Pragmatic AI: How to gain trust with user-centered AI adoption with Meagen Voss

This video features Meagen Voss at DjangoCon US 2026 in Chicago, Illinois, USA.

Pragmatic AI: How to gain trust with user-centered AI adoption with Meagen Voss
0:27:23
Published October 26, 2026
88 views

With some CEOs of large AI companies claiming AI cures for cancer are just around the corner, you might feel like you’re a fool for not integrating AI right away in your software. This talk will show you why resisting that pressure can be a good thing and how taking a steady and thoughtful approach to AI adoption in your open source project can help you stand out.

At the Wagtail CMS project, we recognized the transformative potential of AI tools but we also observed that rushed AI adoption has led to poor implementations, costly mistakes, and user backlash. Not to mention, many people understandably have environmental and ethical concerns about these technologies.

I’ll show you:

The importance of creating guiding principles for AI adoption
Why we decided against incorporating AI in our core product
How we designed our Wagtail AI package to give our users agency and choice

I’ll demonstrate how our approach has built loyalty and goodwill among existing users as well as how it has attracted interest from users who prefer a more practical, evidence-based approach to AI adoption.

This talk was presented by Meagen Voss at DjangoCon US 2026 in Chicago, Illinois, USA
https://2026.djangocon.us/talks/pragmatic-ai-how-to-gain-trust-with-user-centered-ai-adoption/

LINKS:
Follow Meagen Voss 👇
On Mastodon: https://fosstodon.org/@vossisboss
Website: https://wagtail.org

Transcript

4,459 words · auto-generated · en Show

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

0:14

Speaker 1: Hello everybody. Uh fun fact, Megan is an Irish name. It means I burn in the shade. So, y'all, I'm here today to talk to you about uh pragmatic AI. Um, if you want some more details about me, here's all my info. I'm a member of the Wagtail Core team and I've been a member for a few years. I work for an agency in the UK called Torchbox. And I'm going to be talking to you a lot about the Wagtail CMS open source project, which we run today. And so first I want you to meet Wagtail because not everybody who's used Django has used Wagtail. Wagtail is built on top of Django. It's a content management system that balances kind of this nice

1:01

Speaker 1: smooth user experience with blocks and uh different like it provides a natural writing experience for content creators. And another way to put it for Django people usually is why maintain your own Django admin when you can have something nice like Wagtail And you know, it provides a lot of those features that uh the people who manage the content kind of expect and need, but also has all of the power of scalability with Django. And also you can combine Wagtail with pretty much anything you can do in Django, you can also uh pair up with Wagtail. And some of Wagtail's bigger users include Google, the National Health Service, NASA. I want to give a shout out to Linkin

1:47

Speaker 1: Loop in the room because they recently redid Firefox. com with Wagtail. So that is a big coup for us. But we have a lot of other um academic users. The Penn School, uh Wharton School business has been a longtime fan, and we serve some pretty big folks in the world. And I think in any story about AI, it's important to look at the humans behind the software. And so this is the Wagtail team. At Torchbox, we have six people. uh working on Wagtail who kind of make up the core product team. That includes uh myself, uh Matthew, who's kind of our lead soft engineer software engineer. Sage, who is another one of our engineers, and then Tebow,

2:34

Speaker 1: who's our engineer and also kind of more of our product lead, he's stepping into a big strategy role as well. And then Ian is the one who worries about the money all the time and keeps us in line. And then Abigail also helps us keep everything on track with project management. But it's not just us at Torchbox. We also have a large group outside of Torchbox. Many of the members are here today called the Wagtail Core Team because it's important for us. To make sure that this project just doesn't become focused on Torchbox and Torchbox's clients. Like I said, we serve a lot of people, and it's important to make sure that we keep that focus. But the reason I bring up like kind of the scale of that as well, the scale of our users, is because, you know, when 2022 came around

3:20

Speaker 1: uh and chat gtp happened, we already had plenty to do as a six-person team. And then all of a sudden uh AI came And uh we had to and there, you know, pretty much uh that that day changed everything and we saw the rise of this huge hype wave. And I'm not going to read all the different things that people have claimed about AI. Some of them are based in truth, some of them are more science fiction than science reality. Um it, you know, it what what it did do ultimately with all these different things is it made it really hard to distinguish the noise from the actual utility of AI.

4:05

Speaker 1: And for a small team like early on in an open source project, when we're already being careful about what new features we put into our project, Uh it you know, we we were kind of like we're not going to like we we need a better idea of what the actual benefits of AI are. and a better understanding of it before we dive in. And you know, it was clear to uh like and um With all these different claims that were being made, like there was a similar behavior, like similar to the blockchain hype, there is always, I'm sure what Everybody has at least one of these people in their life. This meme is a little uh hard to read, so I'm gonna go ahead and read it for you. This is the big forehead fish staring at a diver

4:50

Speaker 1: and it says Aren't you using AI? Why is it still taking you so much time? AI is magic. Are you a muggle? New best AI tool launched just today. And then the other person is just trying to get their work done And that's kind of what it felt being in the the middle of all that. And you know, our users are asking us questions, being like, and then also our salesperson , always the salespeople, being like, when are you gonna have something to do with AI Uh and so we had to honestly th slow our roll a bit and think a bit about uh what we wanted to do because There were a lot of potential impact and risks. And at the time we really didn't know a whole lot about these impacts and risks. Like it was clear there was going to be an environmental impact.

5:38

Speaker 1: We're only now just getting to understand some of that Um there are obviously user um user safety and labor rights issues that could potentially be in play. And particularly for the clients that our agencies serve and other agencies serve, the the ones that kind of made us be like, okay, we need to really think about this deeply, are around kind of these legal and regulatory gray zones about data and uh, you know, code ownership and things like that. as well as costs and RI, and then also uh proven this your digital sovereignty, particularly for we serve a lot of people in Europe. Uh, you know, there's lots of concerns about like, well, can somebody just take this model away tomorrow? And lo and behold, our president here showed them that concern was valid.

6:23

Speaker 1: Um and so the reality right now is totally catching up with the hype. These are not even like the freshest headlines that can be pulled off. Like I didn't even Uh I've given this talk enough times now that I can't keep up with the amount of headlines that keep getting thrown in there. Uh but the other one that's being reckoned with right now is that, you know, a lot of people are finding that this technology is super expensive And a lot of these initial pilots are actually not returning a whole lot for the people who are trying to use this technology I mean that like the frontier models uh finding out just how expensive they are was kind of a big shock for a lot of people. Well maybe not a lot of people. I think some of us always knew that, you know, that there was some subsidizing being happened. But the fact that open AI

7:10

Speaker 1: like lost $34 billion in a single year. That is like $1,221. 78 per second in 2025. That is an absolutely ludicrous amount of money to lose. Like it's absolutely insane that this is at how expensive this is. Um and so Like, you know, and it's so it was you know, seeing that stuff coming out and like the fact that these big companies are throwing this much money in it and pushing it so forward, like it definitely got us thinking that, you know, this is maybe not something we want to like throw in right to Wagtail right away. On top of like, you know, figuring out all that stuff, uh, you know, as a small open source project, a lot of these tools opened up, you know, the slop.

8:00

Speaker 1: uh issue like this is gonna date me a bit. I'm showing a reference from uh My Little Pony here in the 1985 film. Open source in 24 was like the ponies like building up their little spring festival and having an you know preparing for a nice time. And then uh in twenty twenty six is when this smooze came and absolutely inundated people with uh you know all the repositories with slop. And our community was no exception. We probably banned more people this year than we have in the past 10 years for this stuff. And so having to uh manage all that on top of these bigger questions was definitely a challenge for us But ultimately, what came out of our thinking and our discussion, and the reason that we decided that we absolutely did need to stand up and have something to say about AI

8:53

Speaker 1: was the adoption trend. Like unlike blockchain and some of the other previous hype cycles, so many people were adopting these tools. Like this this curve chart here shows that it took years. almost two or three times as long for people to adapt the internet or the personal computer than it has for agentic AI. That curves the tiny little one all the way on the left hand side. Like it is meteoric rise in terms of adoption. And we needed to have an answer for our users who were re increasingly both concerned and some are also excited about this world where you have more AI agents flowing around the internet. And so what did we do about it? This is uh how

9:39

Speaker 1: we decided to create our approach to AI. This mostly came out of um some discussions between our product uh lead Tebow and uh Ian, my boss, uh, were the ones that kind of started it and then they of course brought it to the wider group for approval and to make sure that we were happy with what they came up with And so pretty much we published this blog post in 2025 that contained what we call our guiding principles for AI. And what are those guiding principles? First of all, no AI dependency will ever be in Wagtail Core. We are not going to force AI onto anybody who doesn't want it. And that is particularly important

10:25

Speaker 1: for we have a lot of governments using our software. And again, those legal and regulatory gray zones are ones that we want to respect and give people full control over what they decide to do. And then our second principle is to take a responsible approach to AI whenever possible. If we can, for example, choose a model that has a lower environmental impact. uh then we're definitely going to go for that option. There are some people who can who can argue in good faith that there is no responsible approach to AI. But uh this principle is included mostly to state that whenever uh there is a more responsible option, we're going to take it. Our third principle, we will always remain model and provider agnostic.

11:13

Speaker 1: We are not going to pick a horse in this race. We are not going to force anybody to use a particular third party. uh model or a particular third-party package. Only the right AI. And this one is where like the community will help us shape this particular principle. Uh because like we don't want to shove AI into uh software just because it's AI. Uh we want some actual evidence that it might actually make something faster, that it might actually help people do things. For example, alt text is the one that we use as an example all the time. It's better to have AI-generated alt text and have it actually exist than have nothing at all. uh and it saves people so much time. So that's an example of like something we're definitely pushing.

12:01

Speaker 1: Um and the fifth principle is the most important one which is keeping a human in the loop whenever possible. We don't want anybody to ever feel like their AI is controlling them. We want them to be in the driver's seat. So um and this is our Wagtail AI package. Uh you know, we do what pant package maintainers do and we went and built another package uh to give people the the option to use AI if they want to And we built this package. Well, Tom started it before these principles exist, but we definitely have shaped it to uh to reflect our principles. And so it contains multiple vendor options. And if you don't like any of those vendor options, you can add your own. It has uh the user gets to decide how much AI

12:47

Speaker 1: to use and where, like down to the specific fields within Wagtail, you can be like, hey, I want to use AI here, but not here. And AI goes here or not there. Um it gives you pretty much total control. And also there's user control over the prompts and the token budget. For developers who like want to set their content people free, the content editors actually have access to the prompts too, so they don't have to pester you about that. We put that feature in to make it a bit easier. And our package focuses on like empowering people on kind of these common repeatable tasks in publishing, like the alt text example I used earlier We also give people the power to summarize their content in summary fields, and then we give people

13:35

Speaker 1: there's a like writing feedback option. And if you are the type of person who does like AI to draft, you can use Wagtail AI to set up an initial rough draft. sequence with AI as well. And there's lots of these like features that are aimed to actually save people time and to aid the human who's driving this as well. Other things that we've done within our community and outside of our community, we've conducted an AI user survey. We definitely found that a lot of people are using it and we uh got some good insights. uh onto where they needed support, usually around skills uh and providing good examples of how to use AI in an effective manner. Um we also

14:20

Speaker 1: have made it easier for our docs on Wagtail uh docs. wagtail. org to be easier to read by AI agents. We provided some better formats for that so that they can uh learn our stuff better. And then we've provided actual like kind of tested skill files. And one of the things that we're doing that not a lot of other places are doing is we've been measuring the carbon impact. of various models. And this chart here on the right shows kind of the average carbon footprint of all the tasks that Wagtail AI can do. The most expensive one understandably is the content feedback one because that involves a lot of tokens. But like most of these are relatively small impacts.

15:08

Speaker 1: Like they're not nothing, but uh you know when you compare them to like having a video autoplane on your website, they're actually pretty small. So what are our next steps for Wagtail and AI? Our next major release, which is coming out tomorrow. uh will have a fully supported write API. It's a very rich API. I got to play with it last week. If you want to set an AI agent loose on your Wagtail website, you can use that out. Uh definitely don't do what I did in my demo though, which I dumbly gave uh like well I wrote a script instead of uh setting an aging on things, but don't give it super admin powers even on a local demo. Just bad practice

15:53

Speaker 1: Uh uh we're gonna continue uh do sustainability and quality research. I think one of the important roles for us at the center of all this is to help people figure out which models are the most efficient ones for them to use with our software in particular because that's the name of the game in AI now, like everything is different. Like you can't write an evaluation for code everywhere. You have to look at your specific tasks. And we want to help people figure things out for Wagtail specific stuff. And then we're going to keep focusing on the DX experience and providing, you know, some support for people who want to use AI in addition to the people who don't use AI, because honestly, having a new command line tool would make both groups happy. And so we're looking for wins like that.

16:40

Speaker 1: All right. So our goal overall, um, because nobody knows Let's be honest, nobody actually knows what's going to happen next with AI. And I I bring up this as an example of proof of that, is that I don't think a year ago anybody would have thought that the open models would be where they are today. I've seen this chart at multiple different conferences now where they're catching up. Uh and you know, the world where we were thinking we were heading towards where we have one or two super big players. is not coming true. And it's a relief to me, honestly. So what we offer our strategy as an open source project around AI and our you know, as as

17:26

Speaker 1: as human beings in this rapidly changing world, we want to give people resiliency. That's what these guiding principles are meant to do. And like even as things are changing around people, we want to give them a sense of control. uh feeling that they can that their software is theirs and not uh somebody else's that can just be changed or taken away uh at in the blink of an eye. And we think that that uh and we're seeing some evidence that that that people are responding to that. These are our Pi PI downloads. We had a huge surge in interest this year. You know, the spike has come down over the summer, which is pretty expected. But the fact that, you know, people we're seeing a continuous increase in people being interested in it has been a good indication that we're heading in the right direction.

18:14

Speaker 1: And then also we have some people who are like shouting us out on the internet and in the communities, like Tim here uh, you know, uh made a point on uh LinkedIn that, you know, that we are never going to force AI on people. And so like the we're seeing some indication that the community is rallying around us, that people are responding to this in positive ways. And so like the urge to like you know shove AI into every like little nook and cranny of your software is something that I think should be resisted because people want choices. And I want to end with a quote from somebody who uh is like if you're not following Morton Rand Hendrickson, you definitely should be, he says a lot of smart things around AI and humanity.

19:04

Speaker 1: Part of the reason I connected with him is he also has a background in content management systems. But this quote is listen to the big AI companies and you quickly start believing that their tech is inevitable And they are the only ones who can save us from it. When we start believing the myth marketing of companies whose financial future relies on that belief, Our inner subjective reality is distorted and we end up building the future their investors need instead of the future we want. We are building a future where people should feel empowered to use AI to solve their problems. We want to build a future where people feel like AI is their ally instead of their enemy.

19:52

Speaker 1: We want to build a future where money is not the only metric that matters. We want to build a future where people will always have choice. And if that is a future you want to build, then join us. You can find out more at Wagtail dot organize. Also, real quick before we get to questions, please come to Wagtail Space. It's uh happening in November 18th through the 20th. I'll throw a link on the Slack too for everybody to come. It's totally free and it's online so you can come from anywhere in the world. And I also would like to thank my employer, Tarchbox,

20:38

Speaker 1: for giving me the chance to be here today. And here is a QR code if you want the slides. Any questions?

20:50

Speaker 2: Thank you for your talk. I was curious, I believe you you mentioned that you uh compared or at Wagtail, the Wagtail team compared the um impact Of different models. And if you had more information on that that you'd like to speak about?

21:06

Speaker 1: So uh the blog post I'm gonna refer you to here, uh two of them. uh is uh we have a blog on comparing the different open weight model providers. Um So Tebow has a lot of different inflammation that we collected on the different criteria for looking at them. Pretty much uh the ones that uh he found uh worked really well for Wagtail and actually a lot of them overlap with Django too. Deep Seek is one that does particularly well with Django, and it's not surprising it does well with Wagtail as well. And then the Gemma 4 models from Google and GLM5. So these are like he has a lot of good information here on the criteria that were used.

21:54

Speaker 1: I argue that we need to do a bit more transparency around the score, the specific scoring criteria. Um, but that's something that we're definitely working on. Um and then the other one is the carbon footprint blog post. That one is is a good one for the different uh this was when he did the Wagtail AI test, but again he went through uh some Looks like we used Kimi K2 for most of this one, but he might have some more details in here about other ones that he tested that I can't remember off the top of my head. So I highly recommend digging into that data.

22:46

Speaker 3: You mentioned that you modified the docs to make them uh easier for AIs to read. What was involved in that?

22:57

Speaker 1: Mostly making sure that the you know everything appeared in markdown format. And also like if we go to our docs website here, I can show you we also added a little interface. Up here it's under view source. I think actually there might be a better demo on the guide website. uh that we use for user manuals. Uh let's go to that one. We we added uh like copy options So people who use like the guide website is more for average users and content, like editors versus developers. But we provided ways for people to access

23:45

Speaker 1: Markdown, do the LLMs. txt for the website and things like that. Just basically gave people more information that they can feed to their agents and their chatbots. So previously it was like you could get the MD files from our repository, but they weren't quite as organized.

24:08

Speaker 4: How can I get involved in Wagspail Wagtail Space twenty twenty six to learn more?

24:16

Speaker 1: Well you you can uh you can definitely uh go to uh wagtail. org uh to get more information on that. You can talk to me uh and then um we are looking four speakers right now actually. So if anybody has any uh great ideas and it doesn't just have to be about Wagtail. We want uh people to come talk to us about anything interesting they're doing in Django and AI. And particularly sustainability and accessibility are two things that we're real keen to get speakers on. So I'll definitely put some links in the Slack channel for everybody.

24:59

Speaker 5: Hi. Lincoln Loop, thanks for the shout-out. Really appreciate that. One of the questions that I have is that Over the past year, I have seen most of our clients shift from I want to go in and use a web app to build something to I want to just take all my documents and files and whatever else and have Claude build out the thing for me.

25:19

Speaker 1: Yep.

25:19

Speaker 5: Um, and I saw some of that in the talk here, and I think I've seen some buzz around it. But as beautiful as Wagtail is. pointing and clicking and moving things around is never gonna be as easy as here's a document, just build it for me, you dumb clanker.

25:33

Speaker 1: Yep.

25:33

Speaker 5: Um what is the How do you see Wagtail working in that ecosystem because having a hard time fighting that when some of the other CMSs have that functionality or will have those hooks? And I know they're not hard hooks to build But um how do you see that vision fitting there and how people

25:52

Speaker 1: I I think we're definitely heading in that direction and kind of the big blocker for the longest time was adding that right API. Um and now that that's in place, I know that Tebow's been experimenting with uh with a doc builder. Um so like I I don't know a timeline at all. Uh you know, and and uh honestly, like, you know More sponsorship dollars help help shift perspectives and uh help move things forward in that regard. Always uh always open to that. But I think that, you know, overall content management systems in the future, we have to start thinking about the fact that they are largely going to become clearinghouses and sources of truth for these machines. Uh we have to start thinking, like, you know, I I I love to brag about how pretty Wagtail is, but there honestly may be a future where that doesn't matter

26:44

Speaker 1: as much nearly as uh you know how efficiently can it move data around or can it approve permissions and things like that. And so that's we're kind of laying the foundation for what like whatever direction that tends to go in without losing our roots too. Like we want to make sure that we never forget that humans are the ones who matter. I think that was the last question. Thank you, everybody.

Presenters

Conference

DjangoCon US

August 24 - 28, 2026

Chicago, Illinois, USA

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 Meagen Voss

More videos from DjangoCon US