From Copilot CLI to Cloud Agents: How Brittany Ellich Builds at GitHub
35 minAudio
In this episode, I sat down with Brittany Ellich, Staff Engineer at GitHub, to talk about what software engineering looks like when AI is part of nearly every step of the workflow. We got into her path from a smaller medical education company to GitHub, what it means to work on enterprise billing and metered usage, and why billing systems are such a fascinating engineering problem: speed matters less than accuracy, and in distributed systems that’s a surprisingly hard thing to optimize for. We also explored how GitHub’s different cloud and enterprise offerings complicate that work, especially when data residency and product flexibility enter the picture. From there, we went deep on AI-assisted development: async vs. sync coding, delegating work to cloud agents, using GitHub Copilot CLI and VS Code in agent mode, and how tools like Spec Kit and SQUADS are changing the way features get planned and built. Brittany shared how AI has shifted her role toward product thinking, architecture, and review, and we talked honestly about trust, pride, technical debt, junior engineers, and where the profession may be heading next. It’s a thoughtful, grounded conversation about adapting to a fast-moving moment in software without losing the joy of building. Brittany Ellich website: https://brittanyellich.com/ GitHub: https://github.com/ Microsoft: https://www.microsoft.com/en-us GitHub Actions: https://github.com/features/actions GitHub Enterprise: https://github.com/enterprise GitHub Enterprise Server docs: https://docs.github.com/en/enterprise-server@3.17/admin/overview/about-github-enterprise-server GitHub Copilot: https://github.com/features/copilot GitHub Copilot CLI: https://github.com/features/copilot/cli GitHub Copilot CLI getting started docs: https://docs.github.com/copilot/how-tos/copilot-cli/cli-getting-started Visual Studio Code: https://code.visualstudio.com/ OpenAI Whisper: https://openai.com/index/whisper/ GitHub Spec Kit: https://github.com/github/spec-kit Spec Kit documentation: https://github.github.io/spec-kit/ SQUADS CLI: https://squads.sh/en/docs/cli CascadiaJS: https://cascadiajs.com/ Craig Dennis talk, “AIdentity Crisis: Can I be Proud of This?”: https://cascadiajs.com/2025/talks/aidentity-crisis-can-i-be-proud-of-this Hacker News: https://news.ycombinator.com/ Rust programming language: https://rust-lang.org/ React: https://react.dev/ Stack Overflow: https://stackoverflow.com/questions LeetCode: https://leetcode.com/ OpenAI: https://openai.com/ Gemma: https://deepmind.google/models/gemma/ Gemma 4: https://deepmind.google/models/gemma/gemma-4/ Anthropic Claude Mythos: https://www.anthropic.com/claude/mythos
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
Barry LuijbregtsOkay. Looks like it's going. It is. So hi. Welcome.
Brittany EllichThank you. Yeah. Thank you so much. This
Barry Luijbregtsis the first time that we meet. So maybe I'll introduce myself a little bit so that you know who I am. I'm Barry. I'm a cloud solution architect working for Microsoft, The Netherlands. I just started there in March. It's now April, but it feels like a year ago because it's well, you you work for a big company. You know how it goes.
Brittany EllichYeah. Everything changes fast.
Barry LuijbregtsYeah. It's crazy. Learn so much, do so much stuff, see so many customers. It's it's lots of fun. So maybe you can take it from there and tell us who you are.
Brittany EllichYeah. Absolutely. My name is Brittany Ellich. Thank you so much for chatting with me. This is exciting. I am a staff engineer at GitHub. I am currently working in the enterprise space, and I'm very just passionate about using AI to make things work better. I think that, you know, if you're an engineer right now, this is a really this is, like, the exciting space to be in. So yeah. Isn't it cool? Yeah.
Barry LuijbregtsIt's it's so cool. It's just such an exciting time. That's also why I kinda joined Microsoft now because this is the most exciting time I've seen since I've been working in software. It's just so much fun. So how did you end up at GitHub?
Brittany EllichYeah. I've been at GitHub for almost four years now. And I was previously working for a as a software engineer at a medical education company. It was pretty small. And I've done I've done a couple of other gigs too. And, yeah, I was very excited about GitHub to get into developer tools. And that was my first foray into it, and I realized that I actually really enjoy working on developer tools. And I like GitHub specifically because the product is very engineering driven, and so I feel like I get to sort of flex my, like, product skills in addition to my engineering skills, which is really fun. So I worked on GitHub actions for the first two years, and now I am in the enterprise space.
Barry LuijbregtsAh, what does that mean in the enterprise space?
Brittany EllichI work in billing, which is, like, it's just every it's one of the things that falls under under enterprise. And so, like, looking at things like enterprise licensing and subscriptions and things like that. I specifically do most of my work on metered billing, which means, like, the usage based billing, like, Copilot requests and actions things and stuff. So, like, each one of those needs a line item to bill for it, and that's that's what I work on.
Barry LuijbregtsAh, okay. And and how do you do that? Do you use AI?
Brittany EllichOh, I mean, we use AI quite a bit for development, definitely. But one of the things that's kind of interesting about working in billing, like, in a lot of other roles that I've had, you really worry about things like speed or, you know, some other engineering metric. Like, it's the most important thing, and you're always driving towards that. Whereas in billing, the the speed at which people get billed is a little bit less important, and it's more about accuracy. So that's sort of the thing that you're always optimizing for is trying to make sure that the
amount that people are billed for is as accurate as possible. And that gets really complicated when you have, you know, many, many, many millions of event all the time. Yeah. It's hard to do, but it's fun.
Barry LuijbregtsYeah. I can imagine. That is such an interesting space, though. Because, yeah, it doesn't really matter if it's a minute later that you show it. It just needs to be super duper accurate.
Brittany EllichYep. Yeah. Yep. Which is really hard in a distributed system, actually, to optimize for accuracy because there are so many things that can go wrong and so many ways that you can miss. Usually, if something goes wrong, then you on the side of underbilling. But, like, you also don't wanna do that because that means that your company's making less money. So avoiding overbilling is a lot a much easier problem than avoiding underbilling.
Barry LuijbregtsUh-huh. Yeah. So if you make a mistake and you underbill, then
Brittany EllichGitHub can lose millions. Yeah. It's true. Yeah. We try not Yeah. To do No pressure. Yeah. It's it's pretty interesting too. I like the fact that there's a lot of different customers for it. Like, it's not just like the customer, which is like the GitHub user, but it's also you have like the product customer where like the actions team or the copilot team or whatever, we're always trying to optimize for their experience so that they don't have to think about billing. Ideally, as like a platform, that's something you wanna optimize for,
as well as, you know, the support team. And then we have, like, the business side of things where we have to make sure that we have accurate data for the business to be able to make decisions. So it's a pretty interesting role, I think, from that perspective.
Barry LuijbregtsAnd then how does that work? Because GitHub is a is a SaaS platform. Right? It runs in the cloud, GitHub Cloud, GitHub Enterprise Cloud, and there's different clouds as well. Right? Like data residency, as in GitHub in Europe or in different places. And you can also host it yourself. Right? Yes. Yeah. Yeah. There's a lot of different customization options,
Brittany Ellichwhich is the other fun side of the billing coin. I feel like it's like all of it is edge cases. Everything is an edge Yeah. So lots of different ways that people can have their their enterprise set up because lots of different enterprises are different, and business needs are different. So
Barry LuijbregtsSo is your billing engine then all the same for all these products, or do you use different, let's say, versions?
Brittany EllichThey're more or less the same. I think that billing works a little bit differently depending on what people are using. Like, I think server works differently, and we don't necessarily connect the billing side to enterprise server. So I think that's part of, like, the the benefit is that people run it on their own equipment. And then we have different versions of our billing system for different data residencies. That way, like, the data that is generated in Europe stays in Europe. Same with everywhere that there's data residency.
That way that way, it all is where it needs to be.
Barry LuijbregtsYeah. Complicated. Yeah. But fun. Yeah. That's you fun. Still write code, or do you just talk to an agent all day?
Brittany EllichThat's a good question. And I feel like that's something that's changing, gosh, so rapidly. I think most of what I do now the way that I kind of divided in my head is like there's asynchronous development work that I'm doing, and then there's synchronous development work that I'm doing. And so I typically have quite a few things going on async where I'm like, just just saying like, hey, like, Cloud Agent, do this for me. And then I'm like delegating tasks out to a Cloud Agent. And I try to not take on more than like three or four of those at once because then it just ends up being really hard to keep up with all of the code that gets generated.
And then locally, then I work in Versus Code or with Copilot CLI to do that synchronous work. And that's usually like more difficult things or things that I haven't quite figured out yet. So I don't have all of the knowledge that I need yet to go and delegate that task. So those are the synchronous things that I typically work on. And usually, those I try to do, like, one thing at a time when possible.
Barry LuijbregtsYeah. Because now you you get all these little tasks that you delegate everywhere, and then it kind of fills your whole memory space, your your own memory space. It's difficult to keep track of everything.
Brittany EllichYeah. That's a really hard problem right now, I think, is just, like, dealing with the the human brain part of, like, what am I working working on? On? And how do I get all of these things done instead of just started? It's very easy to start things with AI. It can be very difficult to make sure that they all get finished.
Barry LuijbregtsAnd then when you just said asynchronous stuff, do you mean you ask GitHub Copilot agents to think that it is in the cloud on github.com to do something to create a PR for you? Is that what you mean? Yeah. Yeah. So I
Brittany EllichI'm typically doing something like that for, like, small bug fixes or things like my general rule of thumb when I'm like, is this a thing that I'm going to do? Or is this a thing that Coding Legion is gonna do? Is like, can do I know exactly how to do this thing? And if I do, then I just make sure that I write everything up. I use Whisper right now. That's my, like, tool of choice for transcribing everything very quickly. And then use Copilot to, like, make sure everything is there in the task and then assign it to assign it to Copilot to actually go and do the work. Whereas if it's a thing that I still feel like I need to explore or I'm not quite sure what the approach is, then I'll do it myself.
In in, like, in an IDE working with coding the other agent. There's so many agents now. It is such an overloaded term.
Barry LuijbregtsIt is a very yeah. In in agent mode. Yeah. Yeah. Yeah. It is very confusing because in Versus Code, you have your little chat window and there is agent mode. But really it's just it's different modes that have different tools and different permissions, and then they call those participants as well. And you can have all sorts of specific ones. It's very confusing. Basically, it's just a text box that you just talk to. Yeah. And it just does stuff for you. Yeah. And I would say, I think, like, a 100% of my work now is with
Brittany EllichAI doing it either synchronously or asynchronously. There's occasionally a time where, like, if it's like, oh, like, there's a CSS problem. I'm gonna run-in and, like, fix that really quick because it'll probably be done faster by me than with an agent or, you know, some, like, manual text changes or something. But everything else is pretty much all done now with AI, which is weird.
Barry LuijbregtsIt's very strange. Yeah. And and so when you work on, let's say, a new feature or something, do you plan it out yourself and you go, oh, I need this file and this change here and this thing here? Or do you just say, hey, agent. I just I need this feature. Go plan it.
Brittany EllichDepends on the feature. If it's something that's, like, very large, then I will typically plan that out like in advance. And then figure out how to break it into smaller pieces so that it's not all just like one shot at and done at once. Because I feel like that's sort of where AI still sort of breaks down is like, yeah. If not the AI breaking down, but in like the review process, then, then I'll break it into smaller pieces. Whereas if it's, you know, just a small discrete chunk of work that needs to be done,
then I'll typically use, like, the plan mode first to figure out exactly how I want it done and then run agent mode to do the actual work.
Barry LuijbregtsAh, okay. And and not
Brittany Ellichspec driven development or something like that? Yes. Yeah. I think that's what it's called. I have a a on my very long list of things that I need to look into, spec kit is on there. And I I Yeah. Yeah. Would like to. I did play with squad over the weekend recently, and that was really, really cool
Barry Luijbregtsand very interesting. Explain what that is to the listeners?
Brittany EllichYeah. So it is a group of it's like a group of agents that you can set up and work all together. So you'll it'll it's really easy to install. I think you just, like, NPM install it everywhere and then do squad and knit within Copilot CLI. And then it'll go through and, staff your group of folks. So I think I had, like, a front end developer, a back end developer, like, a tech lead that was overseeing everything, and, like, a test engineer.
Yeah. I think it was, like, five or six different roles. But then they all work together and make decisions and sort of like play back and forth with each other to, you know, write something very quickly. And so I was doing, like, just like a a fun project on my own just to, like, try this out and, like, solve a small problem for myself. And I used Squad for the first time to do that. And I realized, like, that was a really great way to, like, get a lot of stuff going at once and sort of, you know, like, scaffold something into existence that then I could take and, like, shape into what I wanted it to be.
And it's like a more less less crazy, I guess, the way where there weren't as many things happening at once because it was hard to also keep track of all the changes happening at once. But I didn't really need to keep track of them until I was ready to, like, actually shape it into what I wanted it to be.
Barry LuijbregtsYeah. And and do those agents then work truly autonomously, or do you still need to ask them things?
Brittany EllichYeah. So it seemed like the way that it worked is, like, copilot the Copilot CLI would come back and they went and, like, created I I created a spec markdown file and then a design markdown file, and they pretty much just worked off of that and came back and divided that into a couple of different chunks of what to do. And then Copilot would come back to me and be like, alright. They're done. Should we move on to the next one, or do you wanna smoke test the the changes? And so I would, like, go test the changes and come up with, you know, feedback for them on how what I wanted to change from there.
Yeah.
Barry LuijbregtsYeah. So it's it's kinda local offshoring. You have a little team. Yep. They just say, go do this, and then they just go do that. Yeah. They just go do it. It was crazy.
Brittany EllichAnd it was yeah. It's very cool, and it makes me realize, like, just how many things can be done now because I've been a software engineer for long enough that I have like a very long list had a very long list of side projects or ideas on what I wanted to do. And that would usually die at the point where I, you know, start thinking about breaking down the tasks or go buy a domain. Got a massive collection of domains that haven't done anything. And now that's not the you know, doing the work isn't really the bottleneck. Like, you can go from idea to a working prototype in, you know, a couple hours, which is really cool.
Barry LuijbregtsYeah. That's the thing now, isn't it? It's the hardest thing is now you need to figure out what you wanna make. Yes. What is worth making?
Brittany EllichYeah. Yeah. And yeah. I mean, all of those things are not free necessarily to, like, deploy. So now you have to make the decision of, like, what what deserves to exist in my, you know, in my droplet somewhere or for me to actually pay for. And that's kind of an interesting change where I feel like a lot of engineers are having to sort of step into this, like, product management role in a way that they haven't had to in the past.
Barry LuijbregtsI think now you need to think about value way earlier. Because previously, when you had a cool idea, you buy a domain name and you go build a thing because that's the fun. You build a thing. It might not be super valuable, but you keep building it and building it. Maybe you push it and then you think, meh. It's a bit meh.
Brittany EllichYep. But now you just don't bother. Yep. Yep. It's true. Yeah. I feel like it's sort of changed the way that I think about about building things
Barry Luijbregtsfor sure. Yeah. So so squad, would you trust it for your daily work?
Brittany EllichDon't know. I don't think that I would run it on, like, on my work that I'm working on in production. But that's also in part because I don't think that I think it's good for large changes that need to be made. Not so much for, like, you know, a lot of the work that I do is like, oh, there's a small bug fix or a small part of this feature that I'm going to add or something like that that typically it just doesn't really require, like, that much effort going into it. I should try it. I think it would be fun. I mean, that's the beauty of git repository. Right? It's like, you could just run it and see how it does.
And see if it just one shots that feature. You know? Yeah. You don't have to push it to production. Yeah. It's true. You don't have to push it. You can just experiment and see what it's like. I feel like that's sort of been like my lesson with doing AI things is like a lot of my discomfort with like what I would actually do with it is like very much on me. And it's not necessarily on what they can do. It's like, oh, have I tried running squad on, you know, in on my, like, normal day to day code base. Like, I'm sure once I did it, I would be actually very comfortable with it and realize like, oh, there's so much more. But then that comes with, like, the existential crisis of
who am I? What am I doing? Like, all of those things that come up with AI as well. So yeah.
Barry LuijbregtsYeah. I found in the beginning when this started to work, and I'm talking like six months back or something because before that, it wasn't that great at coding. Was good. But, you know, I had a little bit of shame when I pushed something into production that I didn't write myself. It's like, come on. I'm cheating. It's like copy pasting from Stack Overflow or something. It's not not right. Just it doesn't feel right. Mhmm. But now I'm very comfortable with it. It takes a while.
Brittany EllichYeah. So Yeah. Yeah. I think it takes a while to build that muscle and, like I think and there was a really good talk last year at Cascadia JS that I really liked from Craig Dennis too that he did about, I feel like I recommended a lot. That's about, can you be proud of the things that AI builds for you? Yeah. And it was a really interesting viewpoint, I think, where it's like, can you? I mean, it's still your ideas that made Yeah. Still
requires typically, anyway, it requires a lot of shaping. Maybe it didn't. Maybe it didn't require a lot of, like, expertise or interest. But I still think it's really cool when something like that happens. So I land more on the side of, yes. Yes. You can be proud of it. But
Barry LuijbregtsYeah. And you need to learn to trust it a little bit as well. Mhmm. Like, you know, if you would run Squad in production or for production code. And so for for instance, do you still check everything that it does? All the code?
Brittany EllichYeah. I would say the vast majority of things, but I feel like I'm pretty comfortable now just being an engineer in general of knowing, like, what parts are the important parts that need to absolutely work and need unit tests and what parts are a little bit less, you know, like, it can it could do what it needs to do, and it'll be fine if, like, they use a, you know, a slightly different front end component or it's not built exactly how I would imagine it being built. So Yeah. I would say that I still check most of it.
Barry LuijbregtsYeah. Exactly. Your CSS won't cost GitHub millions.
Brittany EllichExactly. Yeah.
Barry LuijbregtsYeah. Yeah. So where do you hope this is going for yourself or productivity?
Brittany EllichOh, good question. I think I this is April, mid April, and the recent mention of mythos is something I've been keeping an eye on because I'm like, oh, it looks like we hit yet another, you know, step change in how how good AI is. So I'm curious to see how that changes things. I lean on the side of, like, leaning in to all of it. Like, there's nothing that I can individually to say like, hey, maybe maybe we should pump the brakes on this AI thing. Like, it's gonna happen regardless.
So I would rather learn how to use it and do well with it. Because I think that, like, if you're going to be an engineer, like, five or even a year from now, like, has to be you have to you have to use it, and you have to use it well because the folks that are doing it are going to be so much faster and, than the ones that aren't. So I try to keep up with it as much as possible and continue using it. But I also I don't know. I think I also appreciate the viewpoints of folks that are like, but I liked writing the code.
For me, it was never really like part of my identity be identity to be like a good coder or really good like, LeetCode. Like, gosh, I cannot I can think of a million other things I'd rather do than do LeetCode problems. I really like building things, and now I can build things so much faster and better and easier than I was able to before. And I think that part's exciting. But I think a lot of folks probably struggle with the fact that, like, things are different now.
Barry LuijbregtsYeah. Things are different. But like you say, you can add value way more quickly now. Yeah. Yeah. That's true. Yeah, you just need to, I think, develop different muscles. Mhmm. Like, you become more of a product owner. Mhmm. And architect and, you know, notch the AI into, no. No. You should look into this or it needs to be scalable. It shouldn't cost 10,000 to to run or something. Yeah. So
Brittany EllichYeah. I think I've seen some recent discourse about how, like, the role of engineer and product manager are sort of merging together to just be, like, a builder of things. And I feel like I have a product manager that I work with regularly at GitHub, and I think that we are both, like, converging on this same similar role where I'm, like, helping her become an engineer, and she's helping me become a product manager. And I feel like we're sort of converging on this one thing that we both end up doing. So yeah. Yeah. That's pretty cool. Yeah. Because, obviously, non tech people can now do tech stuff as well. They just talk into the box and stuff happens. Yeah. Why not? Yeah. And and do you think that
Barry Luijbregtswell, you kinda alluded to it already, but do you think everybody will still be able to be an engineer? Or will it just be a couple of folks because, you know, people can do so much more now?
Brittany EllichYeah. Yeah. I kinda lean on the side where I think that there's so much demand for software that I don't think it's going away. Even in a role right now where I am building with AI very regularly and building faster than I ever have before, our backlog has not gotten any smaller. Because the more things you build, the more things you have to maintain. And the more things change, and the more things people want. And there's also, I think, a story to be told about, like, there are the right features. And you have to figure out what those are and not necessarily build everything. So I think there's enough work to be done in the world of software that I'm not personally worried about that.
But I could see that it would be potentially difficult to get a job anywhere as an engineer if you're not leaning into, learning those AI skills. Because the folks that are doing it are gonna be, you know, five or six times more productive than you would be without it. So yeah.
Barry LuijbregtsYeah. And I yeah. It's it's such a strange world now. Like, have two kids. One is 10 years old already. And I don't know. A year ago, if you would ask me what should she go and do and learn, I would say, you know, software development or something like that. It's good for problem solving and, you know, you'll probably find a job. But now, I'm not so sure if you could do that as like an entry level person. I don't know. Yeah. What would what would you advise?
Brittany EllichYeah. I think that the story for junior engineers is probably really difficult right now. I mean, I know it is. Everybody that I talked to that's like early on in their career is like, what is happening? Why why is this Yeah. Is this normal? No. It's different. It's completely different than it ever has been before. But I like I said, I do think that there's still gonna be a demand for software going forward. And even if it's not necessarily like your full time role, I could see that having those skills to build software, I think, is gonna be important for a lot of roles in the future even if it's not like
being a software engineer. So I still think it's worth doing. I think that we're going to come up with, like, a potential, crisis here in, like, five years if we don't still have junior engineers, because you can't just, like, come out of the gate as a senior engineer. Like, a lot of it is a lot of your job early on is knowing how learning how to build software because, you know, the boot camps and the even the degree programs don't really teach that. They teach, like, this is how computers work. I'm like, this is how you write React code or something like that. But Yeah. Yeah. The actual, like, act of engineering
software is something that you really only learn on the job. And so if we're not investing in those skills for folks early in career, then I think we're gonna have some problems a few years from now. The whole world runs on software. So like, I don't feel like that's gonna go away and anytime soon, barring any sort of like major disaster. And so, yeah, I I think it'll still be a good career to go into. But I'm also wondering the same thing. My oldest is five, and she's like, when I grow up, I'm gonna be a software worker like my mom. I'm like,
maybe. I guess we'll see.
Barry LuijbregtsYeah. Yeah. Nobody knows. Yeah. I I think we're moving up the stack way more and more and more. And and that was already the case. Like, juniors now, you know, they use front end frameworks and they think that's like the the bottom of the stack. They have no idea. And that's fine too. That's fine. That's just how you're raised and how your career progresses. But now, the stack is very different. You just talk into a box, you type something in there, and then something happens. You don't need to really even see the code.
And this is the worst it's gonna be. Like, a year from now, can you imagine? Mythos like something? Do we even need to have c sharp anymore? I don't know. Maybe the thing just makes something that just compiles and runs.
Brittany EllichYeah. That's true. Yeah. And I think there's a lot of talk to you about, like, languages that are, you know, optimized in one way or another. And I think people might lean more towards those, and towards rewrites in, you know, maybe it does make sense of what the Rust stations have been saying for a while to just rewrite everything in Rust because that's that's the language that's apparently the in the meta right now. Yeah. I I'm curious to see how it goes. I think it's really uncomfortable
not knowing what the future looks like. And I think it's like a good reminder that, like, nobody does, even the folks predicting everything. Like, nobody really knows. And we're all uncomfortable at the same time. So, like, we're all in this boat together, I guess.
Barry LuijbregtsBut But let's let's let's just choose to be optimistic. Yeah. Because why not? I think also a part of the optimism may maybe is all this stuff is currently is written by humans still, kinda. There's so much code written by humans, and that means that we have loads of technical debt. And that can't be solved just in a couple of seconds by AI. That doesn't work like that. Mhmm. So I think there's still lots lots of work to do for us as humans.
Brittany EllichYeah. Yeah. I mean, even now with, like, static site generators existing and stuff, there's still, like, demand for so many, you know, small businesses that aren't online yet. And so, like, I think there's gonna be plenty of stuff to do just to get software enablement out there for everybody and to maintain it and to keep it going for a long time. So, yeah. I think it's still gonna be around for a while. I'm optimistic. Yeah.
Barry LuijbregtsGood. Me too. So as you work with with billing and metering and also with the GitHub Copilot premium request units, you probably work on those or see those or something. I do. Yes. Do you see those being used a lot or overused a lot or that people run into limits?
Brittany EllichI think it kinda depends. It seems like in a lot of places, there's, like, a lot of power users and then, like, a trail like, a big long trail of, like, folks that don't use it as much within, like, places. And I feel like that's the only trend I've really noticed is like, there's some people using a ton of AI and then less as you get on the stack, which which I think is interesting. I actually I look at my own usage quite a bit just because this is a space that I'm in and I'm very interested in And I feel like I use a lot of AI, but I really don't.
You know, I compared to a lot of the folks, maybe that's just the way that GitHub is run where there's just a ton of folks using a ton of AI. But I personally don't use a ton, so I don't really know how to, like, how to gauge. Is there somebody that uses a lot and that's the way to do it? I think there's been, like, some talk online about how, like, oh, like, if you aren't using your salary and tokens every month, like, are you even an engineer? Don't think I use, like, an extra $100 or something like that in usage beyond my normal
amount. Oh, great. Yeah. That seems like a very reasonable trade off for the amount of productivity I feel like I gain out of it. So I don't think that that discourse is gonna be around forever where everybody's like, oh, you gotta use as much as you can, and you're gonna be the most productive engineer. I don't think that there's, like, a linear relationship between token usage and productivity from what I can see. But No. No. Tokens are expensive. They are. Yeah. Yeah. I'm curious to see when Mythos comes out. I feel like it's probably maybe it's gonna be more expensive. I don't know. I don't know.
That's what I'm most excited to see soon.
Barry LuijbregtsYeah. I think it's an OpenAI. They also have something similar. Right? That's coming up. Don't remember the the code name. It it was an interesting code name. I don't know. Mhmm. I'm sure they have it already, all all this stuff, because once one company drops one,
Brittany Ellichday after the other one does it. Yeah. So it's gonna be wild. Yeah. What else I think is interesting is it seems like recently there's been a lot of talk about local models actually, like, getting good. It seems like they're maybe, like, six months to a year behind, like, the frontier models, which is actually great because six months ago, development got a lot better. And so if local models are hitting that inflection point as well, then maybe we can you know, maybe the conversation about the cost and everything like that is changing soon. We'll see
Barry Luijbregtshow that changes. Yeah. That'll be interesting for GitHub Copilot as well because now with Copilot CLI, it can run local models since a couple days. And there's now Gemma four, for instance, Google Gemma four, which runs locally. Haven't tried it yet if it's fast enough or capable enough, but that'll be interesting. Yeah. Like, how will billing go? What what's the value of Copilot? Just a harness? Or
Brittany EllichYeah. Yeah. I think that the there's going to be more changes than we anticipate as things actually settle out. Like I said, nobody knows what the future is going to be.
Barry LuijbregtsNo. No. Exciting. Okay. So what are you excited about working on next?
Brittany EllichYeah. I think I've recently so I think up until now, I've really been using AI tools more as like, I need to create this feature or fix this bug or something like that. And I really wanna spend more time using them, for refactoring. I think that that's like one of the bigger opportunities is like how do we now refactor a lot of this existing code to like optimize for how, you know, the agent would build this thing. So that it theoretically works better when, you know, Copilot or something like that is working on it.
Yeah. I think that's gonna be an interesting space both with, like, these more advanced models coming out that are, you know, able to find all of those security exploits. You know, it sounds like we're supposed to be rewriting all software in the next few months to avoid that. But also to see, you know, I think that a lot of these models are good enough now to to take the wheel for the most part on building software. And so what does that mean for all of the existing code and all of the tech debt that's out there already that we need to go handle and deal with? So I'm interested to do more refactoring with Copilot and see how that changes things
with my own workflow. Yeah. And get rid of some technical debt and stuff. Yeah. Mhmm. I would love to see the end of a backlog one day. That sounds nice. I don't think that it could happen. Like I said, I think that, you know, you never get to the end of the to do list, but I would love to see the end of a to do list and be like, we did it. We build it all the software.
Barry LuijbregtsWell, you're friends with the product manager, so maybe maybe you can make something happen. That's true. That's true. We can
Brittany Ellichalso get there by just removing everything. And yeah. Yeah. But I think that's gonna be interesting.
Barry LuijbregtsOkay. Very cool. And then maybe finally, is there anything exciting coming up in the GitHub world? Like features or capabilities that we should keep our eye on out for?
Brittany EllichThat's a good question. I don't feel like I'm plugged in enough because a lot of the things that I'm working on right now are not shared publicly. So probably. And you'll have to stay tuned to see how those actually work out. Cliffhanger. Yeah. Cliffhanger. Great things are coming.
Barry LuijbregtsOkay. Exciting. Yeah. We'll keep an eye out for that. Yep. Excellent. Well, thank you very much for talking with me today. This was fun. Yeah. Thanks for go and and find out more about you?
Brittany EllichYeah. So I have a website that sort of aggregates all of the things on my socials and all that, and it is britneyelick.com.
Barry LuijbregtsExcellent. Cool. And from there, they can also find your podcast.
Brittany EllichYes. Yeah. I've got a podcast where I chat with some of my some of my engineering friends about folks that are very dedicated and excited about the world of software engineering. I write online too, write some blogs. I just got on the front page of Hacker News the other day for the first time, which was like a career milestone. So I'm online a lot. Probably more than I should be. Yeah. Thank you.
Barry LuijbregtsOkay.
Brittany EllichThank you very much. Yeah. Thank you for having me.
In this episode, I sat down with Albert Tanure, a Senior Cloud Solution Architect at Microsoft, to talk about what AI is really changing in day-to-day engineering work - and what still matters just as much as ever. We...
In this episode, I sat down with Heena Refai, a Cloud Solution Architect at Microsoft, to talk about how enterprise AI has evolved from the big-data and traditional machine learning era into today’s generative AI and...
In this episode, I sat down with Rob Bos to talk about what’s happening across GitHub, Azure DevOps, GitHub Advanced Security, and the fast-moving world of GitHub Copilot. Rob shared how his work evolved from DevOps...
In this episode, I sat down with Anton Sizikov from GitHub to talk about how AI coding tools have evolved from “smart autocomplete” into full-blown agentic workflows. We unpacked the early days of GitHub Copilot, when...
In this episode, I sat down with Jon Galloway to talk about what agentic development actually looks like in practice inside Microsoft and out in the real world. We explored how AI-assisted software development is...