64: Coding Interviews in the AI Era | Beyond LeetCode for Career Growth

Brittany Ellich (00:00) Welcome to the Overcommitted Podcast, your weekly dose of real engineering conversations. I’m your host, Brittany, and today I am joined by

Bethany (00:09) Hey, I’m Bethany.

Erika (00:11) Hey I’m Erika

Brittany Ellich (00:13) We met while working on a team at GitHub and realized we are all obsessed with getting better at what we do. So we started this podcast to share what we’ve learned, where you talk about everything from leveling up your technical skills to navigating your professional development, all with the goal of creating a community where engineers can learn and connect.

Today on Overcommitted, it is just the three of us. So I figured let’s kick it off with our normal question for all of us. ⁓ what is something you are building or obsessed with right now? And I’ll start with Bethany.

Bethany (00:46) Man, I am dreading this question because I feel like I I’m not quite sure. I guess if I had to say or my what my mind goes to is the GitHub app just or the GitHub Copilot app just came out. And honestly, this is probably one of the first like not trying to shell or anything, it’s probably one of the first tools that ⁓ truly has changed how I work and how I like

write code so that’s been really fun to play with and experiment with and try to nail down a good workflow. ⁓ but yeah other than that just kinda enjoying the the weather and ⁓ and also trying to ⁓ run a 5k right now. So I actually am doing it.

Brittany Ellich (01:28) Nice! That’s awesome. Cool.

Bethany (01:32) Erica,

how you?

Erika (01:33) Other side of the AI spectrum where I recently tried to like build a skill. like I’ve been I’ve been implementing more like sort of like repeatable AI context. and like it is not as easy as it seems to get it right. Like

First first pass at this like skill generation, it got like everything almost completely backwards. Like there’s like a V1 and V2, and I was like, this is V1, this is V two. And I started testing it. It was like, this is V1. And I’m like, nope, that’s not right. So yeah, I don’t know. I mean, it’s like a good reminder that, you know, don’t take anything without checking it.

‘Cause I think I’ve like gotten pretty comfortable with being like, yeah, that that seems right. so yeah, like very much testing and like, you know, verifying the output of what

So yeah, that’s kind of I I won’t say I’m like obsessed with it because it is like really frustrating. Like I’ve been like manually testing because when I’ve had copilot or whatever like AI I’m using, when I have it like test itself, it’s like looks good.

Brittany Ellich (02:52) Mm, let’s get to me.

Erika (02:54) Yeah, it’s a little bit frustrating and feels kind of manual at this point. If anyone has tips on how to do this better, I’m all ears. ⁓ yeah, but ⁓ I guess it is gonna be more a part of my life. So ⁓ I guess, you know, the obsession is like, I know it’s important. yeah, wanna get it better, get it right. Yeah, what about you, Brittany?

Brittany Ellich (03:16) Yeah, so I ⁓ just switched jobs. So I’ve got a lot that I’m learning, which is what we’re gonna talk about, I guess, throughout throughout this is you know, learning and picking up new jobs. but one of the things that has come with that is when I was at GitHub, I had like free, unlimited usage of Copilot. ⁓ and so that was pretty much all I used for my AI tool of choice because why would I pay for something if it’s free? ⁓

That is no longer the case. So I have been spending some time getting to know a couple of Claude tools, ⁓ the Claude Cowork specifically. ⁓ and then I guess ⁓ Claude Code as well. ⁓ but yeah, Claude Cowork. I some of the one of the things I’ve been obsessed with is I like rewrote, re put up together my obsidian template for like this new job. And I’ve been using cowork to help manage it and like keep my to-do list up to date and like

Help me go work through notes ⁓ from different meetings I’m in. ⁓ and I’ve been very obsessed with with that. I think that’s been really fun. ⁓ I’ve also just been very much just like vibing my way through it. And ⁓ you know, it’s I don’t think that I have anything valid to provide to you, Erica, unfortunately, on like building skills. Cause I’m like, yeah, yeah, that works. Or I’ll forget the skill exists and be like, can you do all these things again? So I still need to like, still need to get there.

All the way, but ⁓ but yeah, it’s been really fun to try out new skills and there are some things that I do miss about copilot too. I’ve realized that I’m like, mm, I really miss like cueing things. Like cueing isn’t really as much of a thing in Claude Code as it is in Copilot, and I used that heavily. ⁓ so yeah. ⁓ anyway, what we’re talking about this week. Yes, recently I switched to jobs. ⁓ things are different.

Erika (04:55) Yeah.

Brittany Ellich (05:05) Now things are weird. I now work at Blue Sky. and I’ve been obnoxious enough about AT proto at Proto in the past several months to now have found myself a job in that ecosystem. So I’m very excited about it. ⁓ and really enjoying it. But I wanted to talk a little bit about like first, like what is it like to switch jobs right now? And also what is it like to like learn how to do a new thing right now? Because

AI exists. And so like a lot of the traditional playbook that I’ve been using to like learn a new code base is very different from what it used to be. So yeah.

Bethany (05:41) Also, we gotta

know how it actually is pronounced, at proto or at proto?

Brittany Ellich (05:46) I honestly don’t know that that’s a good question. I don’t know. I feel like I always say app proto, but I also hear people say AT proto. It’s like gif and gif, like yes is the answer.

Erika (05:56) Does it

does it stand for something?

Brittany Ellich (05:59) yeah. ⁓ so the AT is the authenticated transfer protocol. That’s what it stands for. Mm-hmm.

Erika (06:05) Yeah.

Okay. So maybe like A T is like the formal way and then at is like the cheeky Q way.

Brittany Ellich (06:13) Yeah. Well yeah, ‘cause you always at people as well, like with their handle and they call it the atmosphere and so it’s yeah. It’s yeah, yeah. It’s been really fun so far though. So ⁓ yeah.

Erika (06:16) Yeah.

Yeah.

Yeah.

Well we’re we’re also curious about the job searching ⁓ experience because yeah, neither Bethany nor I have really done that in a while. So I think ⁓ from ever what I’ve heard it’s a lot different now than it was before. So yeah, how

Brittany Ellich (06:40) Yeah.

Bethany (06:45) Yeah, for context,

we we got our jobs in twenty twenty two and the market was very, very different than what it is now. So it’ll be very interesting to hear.

Brittany Ellich (06:52) Yeah.

Yeah. Yeah, I do feel like, especially if you’re interested in startups, I feel like there’s a lot happening right now. ⁓ I know there’s a lot of people that are like, no, the tech layoffs and everything, but I feel like that’s mostly happening in like these big tech companies. And if you’re okay with doing the startup thing, it seems like there’s a lot of opportunities out there, especially if you have experience. I think it’s really hard if you don’t have experience yet. And that’s one of those like tragic catch twenty twos, like, how do I get experience if you don’t let me have experience?

But I do feel like, you know, if you’re like a senior engineer out there looking for a new thing, you’ll probably you’ll probably be able to find a thing. ⁓ but I think the thing that’s really weird is a lot of these companies have traditionally used like leak code type questions ⁓ to like, you know, assess your engineering shops, which I’ve always hated personally because I’m terrible at them. ⁓ but also like it’s

That’s a weird skill set right now, right? Like, I don’t know. There’s very few times where I’m like, all right, I’m just gonna go solve this really hard algorithm for fun. Like that that’s what AI does now. So ⁓ yeah, it’s really weird. ⁓ and I don’t think that everybody has totally caught up yet. I didn’t interview it at like a ton of places, but like I think ubiquitously across the board from places I was talking to, they’re like, yeah, we’re still.

figuring out what this means for interviews because like that’s been like the goalpost I feel like for a while where they’re like, all right, let’s, you know, use this really hard lead code problem that kept getting harder and harder because people would just get like really good at doing lead code. and now I think it’s is worth even less than it was before. ⁓

Erika (08:41) Yeah.

I feel like I don’t know. I I feel like I want to push back on the idea that like that’s not still valuable to a certain degree because you have to know why it works and how it works regardless of who wrote it.

And it can be hard to explain that. you know, even if you get some kind of generated code, like how do you know that it’s working? Like how do you how do you confirm that, you know, it’s taking the right path? and like I feel like there’s room for, you know, write whatever solves the problem, whether you use AI or whether you write it yourself, but then at the end you have to like prove that it’s working and explain to a coworker like what it’s doing.

⁓ and like that could be the interview. You know what I mean? Like, because that’s kind of the job. It’s like if you can’t explain what it’s doing, then like I’m not gonna let you merge it. I’m not gonna approve your PR if you can’t give me a two-sentence description of how your code works.

Bethany (09:47) Okay.

Yeah, I think like my favorite interviews have been ⁓ at least when I was doing the last round was were like PR review pol ⁓ interviews where you had to literally just do a live code review because I think that really fundamentally tests so much of what it means to be a software engineer so that you can read and understand code, you can walk through it, you can leave ⁓ intentional and empathetic feedback, things like that. and it’s also very hard to cheat.

Brittany Ellich (09:56) Mm.

Bethany (10:18) at something like at a PR review one because you’re you’re kind of doing it live ⁓ versus the ⁓ the ⁓ coding one specifically. but I agree. I’ve never quite hated leak code once, even though I do dislike leak code and I don’t think that’s like really what’s valuable in a software engineer. However, it is really good for just getting people to talk and think like they’re problem solving. ⁓

⁓ factor. So if you just take off like, you have to actually solve this algorithm and it’s more like, okay, you’re like brainstorming trying to collaborate ⁓ on a problem you probably don’t know the answer to, then I think that’s a really interesting ⁓ use case. But I think as soon as you say you have to solve this problem in like two minutes, then I don’t think that’s worthwhile at all.

Brittany Ellich (11:07) Mm-hmm. Yeah, agreed. Especially I think a lot of the problem with them is like their folks will just practice them over and over until they like memorize a lot of the patterns and then it’s like it’s not even really problem solving. I guess it is. It’s you know, pattern matching, but that’s a very different skill set than like, can you take a hard problem and solve it? ⁓ I think. But yeah.

Bethany (11:26) also leads

to biases too. Like if somebody has all the time to study those, I think that’s a very different subset of people or a a a pool than people who might not have time to solve those. So I think that’s another thing to factor in there. But totally agree, Brittany.

Erika (11:43) I would also like if I were designing like technical interviews, like I would put in a section where it was like, you need to come up with like three different ways to solve this problem and then argue like the trade-offs of each. ⁓ yeah, because that’s also like very real life, right? Like there’s not only one way to do it. And like it tells a lot about a person if they’re willing to look at multiple ways and like

To me, if you’re able to like, you know, sort of be less emotional about like one direction that you’re taking, you’re gonna be a whole lot easier to work with than somebody who thinks that one way is the right way to do it and it’s the only way and if you don’t do it this way, you’re wrong.

Brittany Ellich (12:29) Mm. Heart agree. I like that.

Erika (12:31) As of yet, I haven’t seen anyone do that. has did anyone do that in your interviews?

Brittany Ellich (12:35) Yeah.

I don’t think so. I feel like that’s like a really common follow-up question though, is like how else could you do this? Like if you solve a thing. ⁓ yeah, the only other thing I can think of that was kind of weird is like ⁓ because ⁓ some places are, you know, open, like you can use AI for everything when you’re doing this interview. but it takes time and there’s like some latency waiting for it. So like that’s kind of awkward to just be like, All right, I’m waiting for my prompt response. Just give me a minute while while Claude’s thinking for me, I guess. ⁓ yeah.

Erika (12:42) Mm, that’s true.

Brittany Ellich (13:07) That’s a little weird, but it’s the times, I guess, that we’re in.

Erika (13:11) So were the like were the questions and the prompts any different or was it was the only difference like the fact that you’re able to use AI?

Brittany Ellich (13:20) Like it’s just you can use AI. Some places are like not cool with you using AI as well. ⁓ which I think is also probably a really good signal right now. Like, can’t you solve a problem when your internet isn’t working? ⁓ yeah. ⁓ yeah, it was it was interesting. I don’t wanna have to do it again for a long time, because job searching is always really stressful. But it was weird and interesting to see what it looks like now.

Erika (13:34) yeah.

Brittany Ellich (13:49) Today. So ⁓

Erika (13:52) Yeah, so you you got the job, you’re there now. So how’s how’s it been? You’ve been there for two weeks?

Brittany Ellich (13:59) Two weeks. Yeah. ⁓ I think one of the things that I was reflecting on is I think about a year ago I wrote this ⁓ article on the GitHub blog. We actually previously talked about it on the podcast, how GitHub engineers learn new code bases. in March thirteenth, twenty twenty five, so a year and three months ago. and like ten years in AI dog ears because things have changed so much since then.

so I thought it’d be interesting to like look at back at like, all right, what are the things that make sense? Like hands-on code exploration, collaborative learning. I’ll make sure we link this. collaborative learning. I feel like that’s a big thing to talk about. Like how pair programming has changed. ⁓ building documentation also has changed. Learning by teaching, I feel like that’s still pretty applicable. but I think those are the four main pillars. It’s like how do you explore code now? ⁓ how do you learn with other people?

⁓ and how you’re building documentation. All through three of those things have changed pretty dramatically, I think. ⁓ so I thought it’d be interesting to just like get takes on how those have changed and what that means now ⁓ in life. ⁓ yeah. ⁓ so the first one I think ⁓ is hands-on code exploration. That one, ⁓

I think we I’d already called out in there that you could use, you know, Copilot to to search through a code base. but it’s even better now than it used to be. almost to the point where like, I don’t know, I don’t feel like I’m gonna I feel like I’m gonna go to that first versus just like spelunking through the different folder. I feel like the first thing I would always look for is like, all right, what’s the folder setup? Like where are the controllers and the database and everything.

But now I’m like, all right, tell me where all this stuff is and what the mental model is I should be using. I don’t know if you have similar experiences or if you either of you have learned a new code base recently to comment on that. But

Erika (15:59) I’ve been in the same code base, but I did go through a re-org, I guess it’s been like six months, I think now. So it’s not like that recent, but less than a year. So, you know, feels feels recent still. and ⁓ it was only like an additive re-org. Like we basically like took our existing area and then like added a whole

like half a team’s area on onto what we onto what we were already doing. So yeah and you know I like I’ve been finding that for me the like best way to understand what’s happening is still through user flows and like user flow diagrams and that’s even like through to the back end like

Brittany Ellich (16:28) Classic.

Erika (16:53) For me, it’s like, okay, where do where does this start? Like, how do I know it’s broken? And like, how do I know, you know, what the part like what the important parts are in between? So, like, if it’s a bug, I’ll try to replicate it. I’ll try to ⁓ like write a test. Ideally, I’ll try to like, you know.

Do a debugger session or something like that to like know what’s what the code paths are that we’re hitting. and then I’ll look at like production telemetry data. and yeah, for for me that’s like the right place to start in like grounding in like a mental model of like a certain user flow. ⁓ because I find that I get

If I get too in the weeds too fast, I’ll just like totally lose the plot. Like I’m just like, that’s interesting. Yeah, like this query and like that. And like, yeah, we can make that optimization. And it’s like, I just totally like lose whatever I’m actually doing. So ⁓ focusing it on like, you know, this is what’s broken, or like this is what I’m trying to do, and like tying it back to like the actual customer experience, ⁓ or like feature or bug or whatever.

is is still like the right way to start for me. it’s hard with like a whole area because there’s so like it’s like combinatorial, right? It’s like, okay, like for example, so apps, like GitHub apps or OAuth apps. Like that’s the new thing that my team has taken on. and like there’s this whole idea of like tokens within apps. It’s like

I have never done like a deep code dive into like how tokens are minted, how they’re stored, like how they’re invalidated. and like I finally like wrote down all my questions. Like, these are all my questions about like all the different ways that like you know tokens go through these life cycles. And I’m like, it would take me like a week to, you know, actually go through each user flow.

manually and like validate each thing. ⁓ which like I’ll probably like split up and do over the next month or so. but yeah it it can definitely feel like a lot to kind of do it that way. Like yeah the the 3000 foot code view like

I guess can kinda get you started, but I feel like that’s also like a dangerous place to be where you’re like, I know something and then you tr start to try to do something and you’re like, Ooh, wait, no, I don’t I don’t know how this actually works

Bethany (19:37) I can totally see you going through that flow, Erica, like with your you have you’re so curious and like are you so good at like really diving into things. I can I can definitely see that. ⁓ but that’s cool to hear. I also haven’t changed code bases. I my code base I’m still working on the same one, but it has been going through a lot of refactoring.

which is very exciting. ⁓ it was very due for one, but it does now mean that like all my mental models are outdated. And then we also traded some AORs from our team to other teams because we were very overwhelmed with ⁓ the new sc scope of everything.

and AI apps as a whole. So it’s been really interesting trying to help others learn that and then hand it off and then also learn ⁓ basically a evolving code base. But I think I’ve ⁓

I’ve always had success with like just in time or JIT ⁓ learning, ⁓ because otherwise I I do get quickly overwhelmed if it seems like an infinite amount of information to learn or store my brain. It’s just impossible for me to do that. So if I need to know something, I usually will just l shoot off a question real quick to to co-pilot or if I’m trying to

ans get an answer from a different team. I’ll try to self-serve as much as possible with AI, like in their code base, ask a question about like the dependency and stuff like that. Sometimes it works, sometimes it doesn’t, but ⁓ at least it’s ⁓ doing some due diligence before ⁓ poking at other teams. But I I have found it very helpful for visualizing code better because I’m just such a visual learner and and that’s always how I’ve enjoyed learning. So it’s really cool that at a notice I can get

diagrams even if folks didn’t diagram it in the first place or if the diagrams are outdated so I really enjoy that.

Brittany Ellich (21:32) Nice. I like that. Yeah, I still feel like something that has remained the same is like the easiest way to get into something is to like actually solve a problem versus just trying to read it and glean what you can from it. ⁓ although one thing that I’ve found that I’ve really liked now that I can leverage AI for like learning a new space is asking it like, all right, here’s what I’m seeing. What are the logs I could look for? Or like what is the evidence that I could see for like what’s occurring.

to narrow things down ⁓ in the code, like what should I be looking for, you know, in whatever dashboards you have access to. ⁓ and it’s amazing how much faster I can solve, you know, a bug or a problem like that with that. ⁓ yeah, it’s been really nice.

Erika (22:16) Yeah, for sure. And also like asking why it came up with an answer. if something seems off, be like, why did you say this? Like, yeah. See how it responds.

Brittany Ellich (22:21) Hm.

Yeah.

Yeah. It’s a weird week too. Fable was just released and I’m still trying to figure out whether like is this is this it? Like is this the thing that now it can do everything and it it’s all correct or whatever? I don’t think so just yet. But ⁓ yeah. Weird weird week wondering when when our jobs will change to where we are no longer needed. ⁓ whether or not that happens. ⁓ what about

Collaborative learning. Pair pro how’s pair programming going for you all at this point? Like is this a thing that you still do? ⁓ and what does it feel like now?

Bethany (23:05) I would love to know how you all are pair program. I saw Erica nod, so I’m really hoping you have some tips because to be honest, I mean, we chatted with Dennis about this a while back on how you have to be so intentional about maintaining that social fabric. ⁓ and so it it’s tough. You have to you don’t really need to pair program as much, ⁓ since you always have a pair programmer with you now. ⁓ but it’s still good for your for your team, for your

well being to be able to collaborate with others. So ⁓ really looking for some tips on that.

Erika (23:39) Yeah, I have recurring pairing sessions set up with everyone on my team. you know, they run the gamut from like I have this thing that I’m working on that I want to ask questions or brainstorm on, to like talking about the team, like, you know, maybe something that happened, maybe we had a release, like let’s talk about how it went. ⁓ and then like there are times when we’ll like dig into the code. ⁓

You know, honestly, like I don’t feel like it’s meaningfully changed that much when we do get into the code. Like, kind of like you’re talking about the interviewing, like you have an extra tool now, but like, you know, somebody will run something on their prompt or whatever, or just like come up with an idea or whatever. and like

you know, you you’re you’re still implementing the code, like, whether it’s like through typing or through prompting or whatever. I guess I usually do find that pairing, like I write more code manually than like prompt the AI to use it. I think ‘cause like I don’t usually like one shot, you know, it’s like when I prompt it’s like, okay, then we like iterate on it or whatever. ⁓

But if, you know, it’s like, we wrote this thing, now we’re gonna write some tests, like I’ll fire off like a prompt to add the tests or something. yeah, but like I said, I I always feel like there’s there’s something to talk about in a pairing session. ⁓ it might not always be like code itself, but yeah, there’s always always something.

Brittany Ellich (25:26) Yeah, that makes sense. Yeah, last week I was co working with some of my new coworkers. ⁓ and I was so excited because I’m like, Yes, we’re gonna be in an office together. We’re gonna get to like, you know, pair like in real time or whatever. And we did do a let a lot of like very explicit, like, all right, we’re gonna meet and talk about plans or whatever. ⁓ but then like when it we had like spare time to actually do work work, it was just like, well, I’m gonna have my claws do this. Well, you’d have yours do that and then

And then maybe talk about like, I don’t know, something fun or whatever while we’re waiting for responses to to come back. ⁓ which I think is, you know, I I I like the idea of having regular like intentional pairing sessions, like you said, Erica. I think I’m gonna set that up ⁓ because I think the value is still there, even if it’s not like talking about like, all right, let’s work through this code problem, but like, you know, what is our mental model of this thing and does it align?

I think that that’s probably really valuable still and something that I think I’m missing for sure.

But yeah, different world now. ⁓ what about documentation? I think one of the reasons that I originally wrote this is I wrote a bunch of documentation on ⁓ the on some of the billing things when I first started on the billing team at GitHub. ⁓ and it ended up being really useful documentation for like the entire repo. ⁓ now I feel like documentation is almost a little bit more disposable.

because it’s easier to keep up to date in a lot of ways, which means that, you know, maybe it’s not adding this these docs you created to the repo unless it’s like, you know, actually something that you know is gonna live long term. so I’m curious, what are your thoughts on writing documentation and code for code in general now? ⁓ is this still still something that you’re doing regularly? Have you figured out a good way to keep it up to date automatically? ⁓ and you know, have has your opinion on docs changed at all?

Bethany (27:21) Ugh, this is such a good one. ⁓ because I think in theory it’s easy to keep them up to date. But I don’t know if in practice it’s still easy to keep them up to date. my team’s definitely been experimenting with like workflows to track changes and make updates to docs, which I’m really interested in seeing how that’ll that’ll go. but yeah, I think the the age old problem still exists that if you have

documentation in disparate places, it it still is hard to know to update docs unless you have CI checks or like agent instructions or or things in place to the culture in place to ensure those documentation that documentation gets updated. I do still think documentation is very valid because I think I still read docs, but I think even if you’re not reading docs, your agent has to read docs. ⁓

to understand how to do things, ⁓ a lot of the times. ⁓ like you can say it just w uses code, but I think

it it really depends on the system. If you’re if it’s a simple project, sure I can just read the code, but if it’s spans multiple projects, I think it gets a little tough after after that point. So I think our playbooks have gotten super important ⁓ to me personally to make sure that those are up to date, especially as more and more code is going through and you’re not able to actually keep up with the mental model of all the code going through. I think to me playbooks are the things that are are top tier for keeping updated. So you’re on call

all can immediately be effective and ⁓ understand how to work on something. So I think that’s my my kind of strategies. At least at least the playbooks are are golden. But Erica, I know you’re passionate about this, so I’m very curious to hear your thoughts.

Erika (29:10) yeah, I I feel like to me docs serve multiple purposes and one of them is generating alignment on something. ⁓ so whether that’s like technical direction or like

I don’t know, team, team practices, like, ⁓ you know, part of part of what makes docs valuable is that like they can be commented on, they can be edited, like, you know, they can be approved or denied. Like, so I think, you know.

Like we’ve said multiple times, like the human element is never going away, you know. ⁓ even if your doc is like this is how we use AI, or like this is how like you’re a robot, like this is how you’re supposed to work. Like that in itself, in my mind is a like a piece of documentation that can be, you know, approved or denied. So yeah, I

I agree that like code documentation is really easy to get out of date. but like I do think a certain level of like inline commenting can be really helpful for like anybody reading or reviewing your code. ⁓ so, and that’s like again like a form of documentation. You know, I think the pure like readme.md.

I still appreciate it because you know it’s like then I don’t have to do that work myself. Like like you said, like it still does take time to like fire off a prompt. So if it’s there and it’s up to date and it gives me the five commands I need to get started, great. You know, like ⁓ and some things are like not obvious from the code, like

get back to tokens, it’s like what permissions of my does my token need to access these resources? Like I can probably figure it out, but like if you tell me then great. Like I don’t have to do that guesswork. And it could take a while to kind of like suss out that information. So those are the things that kind of come to mind for me. yeah.

What about you? I mean, yeah, you were like the documentation master. Like this was your this was your thing.

Brittany Ellich (31:39) I’m still very passionate about di taxis. I’m gonna I’m gonna add that that link. ⁓ yeah, I’m still a big fan of di documentation personally because I feel like that’s how I you know, work through things as I write it down. and that is why I also write a lot of things on the internet. So I think it’s still valid, but I don’t think that I’m gonna like

I don’t think that en as many things need to be documented as there was before. Like I think that, you know, like write down the specs for how a thing is supposed to work. But how the thing actually works is probably better to infer from the way that the code is written than, you know, whatever your intentions were when you created it, because those can differ quite a bit. ⁓ and it’s hard to keep those up to date, especially as code is moving so much faster than it was in the past.

⁓ I think that it’s really, really hard to keep those things updated.

Yeah, I don’t think I’m committing as much documentation as I used to be because I’m nervous, more nervous, I guess, about things being out of date. and but I think that that’s still like a good framework for figuring out what should be documented is the the diataxis framework, which is like what is your reference code? What are the features that are supposed to be in this? ⁓ you know, where are things like you know, any sort of proof of concept or ADR, I think.

is still super valuable, especially because then that just becomes, you know, an artifact of a thing that you end up using quite a bit ⁓ when you’re building and you can like measure against it to be like, did we, did we do this the way that we said we were gonna do it? ⁓ so yeah. I feel like I use more ⁓ more AI tools to write documentation now and to like make sure that the documentation that I write actually makes sense.

Without. I don’t know.

But maybe that’s still the same amount that I did a year ago. Because I don’t know that they’ve gotten like dramatically better at writing documentation.

Bethany (33:36) Yeah, I was just gonna ask you both what your thoughts on AI writing documentation was. Cause I know some people draw a line at AI writing for them or writing in their voice or or something, but others are like, yeah, you’ll stay better up to date or probably can be more clear that way. So very curious, that perspective. But A Britney, it sounds like you do use AI to write, at least like with documentation and and things or to check your your writing.

Brittany Ellich (34:04) Yeah, I would say it’s mostly like as an editor to be like, all right, does the thing that I wrote make sense how I wrote it? I feel like I’ve I’ve been wanting to I’ve had this idea for a a comic rolling around in my head for a while now where did you ever see that meme where it was like ⁓ where do you draw the line and it was like a pita thing where there was different animals on the image and it was like spanning from like dogs to like

You know, like horses or whatever. Like, where do you draw the line on the animals you eat versus the ones that you like, you know, don’t. ⁓ and somebody went in like with a pen and like drew on there, like, all right, this is, you know, every day, yes. Or this is, you know, economic collapse. This is where I would draw the line or whatever, you know. I feel like there’s a very similar meme to be made about like where you feel about how you feel about AI generation. ⁓ like I’m

Not a big fan of AI generated images. Like I wouldn’t do it myself. I don’t necessarily like judge other people for it, but I know that there are some people that are like no AI generated art whatsoever. Same with like, you know, AI generated text. Like I’m not gonna use AI to generate text, but I’ll use it to edit. and I think that like there’s some people also who like are very ⁓ at a different spot in that spectrum. ⁓ AI generated code, like yeah, that’s standard now. I think that’s just

normal how you know I will pretty much accept like yes this does what I want it to. Great. Let’s use it. ⁓ so yeah I think that’s sort of where I’m at on the spectrum. Use it more as an editor. I think I’ve tried AI generation of actual text in the past and it just never really ⁓ usually when I’m like taking the time to write a thing I already have something in mind on what I want to say. And so there’s not really a a ton of value in saying like hey help me write this thing.

⁓ but help me look at this from the perspective of somebody who’s in like communications or something. Like how does this come off ⁓ externally? ⁓ I think that’s really helpful.

What about you both?

Erika (36:08) Yeah, I go back and forth, because it’s so it’s so easy to like go from blank page to something with like AI generated text and like I feel like it kind of depends on the audience and like what it is. Like if it’s I don’t know, if it’s sort of like a summary or like a PR description or something like that, like I’m more

fine with AI generation because like it doesn’t it’s not like it’s not like a piece of content that’s like meant to be read kind of as ⁓ what am I trying to say? Like I guess I and I don’t really know. I don’t really know where the line is ⁓ for me personally but like

I think if it’s like a broader discussion post or something like that, like I’m less likely to generate it.

Yeah. So I don’t I don’t know where that where that line is. ⁓ I can’t like really define it, but maybe it’s sort of like an internal versus external audience or something like that. Like internal to my team or yeah. Like personal notes too. I’m like, sure, just like generate a note and it’s fine.

Brittany Ellich (37:30) Yeah, I feel like this is a thing too where every time I think about it, I’m like, whichever way this ends up going, like whatever I say now is probably gonna get me canceled on the internet in the future or something. Like I’m gonna say the wrong thing and people are gonna be like look back and be like, she used AI for that, or you didn’t use AI for that. And I feel like either way, like something’s gonna somewhere in this story is gonna go wrong. So I’m like, I’m afraid to take a stance anywhere.

Bethany (37:30) Yeah.

Erika (37:56) Well well

everyone apparently already thinks that we’re AI generated, so maybe this is the point where we say that we are definitely not AI generated.

Brittany Ellich (38:01) That’s true. Yeah.

AI

generated podcasters right here. That’s true.

Bethany (38:11) No, we are not we

are not on that AI podcast network. ⁓

Brittany Ellich (38:15) Mm-hmm. We can

Erika (38:16) I swear.

Brittany Ellich (38:16) do the do the do the do the face thing or whatever where you’re like, look, AI can’t

do this.

Erika (38:22) Look at all plug of my fingers. I promise I’m not AI. ⁓

Bethany (38:26) And before AI does be able to do that.

no, I think like I mean, talking about thing being cancelled on the internet I mean I think like any stance is gonna get cancelled on the internet by someone, so it’s just kinda kinda say say what what you feel in the moment. Maybe it’ll change, maybe it won’t, but ⁓

I I think these are really interesting perspectives and I love the like thought that like well it it it depends on the audience for sure. Like I’ve definitely generated like notes for myself and and things like that. Like I’ll type something that I’m like for sure would not make sense to future me, but like AI can turn it into something that might make sense to future me. ⁓ but yeah, I I mean I used to well

I tried to not let ⁓ like co pilot post as me. ⁓ I mostly do now just because I can’t figure out how to get it to stop posting as me. Like I’ve told it to not do that, but it still does. So I’m like, okay, you know what? At least it’ll be clear about what things are and people will know it’s not me because that’s not how I talk. But ⁓

I d I do think that it can be very verbose and I like I don’t like reading through AI text a lot because it’s just like really hard to parse a lot of times. Like I’ve even tried to make skills to say, okay, you are responding to somebody who does not know this code base, like it’s just they’re integrating into this. Do say the simplest thing possible without leading into what the implementation details are. ⁓ and that’s it. And it just struggled so much with that. I’m like

Two sentences max. You can’t you don’t get any more than that and it struggled. ⁓ so I think like with the tools now they’re just not very good at writing and I don’t think they’re very effective at like communicating, except i in ways that probably would help it with like its thought process or something like that. So I yeah, if it’s anyone outside of me reading it, I typically won’t won’t generate it at least I’ll I’ll

maybe have it at it like you do, Brittany. But yeah. I think a at this point it’s just not not there yet for writing.

Brittany Ellich (40:42) Yeah, makes sense. Nice. Well, this has been a fun ⁓ trip down blog post memory lane. ⁓ but I think we’re just about at time. So, I think that we need to move on to our fun section. And speaking of AI generated, there are two options that were both AI generated.

Yeah, agreed. Okay, yes. Two Truths and a Lie, new job edition. Each host tells three short stories from a first week or first day at any job they’ve had. The other to guess which one is invented. No sourcing needed, pure stories. Great. Do do you need any time to think up?

Erika (41:04) I think confessions won.

Bethany (41:24) Yes, it’s been over four years since I last started a job. I don’t think they were that dramatic either.

Brittany Ellich (41:24) Good stories for this.

Mm. Yeah, I only have like one very f fun like first week at a new job story. ⁓ and it’s from a long time ago. So, ⁓ yeah.

Erika (41:42) I mean they don’t have to be, you know, anything crazy. In fact, like, yeah, the more benign, the more less the less likely we are to think that the lie is a lie.

Brittany Ellich (41:47) Mm-hmm. That’s true.

Okay, I’m gonna take a second to write something down so that I can read from it.

Erika (41:59) Okay. ⁓ I guess I can go first. okay. ⁓ first thing is I started ⁓ my like new joiner class with somebody who I’m now on the same team with.

Bethany (42:00) Needle.

Erika (42:27) second one is ⁓ I merged a change into GitHub monolith my first week on the job. And third one is that I did all of my onboarding on time, like on schedule.

Brittany Ellich (42:55) That has to be the lie.

Erika (42:55) Yeah.

Bethany (42:58) Yeah, I don’t think I even finished mine.

Brittany Ellich (43:02) Still have some onboarding task or whatever sitting there in a GitHub issue that has never been closed. ⁓ yeah.

Bethany (43:05) Yeah. Yeah.

Erika (43:08) That was not the lie.

The lie was that I contributed to the GitHub code base in the first week. I did not. I think it was like the second week. Yeah.

Brittany Ellich (43:14) ⁓

Okay. That makes sense. And ⁓ We should have known that. Actually I shouldn’t have, because I started like a month after you. But Bethany was there, so

Bethany (43:18) Splittin’ hairs there.

Erika (43:27) Yeah.

Bethany (43:27) I did not keep track of your contributions, believe it or not.

Erika (43:30) No,

I can

Brittany Ellich (43:33) Your first week contributions. I know that there are ⁓ DX right now is doing a thing where they like measure your time to your first 10 PRs, I think, as like a measurement to see like how much more quickly people can get onboarded. That’s like your onboarding time is your time to your first 10 PRs. ⁓ but yeah, I don’t know about the first one.

Okay. ⁓ I’ve had a lot of jobs ⁓ throughout my life. ⁓ so thinking about some some good ones. ⁓

The first one, I started a new job and met some very cool people that were part of what we called the new crew and they became really good friends of mine.

Bethany (44:18) If that’s a lie, I’m

gonna riot. They’re terrible people.

Brittany Ellich (44:22) They were not cool.

Okay, second, ⁓ I once left a job in the first week after having to clean up puke ⁓ multiple times. ⁓ and third, I I once had to go to a training where I learned how to measure sound decibel level.

⁓ within a ⁓ hotel at very last ⁓ god I can’t even do a lie. yeah I like I started in a place and then it just like shit I need more details.

Bethany (45:05) What’s up?

Erika (45:06) Wow.

Bethany (45:13) I was like these are so hyper specific.

Brittany Ellich (45:16) The ver very, very specific. ⁓

Erika (45:18) Well that was easy.

Brittany Ellich (45:20) Yeah. I did leave a place though.

I I got a job at a pizza place when I was in high school and I was like the smallest person that worked there and somebody threw up on the twisty slide and I’m like, all Brittany, you gotta go clean it up. I was like, god, this is terrible. ⁓ so I did, and then I had to do it. They like threw up a second time on the same slide and I was like, All right, I’m out. This is it, I’m done. No more. Yeah. Yeah. It was yeah.

Erika (45:33) Awolo

Wow yeah.

Bethany (45:45) Honestly,

I respect that. I think that’s valid.

Brittany Ellich (45:50) Most assertive I ever

was in a job. Bethany, what’s yours?

Bethany (45:55) All right. so for my very first full time job, I was the only one who wore a dress on the first day of the job. for the my second one, I wrote my first lines of Ruby my first week at GitHub. ⁓ and my third one is I never got my welcome package from GitHub.

Erika (46:15) Feel like your welcome package is a lie.

Brittany Ellich (46:17) Yeah, same.

Bethany (46:19) It is. I got two welcome packages. I feel like it’s long enough that I feel okay telling people that I was like, I can’t tell anyone this.

Erika (46:21) Yeah.

Brittany Ellich (46:23) Even better nice.

Erika (46:27) Yeah.

Brittany Ellich (46:30) They’re not

gonna make you t make you send back your

Erika (46:31) I’m like, I know you well enough to know that you would have followed up on that. Like, excuse me, where is where is my sweatshirt?

Bethany (46:35) Yeah yeah.

Yeah, yeah. I would’ve. I love some free swag, honestly. What?

Brittany Ellich (46:39) Yeah.

Erika (46:41) Mm. ⁓

Brittany Ellich (46:43) Did you get two sweatshirts?

Did you get two two handle sweatshirts?

Bethany (46:49) I did, yeah. Yeah. Yeah.

Brittany Ellich (46:51) What a flex.

Erika (46:54) You can clone yourself.

Bethany (46:56) Yes, it’s true. Someone composes

Brittany Ellich (46:56) Yeah.

Bethany (46:58) me.

Brittany Ellich (47:00) ⁓

Excellent. well, this was really fun. It was great to chat with you both. thank you so much for tuning in to Overcommitted. If you like what you hear, please do follow, subscribe, or do whatever it is you like to do on the podcast app of your choice. Check us out on Blue Sky and share with your friends. Until next week. Goodbye.