Go back

How Wistia Ships 100+ Code Changes in 2 Weeks With One AI Agent ft. Chris Savage & Brendan Schwartz

44m 41s

How Wistia Ships 100+ Code Changes in 2 Weeks With One AI Agent ft. Chris Savage & Brendan Schwartz

This episode of Unchurned features host Josh Schachter interviewing Wistia co-founders Chris Savage (CEO) and Brendan Schwartz (CTO) about their company-wide AI transformation. Wistia had already improved its shipping cadence in 2022–2023 by abandoning long-term roadmaps and requiring product teams to ship every two weeks. When AI coding tools surged in late 2025, engineering and product pulled further ahead of go-to-market, prompting a February hackathon that produced the precursor to the Wizard, an in-house agent built on the open-source Pi harness after early OpenClaw experiments raised security concerns. The Wizard is one shared agent connected to Wistia's internal systems, usable only in public Slack channels so knowledge spreads and coordination bottlenecks disappear. It can build web apps called artifacts, answer questions using sales, support, and wiki data, and even predict how executives would approach a problem. Automated software factories handle bug fixes, click-testing, and auto-merging PRs, with one shipping over 100 PRs in two weeks. The founders acknowledge cultural challenges: public work feels unnatural, some craft-focused engineers struggle, and burnout from constant decision-making is a real risk. They also note guardrails around PII and customer data. The Wizard costs about $30,000 monthly plus a small dedicated team, and Wistia has reorganized so product and marketing teams are tightly integrated. The founders urge companies to operate as if they have unlimited resources.

Transcription

9385 Words, 48733 Characters

English
Speaker 1we've got like a few software factories running. There's one for machine that's built like a fair amount of code there. It's probably shipped over a hundred PRS in the last two weeks. Most of it autonomously, a lot of work it's found. So it's like looking at production logs and finding bugs. It's click testing as users finding like UI jank performance concerns, and then PR in those testing. What about judgment?
Speaker 2Brendan does, does the wizard push back? You're listening to unchurned brought to you by the Gainsight podcast network. Subscribe to our sub stack at unchurned.gainsight.com where we go deep on every episode, like how one post sales team at cloud beds built over 150 AI agents, that story and more at unchurned.gainsight.com. Hey, welcome to this week's episode of unchurned. I'm your host, Josh Schachter, senior vice president of strategy and go to market development at Gainsight. And I'm thrilled to be here this week with two buddies of mine, the co-founders of Wistia, Brendan Schwartz, the chief technology officer and Chris Savage, the CEO guys. I'm really excited about what we're going to talk about today. You're really like you've, you've turned the company on its head and, and done such an amazing transformation through AI. We're going to talk about all the things. Thank you so much for being on the show. Thank you for having us. Very excited to talk about this and to get both the co-founders on the show together. It, it means a lot to me. So we're going to talk about Wistia's AI transformation. You got this thing called the wizard. I want to really kind of hear about it. How things were prior to the wizard prior to this transformation. You can tell me what the timeline of that was, how many years or months ago, quarters ago that was. And then let's work our way up into what you're doing today, which I know is really cool. I mean, so the last
Speaker 3since like 2022, 2023, we had actually figured out how to ship much faster with the same size team. We had shifted our culture a lot. There was a moment in 2022 when we'd, we'd hired a lot of folks and we were not going anywhere. We were like, well, we're going to ship this thing way faster. In fact, it's like we're going slower. We were like, well, we lost the excuse of basically being understaffed, which is what we always told ourselves was the issue. And we realized we had to shift our culture a lot. And so we kind of moved away from having very long-term roadmaps. We basically told product teams like, Hey, you own this area, but like the one thing we want you to do is ship something every two weeks. And there were projects that before that happened, they were telling us it was going to take six, nine months to do that. We're done in the first two weeks. So that was remarkable. And you could see it in the results. Like you could see it in what we were shipping and the data of how things were performing. Like we had this big breakthrough and that's kind of where we were and we were feeling pretty good about ourselves, I would say, you know, pretty high on our own supply. But then this year really AI started to really deliver for software development, like December, 2025, we realized this is something is changing here in a big way. And it's helping us with like individual productivity, but it's not leaping the company forward the same way that that kind of cultural change had happened before. And that led us towards finding like a new way to operate around AI. Um, and that really the creation of the wizard.
Speaker 2So it started out with product development, right? Like, like a lot of the, the, the AI efficiencies have, um, been, it was with your coding and your, your, uh, release cycle. Um, Brendan, you then decided to let's experiment and see what we can do to have this thing percolate through the rest of the organization. So what was the origin of that?
Speaker 1Yeah, it was exactly that we were feeling. Like engineering and the product org was getting so far ahead. And to Chris's point, we were already, this is like pre AI. When we were shipping so much faster, we had all this issue where it's like products making stuff faster than we can tell the market about it. And this is all, you can, this sounds quaint to say, right. Cause this is pre AI. So things were not moving that nearly as fast as they are now. Um, but then it just started getting exacerbating the problem. Like go to market would kind of catch up, but then engineering a product would be moving fast. Um, and so there was this fear that, you know, like AI productivity, there was this fear it was just going to widen. Um, and so there was a hackathon we had in February and our, um, head of content on the marketing team happened to be in the office leading up to that. And I was showing him kind of all this, this experimentation we were doing with open claw at the time, living in Slack and, uh, like I, and a few other people were doing basically all of their development. And I was showing him kind of all this, this experimentation we were doing basically all of their development in Slack in like January or something like that, or from your phone. Right. So it's kind of now again, as quaint, maybe most people
Speaker 2listening to this are not as advanced as, as you guys are myself included. So you flew by something pretty profound. Repeat that you, did I hear you correctly that you and most of the development team were doing your coding and releases in Slack?
Speaker 1Not most just, uh, like me and a few other people were playing. Okay. You were
Speaker 3literally really clear because actually before Slack was texting and Brendan was like, Hey, I set it up so that I can code on the ski slope. So when I'm riding the lift, I'm getting stuck. Like he would, it would be the Monday and he'd be like, Oh, all this stuff shipped. I'd be like, how you do that? I was like, what'd you do this week? I was like, I was skiing.
Speaker 2So, so this was like a work-life balance thing we had going on. I feel like it's kind of like a little bit of the impetus.
Speaker 1I think it's like, it's easier. I don't know if you're like as a manager or something. So like, I don't have huge blocks of time. And so the time, most of the time that I had was between meetings or like at night, you know, just like between kids stuff or something. So I was like, Oh, I can get this work on my phone. Then I could like, and on the chairlift. Yeah. On the chairlift, I could like be shipping some stuff. So we had been playing around with open claw. Does everyone know what open claws?
Speaker 2Why don't you give us the 32nd?
Speaker 1That was, I'd say like a key innovation. It was an open source project or it is still is this open source project. That was an early version. It was an early version of where, you know, if you had used Claude or, or chat GPT originally, it was like a glorified search engine. And then that, like the idea was like, you give your agent hands. So you give it the ability to actually take action and do things on your behalf, which now is built into Claude and chat GPT. And you've got like instinct and crock pot and muse and all these other things. This is, they, they kind of like came out of that, like, you know, open claw.
Speaker 2But at the time, were you, were you, were you sticking your neck out there? I mean, you guys are the founders, the owners, like, um, but at the time. At the time to, to be using open claw, was there risk involved in that as far as it insecurity?
Speaker 1Um, I would say we had it pretty locked down, so it could not, this one could not do very much. Like it could work in like one repository in certain settings.
Speaker 2So you had all the guardrails on, we can, we can move on. So you had, you had, you had open claw, um, you're, you're texting to release during the, you know, the ride up Mount Stowe.
Speaker 1And then, so. So I was, you know, showing people it's in Slack. So everyone, a lot of people can see this. And then, so our head of content marketing, I was like, Hey, this is hackathon coming up. You want to just team up and we could build, you know, some version of this for go to market or for marketing or something that could help. Um, you know, he's like, all right. He, the funny thing too, is he was just getting back from parental leave. So he was out for four months, maybe starting in November or September, which is like, as Chris alluded to, that was when there was. That huge breakthrough, uh, with Opus, I think Opus four or five. And he, he came back and he's like, what the hell happened in these like four months? Um, and so we, for the hackathon, we built basically the precursor to wizard. I mean, it's called wizard. It was an open claw that was working, uh, and it was like hooked up to our, more of our go to market stack. And so it could do some really amazing things like, uh, help you answer, like how is website performance doing? Or it could build pages on the website or it could, you know, help you like manage projects in a sauna.
Speaker 2So, okay. So, so you just described part of what wizard does, um, wizard, you know, Chris, you and I, we, we were at the pavilion conference. We were talking and I was like, wow, I gotta have you on the show because what you're telling me this, you've made this the, uh, the gravity of the entire company. Wizard is the brain. It's the core, Brendan. Okay. How does, what is it?
Speaker 1Wizard. So we took, so that initial open claw thing at some point, it like two weeks in, we're like, this is good, but it was to your security question. We're like, this is terrifying. We like, don't own this code. We don't quite know. It's like not super reliable. And we're like, we will build our own version of this, which I remember thinking, like, I, I left like one Friday and I was like, this is the stupidest idea. Like, uh, this like could be the stupidest idea we have, but I'm like, oh, let's see how far we could get over the weekend. Um, and then came back Monday. And then we like added a whole bunch to it, but it, it basically is a, is an agent. It runs in a open source harness called PI, which a lot of these things use, which is really fabulous. If you're building your own thing, it does it, that doesn't do too much for you, but it like provides like a way to talk to LMS. And then most of what wizard is, is like a lot of convention and hooking up to all our internal systems. So. This thing can, you know, see, basically see and do what a human could do inside of Wistia.
Speaker 3There's some key things I think in terms of how it's set up and still works, which is like, it is one agent that everybody uses. It's hooked into all the systems, which, but, and then you can only interact with it in public, which means like in public means like open Slack rooms. And that is required like a cultural shift, right? Because everyone was just doing their own work on their own. And then taking the results of that work. And bring it to meetings and bring it to stuff. And I try to move forward, which is like, you know, definitely making was making. making things more productive, but like things were still bottlenecked in the old way of working. And the thing that's so different about the wizard is because it's in public. Um, and you can see it, it's not just getting work done. It's literally like we're, we're getting things done months faster than we would otherwise, just simply because of the coordination tax of getting people together. Um, you can see everyone's work. So if I build something with the wizard, even in a small channel, that's just me and the wizard, I can tag someone else in and they can see it and they can ask questions and they can refine. And then it's as simple as clicking like go, like it, it can build a code. It can understand the infrastructure. It's like, if you had the best of vibe coding, making pretty things, but it's completely disconnected. Well, if it's all connected, it also has access to all of our data. And we even built like an internal wiki. That's like the brain that's managed by a few people on our strategy, um, how things work with like CSMs and account managers and all that kind of stuff. And so you get nuanced answers back. And what's happening is that things are just going so, so much faster than I ever could have imagined because of the setup, because it's in the open, because it's like one shared
Speaker 2agent. So if I'm going to oversimplify what it is, it's, if you, if you imagine plugging in all of your data sources into Slack, using Slack as a company-wide interface to do all your agentic work and vibe coding, that's what wizard is in public.
Speaker 3Yeah. Except it's as Brian, it says like, it's also an all the go to market. So it's all the other sites. And it's, it's so that's, that's what makes it so different as you can build something or ask a question and it's, it's looking at sales call information and it's looking at support interactions and it can try to guess that like if the support interactions are because of bugs or other, so you get this like incredibly real time view and judgment from the wizard itself.
Speaker 2So like the, the inputs come in from any data source from, from gone calls, from whatever you're using for your wiki, stuff like that. But the, the, um, the actual like interface for it is Slack. And where else it's
Speaker 1Slack. And, uh, it has something called artifacts, which are, if you use Claude or chat, you'd be familiar. It's like basically a build a webpage for you started very simple as a webpage. Then we changed artifacts. So they are effectively like a full-fledged app that the wizard can administrate and you can interact with. So a lot of people interact. This is where it gets, it's like a little hard to explain. You might, uh, make an app like a custom app to do some part of your job. And you will be working in that cost, basically custom software that wizard built for you, um, in a combination of Slack or that web or that app.
Speaker 3We should talk about the, the nurture factor is a good example. Like last week we're in person. We have this new product machine, not that many people using it yet. It's cause launches. Well, I don't know when this comes out a little just launched, but you know, it's in customers cause some customers can access it and stuff. And we want to send emails to the people who are using it and I want to send them and I want to see what's going on. So Brent's like, Hey, look, I use the wizard. I built this thing. And now you can, uh, just in Slack, decide the types of people that you want to segment and talk to. You can create the emails and all this kind of stuff. So he shows us to me on like Tuesday morning and I refine it and I go through and I, I ended up sending like a hundred emails the first day that I look at all the emails. It takes me 15 minutes probably to do it. And someone on our team had an idea for a different segment to talk to. So she's in Slack and she's like, Hey, what if we talked to this segment of folks? Like this would be really interesting. We're like, Oh yeah, that'd be cool. Just like at wizard do that. Five minutes later, it's done. Then the emails are in. I just saw the emails. We send them out. We just started booking meetings with people like in like real time. It was crazy. So it's just what it knows about motion.
Speaker 2Motion is your new motion graphics, um, uh, feature, right? Which you were just showing me, which is, is I'm incredibly excited about. It's taking what it knows about motion from the product requirements and from maybe even the code base itself. I don't know. And it's helping to create product marketing copy that then feeds into your outbound emails based on, you know, probably a single shot prompt that you have there. And then it's publishing that into probably like a Gmail drafts for you. And then you just could do a quick human in loop review and press send.
Speaker 3That's basically what's happening. Um, the only thing I'll add is because it understands the infrastructure of the company, there was a bunch of people in that first one shot that it's like, Hey, this, these 30 folks. You shouldn't be emailing them. Like this person has a meeting booked soon with their CSM. This person just talked to an account manager. Do not email those people, but I can ping those people for you. So all of that was in the first, so it's, and, but it's, you have to, it's like this for everything.
Speaker 2So an example that comes, comes from like, from my own experience just yesterday, um, we've launched a thing called the CCO club. It's a virtual round table for customer success CCOs and, and like the senior, most people at companies. And I had put in the Slack channel, a message to the entire field team. Hey guys, this is truly valuable for your customers and your prospects. Can you please send out an email inviting them to this, to register and sign up for this new club? Um, and CC me because I am personally going to reply all to all those emails, invitation to make it personal. And then the CRO was like, this is wonderful, Josh in the, in the Slack channel. She's like, this is great. Um, but who have we already sent the email to? Because I want to make sure that we're not doubling up on people. So you're saying, well, she's like, this is great. So you're saying machine, because it has access to everybody would have known exactly who was already within the whole organization.
Speaker 3All that stuff you didn't even have to do.
Speaker 1Yeah. Yeah. Yeah. So that is like the true magic of why you want it. Like every shred of context in a single centralized place, because you can do things like that, that in, in the old world is like all this like SAS data pipeline stitching, like you're always missing something. Like, I mean, most organizations, I think do a good job. You get your like 80%. Of it in there. But like, for instance, to do outbound, like to do outbound in our traditional go-to-market stack, like we're not gonna pipe all of our Slack history into whatever is sending those emails. That makes, that makes no sense. But in this case, it also is flagging stuff because it looking through Slack and it's like, oh, Chris, uh, you know, that company's like Brendan's brother's company. So like when you email them, like say that, you know, like that type of context, just, well, not gonna have, this is making me think about,
Speaker 3I'm gonna make a Brendan was out. There was a designer on our team working on a problem and she wanted Brendan's advice and it occurred to her. I should ask the wizard what Brendan would do. And so she asked the wizard what Brendan would do. And it was like, well, he dealt with problems like X, Y, Z in this way. So probably he'd deal with this problem in ABC way. And by the way, there's no like Markdown file of like what Brendan would do. It's not exactly. That's what's that's. And this is, this is the thing I think of like the one shared agent thing where, because it's learning from everybody that just worked, she was unblocked. She did the thing. Brendan saw it is back. It's like, great job. And I think that it's, but this is the cultural difference, right? Because if you, if you can't, if everyone's using their own set and to be clear, we have people who are still using their own agents on things. Um, like people are using cloud, people are using chat GPT for a lot, for lots of different things. It's just that most of the work is in the wizard and it's going at like a faster and faster rate, right? Like as this shift is occurring. Uh, that by having this one system, you're not bottlenecked by me. We're thinking to tell the person by me, giving the person all the context by waiting for the meeting to happen. And so that's where things are just at such a, just a ridiculous rate of increase.
Speaker 2I start to see now you, you mentioned Chris, I think upfront, like the biggest outcome is, is removing the bottlenecks, right? Because I pinged, uh, my CEO, uh, Chuck this morning saying, Hey Chuck, would you mind, um, helping us post something in this public Slack channel? About blah, blah, blah. And like, but that judgment of whether he wants to do that or not, that can actually come back from, you know, from, from a central intelligence that you guys have built the judgment layer of, of your executive team so that I don't have to wait on his response. Right. Cause obviously he's got quite a few slacks that are going on. Uh, I'm not calling him out. Right. But like, it's this idea of like the decision making is now democratized in some way. Uh, so talk about Chris, the. Reiterate for us like the, the, um, the public versus private of the, of the, uh, um, I want to call it a bot, but, uh, of, uh, the wizard, the importance of it being only available in public channels and not splintering off into private channels.
Speaker 3Yeah. I mean, this was a clear decision to the beginning. I think it was really Brendan who pushed the hardest on this. Um, and the idea was we really want to learn from everybody and we want everyone to learn from the things that we're doing. Yeah. Yeah. Yeah. Yeah. So we, we started talking pretty quickly about like, Hey, we run a lot of tests. We run a lot of tests on the growth team. We're a lot of run a lot of tests and marketing. There's a lot of tests that are different parts of the product, but the results of those tests are in little, you know, analysis sheets that human beings are looking at and they're making decisions on it, but nobody else really knows about this. Like, and the, the answer before was reported out in a meeting, you know, reported out in a Slack message. It was like information
Speaker 2overload. That's the dissemination of information is meetings. That's basically what it is, right?
Speaker 3Like that's your arc information architecture. And so the idea here was like, maybe if we have it all in one thing, we can skip that problem entirely. And so everyone's learning from it. So I don't know that Brendan ran a test last week on something and got a result, but it doesn't matter. So I asked my question. It's like, well, Brendan tried this and it didn't work. And it was similar. Do you still want to do it? You look at it like, no, actually I don't have my answer or there's a different way I would do it. So there's a good reason why I'm going to push forward. But now I'm going to do it. I'm going to do it. I'm going to do it. I'm going to do it. I'm going to do it. I'm going to do it. I know I'm informed of the thing at the right moment at the right time, but it's that across like every. surface area everything yeah and so that is the biggest difference is like it's i mean i talk about being months faster part of it's that the ai is doing the work part of it just culturally we can go months faster on things because you don't have the same information architecture of meetings where you're disseminating all the information through
Speaker 1the other piece of it too that was is really important is the the discovery so the whole world is figuring out how to work in these new ways right there's a lot of stuff that has to be invented for lack of a better word of like tricks of how you work with agents and lms to you know get work into the shape that works well for them and by having it all open means it's discoverable so you can cruise around in slack and see it's like that example designer like what is she doing that she has i can see her having so much success like what is what is she doing and then you can get that inspiration and you get more of that diffusion and and learning versus it being if that's all hidden away you're only going to learn about that either at your own pace or through meetings or other structured things which are just inherently going to be slower
Speaker 2i can just see your general counsel cringing at the discoverability that you're talking about
Speaker 3well you know it's funny we've talked for so long like an answer that i i often give when people ask about leadership in general is like the most powerful leadership is modeling like you want to model something and like an example is like let's say you want to have like uh a culture of more feedback like you just need to give feedback publicly or let's say you want meetings to always start on time like you don't tell everyone start the meetings on time you start the meetings no matter what if people aren't there and guess what pretty quickly all meetings in the organization are started on time right i think what's happening with this stuff is we're able to model how we're working we're in our we we have everyone at wistia basically building these brain channels which is like their thinking channel with themselves the wizard usually there's a group of other people also like following along hmm but then if someone finds a new way of working then you see oh brendan has the wizard making this board for him every morning and it pulls up all the major tasks he has and he jumps to the board and he answers the things where like a human input is needed this is it then you say i want that too so you go to the wizard be like give me the board that brendan has so you each
Speaker 2have a slack channel that's like chris's thinking channel brendan's thinking channel it's kind of like your your notepad and your canvas maybe for some dashboards and some some some daily updates and stuff and while you're probably the only ones in it with the wizard anybody could join it theoretically right because that's the rules of the wizard and so it's able to go through each of your morning routines let's say from your private thinking channel or like the little thoughts that you send yourself while you're on the ski slopes and it's able to rummage through that that's what you're
Speaker 1saying yeah i mean there's 25 people in my brain channel so like mostly they're not but they're like you can be in there yeah and i think
Speaker 3the the other thing we're saying is where before you'd have to call in all hands and be like everyone needs to have their own channel you need to create this board we expect you to work like this like we don't ever have to say that because everyone just sees it happening right and adopts it because it's so much faster and easier way to work what okay
Speaker 2let's talk about big brother who's big brother in this situation like what's what are the concerns here about privacy and and that sort of thing
Speaker 1like for from an employee
Speaker 2standpoint yeah i'm i'm just i'm i'm a straw man argument right but like what is there any pushback around privacy yeah i think
Speaker 1there's a ton i think it is working this much in public is an unnatural way to work for most people and i think this is the biggest cultural change we've had to make to get this to work because there are so many reasons that this is hard like you could say i want a private space to do my private thinking because if i feel like i'm on stage and people are watching me you know i can't i'm not going to take as many risks i'm not going to do that and i think the only way that we can really make it work is like everybody has to participate and you have to you have to normalize it and people have to get used to it and say oh it's okay i'm saying and to chris's point modeling like doing a lot of unfinished work like if you look in my brain channel chris's brain channel like i'm doing dumb stuff i'm not purposely doing dumb stuff but like i'm a human being like i have stupid ideas sometimes right like and they're all in there just with the good ones um and i think but i think it has been like it has been a challenge to to do this it is like a more extreme way of uh working right and there are
Speaker 3certain things that have not we've purposely not put into the wizard right so there's like employee information certain customer information that are purposely not put in to not create a larger issue uh for many reasons um so good disclaimer to share well it's just it's worth saying i mean you could be like these people are so ai pill they've lost their minds like no i don't think we're all the way there yet we might have lost our minds but i don't think we're i want to say one other no
Speaker 1for people who are listening on this uh because it was an interesting thing that we got to is i think like in any reasonably sized organization who is using it either you're like you cannot use ai right or you have like these such tight controls on your employee base or i think where most people are are like people have chachi you know some enterprise like chachi t or claude or something like this and you got people hooking up stuff all different ways right like you don't have most organizations don't have as tight control as they want and one thing let's just say on the customer data side we spent a lot of time building like the right way to integrate that into the wizard where it redacts pii and will not put that in there and but by i think by making that the most convenient source to do your work you actually have like better stewardship over customer data and privacy because you don't have people being like well i just got to get this done so i'll just like pick up this thing and then you don't know you don't know yeah yeah you don't know what's happening
Speaker 2no exceptions um yeah like what about permissioning particularly i don't this is the factor or not but particularly as we talk about releasing code there's got to be some permissioning there right
Speaker 1i would say that like through all of this like we've leaned on a principle that we've long had which is we hire smart people and we let them use their judgment right we don't try to build a lot of processes like where it's going to help us but for the most part part it is you lean on human judgment and do not like i don't know hide behind process or other things so we're kind of applying the same thing in this world the agent does what you say right it doesn't just randomly do stuff like or or just then it does then it's you have used bad judgment because you have not understood what the model is doing and how the agent works and so it is the job of like the product teams or engineers to like engineer this in a way that is going to be safe um and work well but like there are a lot of experimentation um basically got to the place where we're like four various lanes of work in our product like built up enough like good ci pipeline good uh agent review pipeline where there's enough evidence it's like gonna click test the full feature or change to be able to build up enough confidence that it can auto merge those things so we've got like a few software factories running there's one for machine that's built like a fair amount of code there it's probably shipped over 100 prs in the last two weeks uh most of it autonomously a lot of work it's found so it's like looking at production logs and finding bugs it's click testing as users finding like ui jank performance concerns and then pr-ing those testing what about judgment
Speaker 2brendan like you know i was a product manager for for much of my career and i came up with some really shitty ideas like and uh chris you know what i'm talking about right and uh and uh you know like the product founder right and uh but you know you sleep on them and you realize they're not as good as they were the night before and then your your ux ui designer tells you just how shitty they are and then the front engineer tells you no no you should be doing this and they're actually better at product than you are i'm getting into a little bit of product therapy here but like do you understand what i'm trying to say like does does the wizard push back and like nope we're not going to up the product
Speaker 1like that the wizard will do what i mean sometimes it'll be like this other per or this team decided not to do that because of xyz that's happened to me before i'm just struggling with the example but i'm like let's do blah blah and they're like you know that this team explicitly decided not to do that for this reason so like well that's a good idea okay sorry um but i would say it's again it's people's judgment one thing we're doing and all i think is happening at other companies you've got because everyone can almost do everyone else's job more so than they can right like if you couldn't write code before now you can and to chris like chris can ship code into the product he's gonna have to learn like he's gonna have to learn more of those skills or judgment about that so we're again our philosophy is not like make the wizard so it's the gate it's the people and like the culture and the company apparatus needs to learn and reconfigure itself to be able to do that so i think it's really important to learn and reconfigure itself to be able to do that so i think it's really important to learn and learn and reconfigure itself to like figure out how to make good software in this era um and with this factory thing in particular a lot of what i'm talking about that's going through these are not feature level things they're like chores bug fixes security improvements performance um stuff i'd also hot take on this like i think the models are quite good their first shot on a lot of feature ideas is like is pretty decent i think human judgment can often get those can like punch them up and get them better but
Speaker 2can like punch them up and get them better but
Speaker 1can like punch them up and get them better but yeah i've seen myself included make a lot of yeah i've seen myself included make a lot of yeah i've seen myself included make a lot of bad designs that once they hit reality bad designs that once they hit reality bad designs that once they hit reality you're like oops to your point that was a you're like oops to your point that was a you're like oops to your point that was a mistake uh and i think the median thing that's coming out of these is like a lot better um than
Speaker 3that it's also like in terms of people shipping something to the product everyone's like even though you have the power to do it it's like i guess anyone could have logged into hubspot and sent an email to a million people people didn't do it you know they're like worried right and so i do think in general people are it's easy to think of the failure state of that happening but actually most people are like kind of too reticent to do it and if anything we're trying to encourage people to do it more where um you know one other thing we've seen is we've been taking people who have never made product before and and they're having a real hand in building the product but they happen to be an expert at the exact thing that they're working on it's like motion handles motion graphics really well and i would give a lot of the credit to one guy on our team who basically built an internet-based app for people to do it and they're like oh my god i'm gonna do it and i'm gonna do it and they're like oh my god i'm gonna do it and they're like oh my god i'm gonna do it and they're like oh my god i'm gonna do it and they're like oh my god i'm gonna do it and they're like oh my god i'm interactive app with the wizard about motion graphics himself and like established all these like different ways of putting them together that then because of the way it was built was like pretty easy to bring directly into the product and like his hand is on it in a way and is he a motion
Speaker 2graphic designer yeah and
Speaker 3his hand is on in a way that it could not have been before and so i think one of the reasons why that part of the machine is so good is because of that and so it's trying to get the actual expert at the problem as close as possible to solving it and it does that again this is just so much cultural change yeah and um yeah i think like it it's it's it's that's what is going to slow this down for most people i think is that the cultural change is so challenging where you know you have most companies have specialists everywhere it's like hey you're not you are a specialist but that really you need to be t-shaped which means you're an expert at this but you're gonna do all the other stuff now they're a designer you're making ads you know you're shipping code to the product it's like what and then well you are and then once you'll get used to that it's like and it's like uh the dopamine doesn't stop flowing
Speaker 2how has it changed the um structure of the org and the role design within the org i mean we just
Speaker 3made a pretty big change where we have a few different products at wistia and we brought um products and go to market like really tightly together so that they're fully dedicated so where or you know you'd have a couple product teams working on something and then you'd have like the marketing team which is kind of like an you know almost an agency everyone is a specialist is specialized in something but has almost like an agency model of like okay we have these five different product shipping things which one are we marketing now instead of saying nope you're just on machine you're just on webinars you're just on video messaging you're just on like core management and now you're in all the meetings when they're making the stuff and your your voice is heard in a different way and their voice is heard in a different way if you're building on the go to side on how it's enabled to customers so it's we made that change a month ago yeah or so and i think it's it's been again like the very quickly seen by getting these folks together everyone feels more permission to ship and go even faster
Speaker 1yeah i think the biggest thing that we're seeing is um the execution of everything is is so fast or you know solved in some cases so it really is about decision making is the bottleneck whereas before it's like oh if it you know took us a month to ship a new version of whiskey.com the home page then we had a month to to make a good decision on it right like you'd match it to the pace actually it doesn't you don't need fast now everything is so fast uh so is there
Speaker 2fatigue because that's what you hear some people talk about with ai that now now you don't have to do the busy work all you have to do is make decisions and geez you know what i don't want to make a decision i want to be like obama and just have the same suit in my closet for every day of the week that is a that is that is
Speaker 3like i think one of the most real concerns
Speaker 1i don't i feel like people are having a grand time with it pers like i think it's i think the it definitely where we have there needs to be a lot of people come together to make a decision i think people have very low patience for that because they're like man this is wasting a lot of time and where it's really working well is like that's why we we reconfigured the teams because we want if we put go to market and product together it's like that team is everything it needs all the resources and a clear lane to execute that they can go as fast as they want and they have all the resources versus before this is i probably go a million miles an hour and then we're not like hooked up to go to market the whole thing is going to be a failure and we're like oh we can we go you know and now it's just it's combined and as like chris mentioned there's the lack of specialization so you've got people on the product who are doing go to market stuff or like sending emails to customers and like this is great this is like how it should be it feels very much like we are starting the company over again like it we are it feels maybe you're
Speaker 2just overwhelmed with all the ideas of what you can push now
Speaker 3without brand well i think it's what i say in terms of it being a concern i think it's like you can be doing six different things at once which i think can be extremely tiring to like context switch between these things which is like kind of like one of the challenges yeah um i i mean i said i've said this recently but like last week on wednesday or tuesday to thursday last week we had like a bunch of the company together in person and like when i said wednesday was definitely the most fun day i've ever had at wistia and it was i mean 20 yeah 20 years 20 years the most fun day okay tell us more and it was just i mean basically it was like sitting down having breakfast next to some folks like in the office and we're talking about something and they're then they're showing me the thing that they just did yesterday and then i turn there's two other people and they're on another project and they have some idea and it's live that afternoon and like every single conversation i was in was like really exciting breakthrough stuff doing on all aspects of the business and i definitely ended that day i was like holy shit like that was amazing and i'm like ready to sleep through the night you know what i mean like i i this i am good um and so i just think it's like uh it's it's very very different way of working i happen to enjoy it greatly but i i am concerned that some people will just burn out because the dope they'll be have too much dopamine overrun with dopamine well let's
Speaker 2talk about performance management so what what has changed or will be changing in those regards like who has not taken to this you know what archetypes of folks maybe have not taken to this as well uh how do you how does this change the way that you view the performance of your team those sorts
Speaker 1of things i think one one uh early one on the engineering side and i think it applies to other disciplines too that we're seeing is if you were in it because you really loved the craft that's been a tough transition because a lot of that craft is going away if you were in it like let's say you were an engineer or you're in product because you wanted the end result you wanted i don't to delight customers you wanted somebody using something that you built i think those people are having a really fun time it was easier for them i think to let go of you know some of the craft pieces which are yeah you know have are like disappearing from some of these things and so that's been a hard i'd say cultural change um for folks and and the other one that's i think tough too is uh you like need people who like to invent stuff or a really imaginative uh like really i think now is like a great time for that and folks who are like i've got a way i work and i've like honed it over the career like is it it's tough because you need to find new ways to work and those people who are like ah whatever i'll try like like and you know they're going to do all that experimentation and find some things that work there are like multiples more productivity that you could get versus in the old world it would be like a five percent or ten percent improvement would be phenomenal to achieve as a team
Speaker 3we did change the performance ladders or levels i should say like at the end of last year i think basically redefining like hey at different levels of the company like this is how much we expect you to be able to use ai to improve like your productivity like i we i think we did it we went through a performance recycle which we do twice a year and then we said all right now we have the new level so like six months till the next one so we're not like we tried to make expectations clear but as brennan's saying i i think is like if you're more worried about what you're losing then this is concerning if you have an easier time imagining what's now possible then it's like more fun does it
Speaker 2allow you to view performance in a different way are you asking the ai about somebody's performance are you asking about the outcomes they're delivering you know what's the biggest
Speaker 3change is that like we're seeing i've seen people who in the old world were good and in the new world are incredible and like why why are they incredible it's often because like you know people they had some vision for how to do something and they didn't have something but they actually were incredibly limited by the structure the way we organize the company the way we set goals um what they had permission to do or maybe they weren't that good
Speaker 1at explaining their vision too that's another
Speaker 3yeah totally i think that's like they very much could be the case like you have a great vision and can't explain it but when you can actually do it yourself the vision is fantastic so we're seeing people who are building things that you can use it and you know like wow all of the details of this are so spot on like thinking about the feeling the customer's gonna have when using it has been, just been done so well and it's the same person who was doing stuff before but couldn't get it translated and so it definitely is rewarding a different type of work and a different type of
Speaker 1thinking yeah cost
Speaker 3how much does this thing cost
Speaker 1i think i mean our team is we're like around 140 people the wizard run rate is like 30 grand a month i want to say right now i think it should be more i think it should be a lot more i think it's it is being underutilized uh currently across
Speaker 3the employee base but the the yeah the but the cost per month has not changed that much because like the models have gotten better at such a rate and the price keeps dropping that even though like if our actual are looking at an absolute basis like usage is up tremendously yeah if the cost has not changed that much so i i mean it would be weird for that to change at this point i think it's not going to suddenly you they're not gonna be able to make it cheaper like no i think it's the
Speaker 2opposite and has it replaced tools and solutions that you've had yes oh yeah yeah okay and i
Speaker 1think there's more there's like there's more to come because again having all that context it is like hard for in some cases for external tools to match that level of context that it has now but we don't want to build every you know like i think there was some in the early part of ai like at the full saspocalypse like everything is gone like there's plenty of stuff that is like a silly for us to build it's like a nice match for this way of working but there are some things that are you're like that's like a it will be better for us in the wizard um and it and we don't use we don't do that much with it currently
Speaker 2yeah but what about total cost of ownership like the team that you have devoted to this what does that look like yeah it
Speaker 1it's not inexpensive so the wizard they're like three or four folks full-time working on that so like call it like a million bucks a year um i'm developing on it
Speaker 2so it's an investment but you know just to uh close this off here chris in exchange for that investment obviously there's a high a huge roi what you guys have described how would you summarize the outcomes that this has driven i mean the outcomes
Speaker 3are just tremendous um ridiculous ridiculous impact in terms of quality of the product scope of what we can go after uh bottlenecks that have been removed you know i kept i think one of the things that i was saying to people last week on wednesday that i wasn't really sure how they're going to react to i'm still i'm still saying because i think it's the truth which is like we need to start working like we have unlimited resources which sounds kind of crazy to say i know um and they were like yeah we do we need to start working like that or we are working like that because the models are getting so good and when you have the operating system of the business when it's ai enabled when the model gets better everything just gets better and so the actual the hard thing is again shifting like what you ask for like asking for more and asking for it faster and getting used to that which is very you know we've been doing stuff for a long time we're both very used to like sure you're used to the same thing hey we're gonna kick off this new project everyone's like cool what is the new product okay when will it launch we're hoping it's gonna launch in like 12 months and then it's going to be good in like three years in and that you know what i mean like it's like these very very long timelines that we're used to work on and this is just requiring a a shift there and i think like but i think that speaks for the results like you can just see them you know i'm
Speaker 2just really happy for you guys when i think of companies that have gone through the best ai transformation so far wistia honestly is at the top of my list you guys finn has done a wonderful job ramps done a wonderful job there's a handful of them but after i've spoken to both you guys about this several times i'm more and more convinced and impressed with what you've really done to incorporate it into the fabric of the entire company and leading by example in that so chris savage brennan schwartz founders of wistia an amazing creator platform go play with mosheen not musheen mosheen these guys just gave me a demo and it is awesome and i can't wait to use that for my social clips and podcast clips guys thanks very much so much for being on the program
Speaker 1thanks josh yeah thanks josh you

Podcast Summary

Key Points:

  1. Wistia's "Wizard" is a single shared AI agent, built in-house on the open-source Pi harness, that is connected to the company's internal systems and can see and do what a human could inside Wistia.
  2. The Wizard can only be used in public Slack channels, a deliberate cultural choice so everyone learns from and builds on each other's work rather than splintering into private channels.
  3. The tool originated from a February hackathon that grew out of Brendan Schwartz's experiments with OpenClaw, which he used to ship code from his phone and even from a ski lift.
  4. Before the Wizard, Wistia had already shifted its culture in 2022–2023 to ship something every two weeks, but AI adoption initially widened the gap between engineering and go-to-market teams.
  5. Automated "software factories" now run continuous work such as scanning production logs for bugs, click-testing for UI and performance issues, and auto-merging PRs; one factory shipped over 100 PRs in two weeks, mostly autonomously.
  6. The Wizard's centralized context lets it catch things humans miss, such as flagging contacts who should not be emailed because they already have meetings booked with CSMs or account managers.
  7. The biggest organizational bottleneck has shifted from execution to decision-making, and Wistia reorganized so product and go-to-market teams work together as fully dedicated units.
  8. The Wizard costs roughly $30,000 per month to run, plus three to four full-time staff (about $1 million per year), with usage rising sharply while costs stay flat as model prices fall.

Summary:

This episode of Unchurned features host Josh Schachter interviewing Wistia co-founders Chris Savage (CEO) and Brendan Schwartz (CTO) about their company-wide AI transformation. Wistia had already improved its shipping cadence in 2022–2023 by abandoning long-term roadmaps and requiring product teams to ship every two weeks. When AI coding tools surged in late 2025, engineering and product pulled further ahead of go-to-market, prompting a February hackathon that produced the precursor to the Wizard, an in-house agent built on the open-source Pi harness after early OpenClaw experiments raised security concerns.

The Wizard is one shared agent connected to Wistia's internal systems, usable only in public Slack channels so knowledge spreads and coordination bottlenecks disappear. It can build web apps called artifacts, answer questions using sales, support, and wiki data, and even predict how executives would approach a problem. Automated software factories handle bug fixes, click-testing, and auto-merging PRs, with one shipping over 100 PRs in two weeks. The founders acknowledge cultural challenges: public work feels unnatural, some craft-focused engineers struggle, and burnout from constant decision-making is a real risk. They also note guardrails around PII and customer data. The Wizard costs about $30,000 monthly plus a small dedicated team, and Wistia has reorganized so product and marketing teams are tightly integrated. The founders urge companies to operate as if they have unlimited resources.

FAQs

The Wizard is a single shared AI agent that everyone at Wistia uses, hooked into all internal systems and accessed through public Slack channels. It can see and do what a human could do inside Wistia, including answering questions, building pages, and writing code.

Wistia deliberately requires public channels so everyone can learn from each other's work and discoveries spread quickly. This also removes coordination bottlenecks and lets people build on each other's ideas instead of relying on meetings to share information.

It can segment audiences, draft and send outbound emails, build web pages, answer performance questions, and flag people who shouldn't be contacted based on company context. It pulls from sales calls, support interactions, wikis, and other data sources to give real-time, context-aware answers.

The Wizard's run rate is roughly $30,000 per month for a company of around 140 people, plus three to four full-time engineers building on it (about $1 million a year). Usage has grown tremendously while costs have stayed relatively flat because model prices keep dropping.

Working so openly is unnatural for many people and has been the biggest cultural challenge. Wistia purposely keeps sensitive employee and customer information out of the Wizard, and it redacts PII to protect customer data.

Wistia runs automated 'software factories' that use production logs and click testing to find bugs and performance issues, then open PRs autonomously—one shipped over 100 PRs in two weeks. The bottleneck has shifted from execution to decision-making.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.