what we're using. because yeah, like are you using it? Yeah. What does that mean? We still haven't teased that out. I mean, I'm personally and my colleague Paul Everett is very excited about what Apple's doing with MLX, which is
modified memory and the M5 chips and M7 chips. basically because if you're you doing agentic flows, the constraint is not compute, it's memory. You have to store everything in there. And that is very expensive. And so if you can do it on your laptop, you know, with 128 gigabytes of RAM or these things coming down the line, then you can not just own your data, but you can run a you know open weight Chinese probably model and you know just have it running for forever.
because I guess the interesting thing is the models they're still in sorry,
Carlton (24:51)
But I...
Will (24:52)
let me finish this thought and then you go. The models are still improving a little bit, but what's really improved, the reason they've gotten better is these agentic flows where they keep trying, you know, already with ChatGBT, it's like how many times do you want me to try? How many tokens do you want to burn? And it's like, yeah, it's better, but like at what cost? Okay, I'm done. Go.
Carlton (25:08)
Yeah, but no, okay. But like you
just said, you just said words like M5, M7, 128 gigs of RAM, no, exactly. This is no
Will (25:14)
Everyone can afford these, right? Everyone can afford five thousand dollar laptops.
Carlton (25:18)
more affordable for the majority of people than, you know, an API costs thousands of dollars a month token subscription, Or token spent.
Will (25:27)
Well, yeah, and we
all right, so as as a last bit, yeah, like Anthropic's gonna release its figures. it everyone I know who says they know something says that Anthropic is minting money on inference, right? 'Cause the question is always, they're subsidizing it. Well, how much is being subsidized? But you know, right now it's a better deal in general to like use an API than to run it all yourself. But that surely has to switch because you you it's unsustainable to
Carlton (25:55)
Yeah.
Will (25:55)
To spend thousands of dollars a month and
Carlton (25:58)
I mean, there's a nice George Orwell quote
Will (26:01)
Ha ha ha.
Carlton (26:02)
about the newspapers in Spain. It's like in most countries, you'll see the different takes on the same news. The left paper will say one thing, the right paper will say another. But in Spain, it's like two different worlds. They'll just say totally different things.
You you've just said that people say that they're making money on imprints, but you read some posts like that, but you also read other posts where they're just hemorrhaging billions of dollars a time, you know, a quarter. So what to
Will (26:28)
No, yeah, there's there's well, they're gonna have to
Carlton (26:31)
believe?
Will (26:33)
I know. Well, all right. All right, let's switch
Carlton (26:34)
Anyway, anyway, one thing from all of this tooling
section, I want to just go is that 75 % of people were still using IDEs, VS Code 47%, PyCharm 26%. That's three quarters still using the IDE. So it's not that we've all abandoned our editors and jumped into the terminal and just driving.
Will (26:53)
Yeah. Well, it's all up for grabs. I mean, it's it's the you know, and I can speak personally, you know, at JetBrains on PyCharm, we're rolling out constant features and try to think, you know, what is the what does a modern IDE look like? You know, it is interesting that you know, these AI first IDEs, so cursor's still around that they're being acquired, Windsurf got consumed. You know, it's really just unc and now there's these agents, so you can pop in whatever model, you know, Hermes is written in Python, it will generate skills for you in the back end. Now skills are
sexy and interesting. They're just markdown files. so all right, let's not get too far afield, but we need to tease this out. And I think, you know, I guess the last thing I would say is web flows do work very well with AI. So like a lot of my colleagues are data science, Jupyter, you can still use it. But just with the cells and everything, it's not quite as smooth. But for web flows, you know, you can ask it to build you an Instagram clone and it'll give you something. Is it good?
not without a lot of guidance and knowledge. You know, you still I s I think, you know, if you could build it yourself, you can build manually, you can build it faster with AI, but you just gotta be know how to do it manually. and otherwise
Carlton (28:02)
Missed.
Will (28:02)
you'll just get so far afield quickly.
Carlton (28:05)
You still got to maintain the thing, It's like, that's, that's next quarter. Okay. Or leave. Yeah, I'd leave. I've moved on to the next country. Anyway, let's get back to the thing, to the survey. There was some, okay, go on.
Will (28:08)
Nah, next quarter. Just get promoted and someone else someone else. all right, deployment. Deployment. Deployment. Let's talk about that.
All right, so just plus a challenge,
you know, fifty percent monolith. Yeah.
Carlton (28:22)
Yeah, well, my take on this, interesting, yeah,
was most people still using a monolith, still on the BPS, still using NGX, still, know, still that as your front end for static files, you know, as bog standard, as bog standard could be. And then for the database, there was a nice split, 33 % self-managing their database and 30 % managed. So, you know, there's this nice divide between like, yeah, do you use an RDS or a CrunchyData or a host, you know, managed database or are you running it yourself? I thought that was lovely.
That whole section.
Will (28:51)
Yeah. Well and I and
I've realized too, like the word managed carries a lot of water in the database section because I was talking with Vercell and Render and all these services, you know, not to not to name a specific company, but you can s call something managed and you just spin up a Docker container and do backups every once in a while. It's managed. You
Carlton (29:10)
Yeah. Yeah.
Will (29:12)
know, the marketing team's happy, put it on the homepage. But if you're running a big important business and you have compliance and all the rest, you know.
that type of managed is a whole separate thing. And so
Carlton (29:24)
So.
Will (29:25)
yeah.
Carlton (29:25)
Let me give you
an example. over the summer, whatever, well, I've done a bit of maintenance and I did two big RDS updates in a day. And one was for a very serious project that has to stay up. And I did their own, lovely green blue thing. And it's just delightful. It spins up the blue. It sets up the logical replication, does all this. And then you could check it all again through your application and just go, yes, switch, switches it fine. Make sure everything's okay. Then you delete the old one.
Wonderful, took me best part of three, four hours to actually go through the whole process and get it done. And then the second one, it doesn't matter if it has two minutes of downtime. So no, let's just upgrade the instance in place. And both were totally appropriate for the application that had like two minutes of downtime on the other one. But the first service couldn't take that two minutes of downtime. So the
Will (30:17)
Yeah, yep, yep.
Carlton (30:19)
difference between something which took me basically less than an hour and took me all morning.
is that difference in caliber. And what's nice about RDS, it's a very serious service. has all the options in the world. it's, can anywhere on that spectrum, you can be, that's what, you know, those are the kinds of check boxes I'm looking for in a managed service. Yeah. If it's, if you're just spinning me up a Docker instance, well, I can do that myself.
Will (30:43)
Yeah. And yeah, I I just think if it's anything serious, I want someone else to manage it in some capacity,
Carlton (30:49)
Yeah.
Will (30:50)
unless it's a toy project. people people
Carlton (30:52)
Yeah.
Will (30:52)
disagree with that, but I just like what's the cost of messing it up? is what I think. But anyways, but it's but
Carlton (30:59)
It's the snapshots,
it's the backups, it's the restores, it's the, you know what, this went down and there it was an hour ago and now I can put it back. That is, you you've got to be a very serious Postgres admin to be able to do point in time replication yourself, right? Restoration yourself.
Will (31:15)
Well, not to get
too gossipy, but Vercel had a booth at EuroPython. And you know, they're they're sponsoring EuroPython. They're, you they're they're killing it. And they've historically been known as front end next.js deployment. So you you'd have a
Carlton (31:27)
Yeah.
Will (31:27)
Django backend hosted on let's say Heroku, though you wouldn't use them anymore, and then front end there. Vercell now has the option of full stack. So just put it all in there and you can do all the branching and all the stuff that they're happy about. The one thing they don't have is a managed database. And I sort of pushed them on that. I was like, well.
It would be nice if you had, you know, render has this, Railway has this, Heroku has this. You we that's how we got into this discussion of what does manage mean. A lot of the team, especially the team that was there at the conference, came from EdgeDB, which and moved over to Vercell. And so they were really doing this. So they knew,
Carlton (31:59)
Right.
Will (32:00)
no, we have no interest in like getting into, you know, well, data I
Carlton (32:02)
Yeah, yeah, yeah, yeah, they've lived that pain. They've lived that pain.
Will (32:06)
mean, databases, you know, you give a nod of respect if someone's like doing databases. like just scary.
Carlton (32:10)
Yeah, no, absolutely. It's serious stuff.
Will (32:13)
Exactly.
All right. Okay. what one other thing that came out is you know, we asked the question of the three most important bits that you like about Django. And it's
Carlton (32:22)
Yeah.
Will (32:23)
the heavy hitters, models, which is sort of our ORM by proxy, admin auth. There's a long tail, but I mean, you
Carlton (32:28)
Yeah. And all the rest, yeah.
Will (32:31)
know, I think most people if push to shove would say the ORM or the admin. That has to be in the top three, five for sure.
Carlton (32:39)
Yeah, yeah, yeah.
Will (32:42)
You know, it's still, you know, fast API doesn't have an ORM. It doesn't have admin, you know, like we can attest. It's not a trivial thing to just spin up.
Carlton (32:53)
Yeah.
Will (32:53)
so so that's good. Stability. starting projects, people are still this is actually interesting to me. so seventy-five percent said they use the start project command. I kind of thought more people would maybe just go let AI take the wheel, but it
you know, based on the responses, people are starting manually and then adding AI in afterwards. They're not just Greenfield, you know, command line, build me this app.
Carlton (33:20)
But here's my question, right? So why would you ever have a random token generator do something which you already have a deterministic tool for? Right? If you haven't got a deterministic tool to build something, then asking the clanker to come up with something, that's brilliant. But if you've got a deterministic tool, what benefit at all is there in having the LLM do it?
Will (33:48)
Yeah. Well, here we get to, know, there's things like, you know, with tooling, like rough is super super popular at forty three percent. So you have a deterministic
Carlton (33:54)
Yeah.
Will (33:55)
tool that will walk right through, or you can toss it into your LLM. You know, like I do think there's sort of this pushback of like, wait, like we can already do these things. We can already check types, we can already check linting, you know, we don't need to not just pay the cost, wait for it, but it's non deterministic, you know, it's like it's really,
Carlton (34:13)
Yeah.
Will (34:14)
really good, but it's not a hundred percent, you know.
Carlton (34:17)
But I mean, surely you have the agent use these tools. You say, look, when creating a project, you start project, use this template. And when linting, use rough or use. So what I thought was interesting with the rough there was rough was at 43 % versus 17 for flake 8 now. So,
Will (34:35)
Yeah.
Carlton (34:35)
you know, I think the ecosystem is moving across.
Will (34:38)
Yeah, I mean look, ru I use Rough. Rough is great. It's not just the speed. They they're incorporating things. You know, black is, you know, rough sort of black is still has usage, but like Rough is kind of one tool to do it all. you know, so Ven versus UV, your favorite topic. You know, Venv is still for managing virtual environments, sixty-three percent, UV is forty-three percent. So UV is definitely growing.
We do have you know, we will see they they you know, they're still that team is still putting out updates. They're now working at open API, so you would
Carlton (35:14)
Yeah.
Will (35:15)
assume that's not gonna be their main focus. but I don't, you know, I don't wanna detract from the amazing work that they did for the community. And, you know, but I don't think we don't wanna give up everything in favor of these Rust built tools. I know how you yeah.
Carlton (35:30)
Well, I'm, you know, just as a business risk perspective,
I'm not going to put my faith in a VC backed tool because it's guaranteed to go, you it is, it's
Will (35:40)
It's not VC well, I guess they're one V C back to another.
Carlton (35:43)
always been bought out now. But the point being that as nice as the individuals in the project are, know, Charlie Marshall, heard is a very nice person, know him, Carl Mayer, he's a, you know, former Django alumni,
Will (35:52)
He is. He is. Yep.
Carlton (35:54)
you know, these are lovely, wonderful people. But
the dull hand of economic determinism owns their project. And so just from a personal perspective or a business risk perspective, I'm not gonna tie myself to their mast. No, you're absolutely right,
Will (36:10)
Yeah. It's not the hardest thing to swap out though. You know, I think that's part of the th like
Carlton (36:15)
but it's not also not the hardest, it's not the, what does it bring? So VM and PIP, perfect.
use the community tools, the stable, long lasting, open option.
Will (36:33)
I do think not to you know not not to just be like a fanboy for UV, I do think the pi you know, it does more than than just virtual environments. You know, the Python package management, okay, there's a deep discussion around how they do that, but it does make it easier, especially for newcomers. That is a real advancement. but all right, let's not get you know.
Carlton (36:52)
No, so
the one thing about UV that I like is the Python installation management. So I actually created a PSA in CLI tool that I used that does the same thing but in Python.
Will (37:06)
I mean it's brilliant and
it's faster because they have their own versions and you know, Pi it it's it's definitely yeah. All right, let's let's let's blast through. So caching, Redis is still king, fifty three percent, no surprises
Carlton (37:19)
Yeah.
Will (37:19)
there. It will be you know, we had you know, Django tasks, like maybe there's some you know, Redis is this yeah, I don't know. What do you I'm
Carlton (37:29)
I think, yeah,
I think the only reason not to use Redis is if you don't need it and you can cache in the ORM and likely that's fast enough. So if you say you're running, you've got Postgres, if caching in Postgres is good enough for you, just use that and that's one less thing in your stack to be running. Cause you know, as soon as you spin up Redis, you've got one more service, one more thing, one more chunk of RAM being used, et cetera. If you...
If you exhaust caching in the DB and you're like, do know what, I need something faster. Okay. Redis is there for you. And, you know, it seems like the, option memcached it doesn't look like it's going to end up being compatible with, free threaded pythons. So it's a little bit, it's like, okay. Is that, is that ever going to catch up there? So I think Redis is probably the, the, the go-to in, in memory choice. there is another one. what was it called?
sop, no. I'll have to look it up. But it was another cache that was claimed to be faster and used in process and whatnot. but you know, it didn't show up in the survey results. Nobody's using it.
Will (38:43)
Well maybe it w maybe it maybe
it wasn't was it I don't know, we should check if it was asked. All
right, moving on. types. So this was kind of interesting. So big percentage of people claim they're using them. no clear tool. There's so all these type checkers. This is something the Python team is working a lot on. You know, what's our default? We want to support all of them. But to your point, you said before we started recording, it seems like it's basically whatever's built into the IDE, you know, like.
Carlton (39:09)
Yeah, well that was one
of the questions. One of the answers was the top answer of what type checker you're using is the one the IDE or the tool provides. it's like,
Will (39:16)
Yeah.
Carlton (39:17)
know, whatever it is that VS code and PyCharm turn
Will (39:20)
Ha ha.
Carlton (39:21)
on, that's what people are using for their type check, which is fine.
Will (39:24)
Yeah, I yes. For for the end user, it doesn't really matter. testing. A lot of people said they were testing, which is great. I know you have a point to make. So fifty one percent said GitHub actions. So so it's easier
Carlton (39:37)
Yeah, for CI and CD.
Will (39:40)
to do CI and C D than before. but you have concerns.
Carlton (39:46)
Well, it's not concerns. It's like, I think the one thing now that is stopping a mass flight from GitHub is the GitHub actions. I think they're the only people offering free, essentially free CI that kind of works. And I think as soon as...
a solution appears as soon as it's like, we could do our CI that way. I think we will see a big tidal wave out because of the degradation of user experience within GitHub. Yeah, but people can, you can spin
Will (40:16)
But where, right? I mean, code Codeberg is making a big stand on non-AI, you know, either is get GitLab.
Carlton (40:23)
up a 4g0. can, like, it's not actually hard to host your own GitHub, bare repos and whatnot. It's...
Will (40:31)
Yeah. But it's hard
and it's hard enough, right? I mean, that's the point. As long as it's
Carlton (40:35)
Yeah.
Will (40:35)
free and it works pretty well, it's just like, do I wanna have to fiddle with something else? Like, no.
Carlton (40:42)
No, entirely, entirely. like, I know, I don't know if they still have it, but JetBrains did have a, you know, a Git hosting,
Will (40:49)
Yeah, team city.
Carlton (40:50)
yeah, Team City solution. I think there are plenty of them out there.
And but it's what's for me what keeps the critical mass? What's the inertia that stops people moving? just we'll move the company over to this hosted version. We'll host our own one in the office for that. You know, what stops that being the solution? I think GitHub actions is the answer. That's just a sort of conjecture about the state of the ecosystem.
Will (41:16)
I mean, I think it's a brilliant
business move. I don't know that you know, the costs
Carlton (41:19)
Yeah.
Will (41:20)
and everything else, but look if something free and works, you know.
Carlton (41:28)
Yeah, no, it's great. We've had it for years. It's just interesting that it's the big one and we'll see. Maybe I'm wrong. Maybe competitors appear and people stick with GitHub, but I can quite easily see there being a kind of cascade effect
Will (41:43)
Mm.
Carlton (41:43)
as soon as that one linchpin breaks that people go, I do know what, let's jump
Will (41:47)
Yeah, well there needs to be a b
Carlton (41:49)
ship.
Will (41:50)
there needs to be a business model, you know, related to it, right? 'Cause no one's gonna do a startup to compete you know, like w who's gonna Yeah. Anyways, but yes, no, it's interesting.
Carlton (41:56)
Yeah, I mean the
reason why GitHub got bought out, right, was because they couldn't keep going. That's why they needed somebody like Microsoft to come and take them under their wing and say, yeah, we'll keep funding you forever, basically.
Will (42:10)
Well, and you know, this is this is public, you know, with the amount of repos generated by AI, you know, the just the costs to to Microsoft of running GitHub are exploding. You know, you could also say their training data for models is growing, but yeah,
Carlton (42:25)
Yeah.
Will (42:26)
I don't know. It's a f hard hard to say. all right, so let's let's wrap this up. What I I I'll give my quick take and then you can give the deep insights.
I
think, you know, Jable Django remains a stable core. There's a lot of movement around tooling, both in the p you know Python world and AI. But Django, I think, is well placed and people seem to be, you know, they're using the latest versions, they're using mostly the things I think you and I would recommend to people. and the last thing is we have as ever this communication problem. You know, what percentage of people know about the new features
Carlton (43:01)
Yeah, yeah.
Will (43:02)
repo? You know, how do people find their things?
You know, you cited 25% said nothing. you know.
Carlton (43:10)
Yeah. Yeah.
Will (43:11)
I do want to shout out the Django board is doing a lot of work to post more regularly on the the blog. And you know, there were community booths at PyCon US and EuroPython staffed by lovely Django members, you know, doing it for free. I guess, you know, Jeff Triplett, our president, has mentioned they're they're doing a search for an executive director. He's managed to find.
Carlton (43:34)
Yeah.
Will (43:35)
funding for that, but it's a little bit of you know, we can complain about it, but in terms of what's gonna happen, we're we're waiting on more people power.
Carlton (43:45)
Yeah. No, I think we do well. Well, I think...
Will (43:46)
All right, sorry. What what do you think? Well no, we we we do great. I mean,
look, we do great. Like I if we saw that people were like on old versions and just totally checked out, what we see is people are like, Yeah, Django's a core tool and I fiddle around the edges, but if you know what it does for you, you're like, Why would I use something else?
Carlton (44:02)
Yeah, and I, you know, when I spoke to Michael Candia, you know, a while ago, mean, we the COVID pandemic came up and, know, it was something that struck me that
It really did hit the community hard. And for a few years after this, it's taken us a long time to recover. But I look around at the community now, think, yeah, it really is going strongly. It's in a good place. And that's lovely to see. And I think technically, we're in a wonderful position. I'm really, really excited about free threading coming through. I just think that's going to just pay div.
Will (44:37)
So maybe Django
and the Med, there'll be some discussions around
Carlton (44:41)
Well,
Django in the Med, I've got a secret project to work on async cursors for the ORM. So we'll see if we get that. But yeah, there's also a list of projects that help boost the free threading readiness of Django. There are a few cases where we do things at startup to populate registries and things like that. Well, those need to be done.
before you enable multiple threads, right? Otherwise there
Will (45:07)
Mm.
Carlton (45:08)
could be race conditions and whatnot. there are the tests we passes with free threading, but there may be cases where there are races that we haven't eliminated yet. So we can't really say, yeah, just go fire it off. But over the next few releases, we should see those issues ironed out. then I think Django is in a really exciting place. We talked about serialization and the API story that's coming forward. I think we just need to see how these AI tools
I think we're still in the early days and I really want us to be in a position where the economics side of it is settled so we can see actually what's the realistic cost that people would expect to face because then I think we could answer questions about, well, okay, how much are we going to use these tools? If they are more expensive than a junior developer, well, I'd just rather have a junior developer but if they're obviously much cheaper, then I would rather not.
Will (46:02)
Yeah,
and I I guess the last thing I would say is the the plight of the junior developer is real. it is very hard to get hired as a junior. I've had many, many people at consultancies at big companies just say, We have no juniors and everyone can say this is a
Carlton (46:17)
Yeah, at the moment.
Will (46:18)
problem coming down the line, but you know, if if you're a manager and you're used to just dealing with pull requests that you def for defined scopes, well, AI can do a version of it and it's
cheaper than a junior dev who will get trained up and leave in two years. I mean, to give one take on it. So Cur yeah, no, currently. Currently, yeah.
Carlton (46:34)
Yeah, yeah, I mean, currently, but
until we see the subsidies around pricing going away. if you get a subscription and you go to usage, it's like, you've used eight times your subscription value in the last hour. It's like, hang on.
Will (46:52)
Well, and I I will say something negative about Cursor. They just hid ha like the pricing of tokens so you can't see the spend anymore because
Carlton (47:01)
Right.
Will (47:02)
you know, it's it's a lever you can pull and
Carlton (47:07)
Yeah, but the example that we saw came through recently was GitHub Copilot. Microsoft got so many things called Copilot, you don't know what's what. But
Will (47:16)
Everything everything AI is co pilot for them, yeah.
Carlton (47:19)
they went from a fixed cost to token billing and people were getting thousands and thousands of dollar bills at the end of that month and were shocked because, we built these workflows around using Copilot at a fixed cost and now suddenly it's a token price.
Will (47:36)
Yeah, well,
I gotta put on my company hat. You know, JetBrains made that move in the fall because they saw that coming and said, look, we're we as a company can't subsidize this. It's going that way. There was a little pushback,
Carlton (47:45)
Right, okay.
Will (47:46)
but people people got on board. I mean, look, the challenge is
Carlton (47:50)
soon.
Will (47:50)
also jet, you know, anyone building around these models, they change too fast. So so
Carlton (47:55)
Yeah.
Will (47:55)
JetBrains is doing things about like we have all this great static analysis that we do. Can we separate the intelligence of the IDEs into a headless
you know, something headless and then feed that into the models. And can we be more efficient and better? And yes, we can, but we'll do it for one model and then the next model three months later completely changes that. So it's it's just
Carlton (48:15)
Yeah. Okay.
Will (48:17)
building on quicksand. And yeah, until it stabilizes out, it's very hard to
Carlton (48:23)
okay. So let me bring this back to Django, right? We've got this idea about, we've
Will (48:25)
Okay.
Carlton (48:27)
got this idea about using boring tech and what's one of the notions in that essay is about innovation tokens.
Well, what is
Will (48:34)
Mm.
Carlton (48:34)
an innovation token? You spend it on an unknown technology. Well, back in the day, it would have been using Mongo or using CouchDB instead of a relational database. Well, these days it might be, well, building with all this new AI tooling that we don't know what's going. Well, there's your innovation token being spent because those things are changing all the time. what's your solid known foundation around which you can experiment with these things when it's jank?
Will (49:00)
Yeah, it's Python. You know, look, Python
Carlton (49:02)
Yeah.
Will (49:04)
For all the internal stuff, Python is growing and relatively stable and in a great place compared to other languages.
Carlton (49:14)
Anyway.
Will (49:15)
Okay,
good. All right. We we went on. do check out the links. Links, the survey will be live, the blog post will be live. give us feedback. We're gonna do the survey again next year. And hopefully not ask all AI questions, but ask better questions to un to tease out what is actual usage in the community and what can the Django do as a result.
Carlton (49:36)
It would be nice to dig in,
dig in and find out what's actually going on there. That would be nice to find out. Anyway.
Will (49:44)
All right. Well thanks everyone. So DjangoChat.com. We will be back in the fall, September, with our regularly scheduled programming. But this was a special episode. thanks for listening. We'll see you next time. Bye bye.
Carlton (49:55)
Bye bye.