A podcast about coding, but not code

We want to entertain, inspire, and motivate you -- or to put it another way, make your coding career more enjoyable.

... With your hosts: Adam, Ben, Carol, and Tim.

Adam Tuttle Ben Nadel Carol Hamilton Tim Cunningham

Episode 277: Some Fun S#!t We Did With AI

MOST IMPORTANT NOTE: We are looking for a new editor. Interested? Email workingcodepod@gmail.com!

In this week's episode the gang, minus Tim who is stuck at work fighting fires caused by AI lighting the world on fire, talks about having fun using AI to solve weird and complex problems.

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

Mentioned in this episode:


Transcript

Spot an error? Send a pull request on GitHub.

[00:00:01] Cold Open

[00:00:01] Carol: you know the good thing about being the government is that they want to try everything, like every tool, like we to see if it's usable, we want to see if it's something we should purchase. And so, you know, we get these like free trials to s to like to see if the software works for us. My opinion is cursor isn't working for me.

[00:00:18] Intro

[00:00:18] Adam: Ha ha ha. Okay, here we It is show number two hundred and seventy seven. And on today's show, we're gonna talk about some fun we did with AI, because that's all we do anymore. Um

[00:00:41] Carol: Mm-hmm.

[00:00:42] Adam: but first, actually before we get to as usual, I have uh a little bit of an announcement. I I just wanna get out in front of this. I also was having an affair with Jensen Huang

[00:00:53] Carol: Ooh, too soon.

[00:00:55] Adam: Uh

[00:00:55] Ben: Better we find out from you.

[00:00:57] Adam: Yeah, yeah. So

[00:00:57] Carol: Yeah, yeah. And then CNN.

[00:00:59] Adam: no. Anyway, but

[00:01:00] Carol: Mm-hmm.

[00:01:01] Adam: first as usual, let's do our Tramps and Fails I should say Tim's not here tonight. he is currently fighting fires. it's a brave, new, interesting world we live in, and and so Tim gets to play firefighter tonight. So just the three of us, and that means Ben, I am coming to you first. What's going on, my friend?

[00:01:17] Ben's Fail

[00:01:17] Ben: Yo, I'm gonna go with some soft failures here. Um one is just that I've been feeling kind of meh lately, which I think is part of why we didn't record last week, because there was stuff came up and I just also couldn't motivate. but the week before that I wasn't here because I went down to CF Summit East, which is the Cold Fusion conference down outside of DC and Reston. And It it's obviously a triumph that I'm motivated to leave the house and put on real pants and go talk to people. Did I have real pants on? No, I had shorts

[00:01:45] Adam: Ha ha ha.

[00:01:46] Ben: on. You know what I I live in sweatpants all the time. So I think part of me probably considered just wearing sweatpants to the conference. So I had to double check my memory. Uh

[00:01:54] Adam: I am wearing the same pair of jeans I've been wearing for at least a week.

[00:01:57] Ben: Yeah, yeah. there's no shame in that. We all work from home. We're all surviving. anyway, so so my my fail is that I don't think I really talked about a single tech-related thing at CF Summit, which made me sad. Yeah, just you're sitting

[00:02:15] Adam: You're talking about like on the hallway track sort of s situation?

[00:02:17] Ben: there, you're swiveling around in my seat, saying hi to people, talking to people in the hallways, that kind of stuff at the at the mixer at the end of the day. It's a one-day conference, so it's not a tremendous amount of opportunity. But yeah, mostly just talked about TV shows. There's some reminiscing about Star Trek, talking about people's insulin levels and and diabetes and just all kinds of random stuff, and just really none of it about technology at all. And I think that that's a byproduct of this AI era that we're living in, where I think people just I don't know if it's that they're not interested in technology anymore. It's more like I I think the AI stuff is so overwhelming. Say it again.

[00:02:57] Adam: Well we're not doing the technology anymore. We're not doing the technology anymore. We're just having our assistants

[00:03:03] Carol: Mm-hmm.

[00:03:03] Adam: do it for us.

[00:03:04] Ben: I know, I know. It's really, it's it's I don't know, it just makes me sad. You know, I reminisce about the times I was at conferences and everyone was just talking about trying this new database or or framework or library or here's some new technique we're exploring with at work. And you're doing containers. we haven't tried containers yet. I I don't know, just like there there always seemed to be such a buzz about all the various little technology things that everyone was trying and it was so fascinating. And now I it's not even like people were bringing up AI and I was steering the conversation away from it. It was like people just didn't talk about tech. And maybe it was the sample of people that I interacted with and it's completely coincidental. But I don't know. It just it it made me a little sad to to

[00:03:52] Adam: What else is there to talk about?

[00:03:54] Ben: yeah. And and and so sort of like dovetailing with that and with the topic of the show as well is I just not not to compare myself to other people, right? That's the root of so much unhappiness. but I hear continuously about all the people really thriving with AI. And I just feel like I am struggling very much to Feel productive with AI beyond very either completely solo projects that I'm doing just for myself or in very small ways. When I'm working with the team at work and we have five people on the same project and we're all using AI, I almost It's like I'm getting more done than I've ever gotten done. And I also feel like I'm moving slower than I've ever moved before. And it's it's this weird juxtaposition of productivity and progress.

[00:04:55] Adam: Yeah.

[00:04:56] Ben: And and like I just I I'm starting to feel like I'm not smart enough to figure out how to connect the dots on getting it from productive to progressive. Aggressive, if that makes sense.

[00:05:09] Adam: It took me a minute to process what you were talking about there, but y you said you're getting more done, but you feel less productive, right?

[00:05:17] Ben: It's it's almost like AI enables so much stuff that me and maybe the team are are like um what's it what's it called when you have a unhealthy relationship?

[00:05:33] Adam: Like a parasocial relationship with a codependent. Yeah.

[00:05:35] Ben: It's like we almost have like this codependence now with

[00:05:37] Carol: Mm.

[00:05:38] Ben: the AI, where I I don't want to go so far as to call it cog cognitive offloading, but I feel like because the AI has such a fine attention to detail and it can ask so many questions and it can and it can find so many little gaps in the way you're thinking about stuff, that We end up doing so much specification about the things that can be done. And then it goes and it fills in all those holes. But then it gets so hung up on so many details. And the more we put into it, the more details it gets hung up on. And then it's like, I feel like I'm wading through clay at some point, trying to just get it done. Because I not not not to just ramble, but it's like I was dealing I I run a retrospective when the when Claude is done so I can fine-tune some of the ways I'm having it work. And it'll s uh

[00:06:33] Carol: that's really smart.

[00:06:34] Ben: it's it's help it's it's been helpful. And it'll say like, uh in this unit test that we're running, I'm finding I I've like Claude, this is what Claude's telling me. I ran this unit test and I stumbled because I was looking for the language inside of a data attribute on an element in the rendered page, but I didn't realize that CFML had been encoding the attribute. And I was looking at it in plain text. And I was like, what are you doing? Like, why are you looking at the value of a data attribute? And it's like, well, because we want to have this confirm, and I want to make sure that the confirm exists on the page so that it we don't have pages that don't have the confirm. And I'm like, but that's not a unit test rule. That's just like a how we build things rule. You shouldn't have to check for that because what that

[00:07:22] Carol: Si, si, si.

[00:07:23] Ben: means is that anytime I change the language of how the page works, now you have to go and change the unit test. that's bonkers. And I didn't

[00:07:28] Adam: Yeah, yeah. So

[00:07:29] Ben: tell it to do that unit test in the first place. So now it's like anyway.

[00:07:31] Adam: Yeah. So

[00:07:31] Carol: Ha ha Ben.

[00:07:35] Adam: this is reminding me of a video I saw in the last week, about how our job is kind of becoming gardeners. I don't know if you saw I know you watched

[00:07:42] Carol: Oh yes.

[00:07:42] Adam: Theo's videos too, Ben. Okay.

[00:07:44] Ben: I I've been on a theo break.

[00:07:46] Carol: Mm-hmm.

[00:07:48] Adam: but he was talking about how, you know, i in some ways what we're doing is starting to resemble gardening, where it's like, you know, it kind of grows on its own and our job is to prune and be on the lookout for signs of bad stuff, right? Look out for the slop and remove stuff. And he did explain, slop kind of begets slop, right? if you if you let junk into the code base, that becomes an example that the model can then, you know, grab onto and use as something that, you like t tests like this? You like tests that verify stuff that doesn't need to be tested? I can do that. And it adds more. Yeah. Um

[00:08:19] Ben: Well, I can do that all day.

[00:08:20] Carol: All day long. Mm-hmm.

[00:08:22] Adam: and so Yeah, that I'm getting that vibe from you. It's like it feels like you need to spend some time cleaning up stuff that shouldn't be in the code base. But

[00:08:31] Ben: Right. And then take that and multiply it by five different people on the same project. And it's it's shocking to see we're all working on the same uh product and we're all sort of basing on some of these shared convention markdown documents. And yet when we do little screen shares about what we've built, everybody's pages look wildly different. And I'm like,

[00:08:55] Carol: Mm-hmm.

[00:08:56] Ben: we're using the same technology. This is wild to me how. I cause when I do it, I'm like, okay, I have this one thing and then I do it again and it looks basically the same, and then I do it again, it looks basically the And I'm like, okay, well, it just knows what it's doing at this point. And then you look at someone else's implementation and it's completely different, and buttons are in a different place, and labels are different, and alert boxes are being rendered differently and in different places, and you're like, How are we generating such different code? It's it you you almost get into your own little bubble.

[00:09:24] Adam: That's that is very much the the thing that the gardening movement is saying we need to be on the lookout for.

[00:09:31] Ben: Right, so then it becomes say it again.

[00:09:31] Carol: We have design templates that help with that. We have design templates that help with that a ton. So we've given it our design standards. So as it's creating things, it has a clear understanding of what the look and feel should be. And then I've also fed everything our help context. So everything that's in our help URLs. So where the user is like, what happens on this page? It has ev all the access to all those images to show like what the screens look like. So as it's building backwards it has information to use. So a lot of our designs look the same now.

[00:10:06] Ben: So okay, yeah yeah. It's I

[00:10:06] Adam: Yeah, so you guys need like a gardening day, right? You just take the whole team. No, seriously. You take the whole team

[00:10:10] Carol: Mm. Yeah.

[00:10:12] Adam: and you like you you just break it up okay, I'm gonna tell it to look for anywhere that's doing toast messages that's inconsistent with other pages and I'll get all to use the right one, right? everybody you divide and conquer and get everything back on track and then document that stuff.

[00:10:26] Ben: So it but then okay, if I can just continue to rant for like twenty more seconds, then I'll be done.

[00:10:29] Carol: Mm-hmm. That's all you get though. Twenty seconds.

[00:10:31] Adam: Twenty more minutes?

[00:10:36] Ben: So we we are trying to refine the the design specifications. But then the issue is like I'll have a design and then someone else has a design. And then we try to figure out okay which one works best and or which little aspects work best. And then we try to bake that into the design. And these design specifications just come in thicker and thicker and thicker. And then eventually you reach a user interface where it just doesn't conform to the design in a necessary way. it's just a little bit different. And then you try to deploy a PR and then the automated GitHub. bot is like, well, this page is being rejected because it doesn't adhere to the specification. I'm like, yeah, I know, but it doesn't really work for the specification in this particular page. And I put evidence of that in the PR. And then the bot will be like, well, but we still need to have it conform to the specification. And I'm like, well, I don't know who to talk to because I don't know how this bot was set up and I don't know who controls it. And it's so it's just, it's, you know what it is? It's like, It the the the metaphor that I keep coming back to is that it just feels like I'm working with a bunch of black boxes. And I'm sort of reaching into one black box and I'm doing some stuff and I'm pushing that into another black box and I'm just waiting for something else to come out. And when it comes out, I have to figure out what to do with it. And I push it back into a different black box and I'm waiting for that black box to report some stuff. And it's just it's it's it's like, you how they say like, if you have a system that's like 0.95% reliable, and then you chain 10 of those systems together. You don't have a system that's 0.95% reliable. You have a system that's 25% reliable because

[00:12:14] Adam: Mm-hmm.

[00:12:14] Ben: of maths. That that's what it feels like. It's like it's not one indeterminate system. It's like we've chained together the entire team of indeterminate systems that then interact with indeterminate bots. And it's it's it's not just indeterminate, it's like indeterminate to the tenth power. And anyway, I just feel stupid. I feel I feel dumb and I and I feel like I don't have the tools mentally to figure out how to make it better. This is the thing I'd say to my wife. I'm like, I feel frustrated and I don't know what to do about it. Anyway, I'm done. That's my failure.

[00:12:49] Carol: Well, I want to add to yours. So I feel like I do good with AI because I'm a mom and I talk to it like I did my kids when they were younger. So today, working through notifications, it tells me on alerting the first thing I should be concerned about is if this database is down, nothing will return. To which I just returned like responded with if the database is down, we have a larger problem than this alert

[00:13:14] Adam: Right.

[00:13:14] Carol: notification going out. You shouldn't be concerned with that. And then a little later it was asking me if I was certain I wanted to change the send to addresses. And I was like, yes, these people have left the company. We just never checked it into source code and they've been like editing this file, the store procedure, via a database like script. I was like, I want to get code back in sync. I was like, don't worry about the the send to list. Like I got that handled. And then there was another thing that was like a third weird, you should have never even questioned this. So my message back was, okay, I've given you three chances to try to be good. If you can't be good, don't talk to me. get your act together. And

[00:13:55] Adam: Yeah.

[00:13:56] Carol: that's when my responses start getting better and it stays on track. I feel like I have to talk to it like a mom. Like

[00:14:01] Adam: You have to threaten it.

[00:14:02] Carol: I've given you the opportunity to try something different. We're done trying. Get back on task. Like clean your room. Like we're not going outside to play yet.

[00:14:12] Adam: I will turn this car around.

[00:14:14] Carol: I will unblog the internet.

[00:14:16] Adam: Yeah. I've I've threatened to unplug uh different models at different times for sure.

[00:14:22] Carol: Yeah. Like you have to at some point say, look, we have a task. The task is this, stay on track. That doesn't help with

[00:14:29] Adam: Mm-hmm. Mm-hmm.

[00:14:31] Carol: design and what you're getting from it, but it does help with like weird o like unit tests that should have never been touched or the tangent of changing your your send to when it shouldn't have even considered it.

[00:14:45] Ben: That was my whole failure. Thank you for letting me rant. It's been a while. Carol, what do you have going on?

[00:14:51] Carol's Triumph

[00:14:51] Carol: I'm gonna go with a triumph as well. I lost my voice for about nine days and it's back. So that's a big giant win. I'm probably gonna cough during

[00:14:59] Ben: Heck yeah.

[00:15:00] Carol: the show, just forgive me. However, during that time I realized my communication style needs work. And I

[00:15:09] Ben: Yeah.

[00:15:09] Carol: think that losing my voice really helped me understand how quick I am to just call someone. and say, explain to me the two paragraphs you sent me because I don't want to read it and let's talk about it. And when I couldn't talk, like physically had no sound coming out, I had to read and I had to think about my response. And I've realized that if I take an extra moment, read what the person took the time to type and send, and then type of a response, it's gonna match their communication style more. Because not everyone wants to call and talk about it. They need time to process my response, just like they needed time to to process what they wanted to ask me or what they wanted to tell me. So um the past two days I've taken time to go, okay. You know, developer has asked for help. They've written it out. Instead of just hitting the call button and saying, Hey, let's chat about it. I'll walk you through it, I'm typing out instructions and I am giving like better information and I'm trying to communicate the way they prefer like to communicate. And I don't think I would have gotten there if I hadn't have lost my voice because I am quick to just start a call and say, handle it in three sentences instead of me having to type out messages. But I'm also the one that creates text stories with one sentence that basically says, Do work. So I'm very direct

[00:16:34] Adam: Mm-hmm.

[00:16:34] Carol: and and few words makes sense.

[00:16:38] Adam: Congrats on having your voice back.

[00:16:40] Ben: Yeah. Now i if I can say in your defense, I do think that a lot of conversations can often be more effectively handled over a call. And and I can only speak from my own perspective. I think I've always been late to call people because uh people people stuff is always more challenging. It's always I always felt like I was gonna interrupt people. And so if someone picked up the phone and called me, that always felt like it was removing the burden of initiation from my shoulders. And so I'll just say as someone on the other side, I think I always appreciated when people jumped, not jumped the gun that's not the right, but short circuited the conversation and just turned it into a call. Cause I always think those got resolved much quicker.

[00:17:30] Carol: So do you think a healthy way for me to approach it would be asking, hey, would you like to have a call and chat about this? Instead of just automatically trying to type up a message? That way they have the opportunity to say, Yeah, let's talk or they could say nothing, crickets and I type up a response.

[00:17:49] Ben: No, I do think asking people or s or suggesting, hey, I think this would be easier to discuss on a call. Do you have time for a Zoom? Something to that effect. I think that's gives people that are out.

[00:18:01] Adam: feel like are are any of these people your direct reports, Carol?

[00:18:05] Carol: I have zero direct reports, I'm not a supervisor.

[00:18:06] Adam: Okay.

[00:18:07] Carol: I just lead a bunch of people and mentor.

[00:18:09] Adam: Gotcha. You're a you're a sidecar. so then that changes it. I was gonna say, like, it's possible these people are feeling like a weird power dynamic where like Carol prefers to talk in person or over the phone, and so I have to do it that way because she's in charge of me. So but if that's not the case, then that doesn't matter. Yeah.

[00:18:27] Carol: Yeah, I hope they don't feel that way. I am the architect and the role is like higher up, but it's not like I sign their timesheets or I control how much they get paid. I just am here to teach the team. So hopefully it's not that. I appreciate your input and I will try to do better. All right. So No, no,

[00:18:45] Adam: Uh certainly wasn't telling you that you needed to do better.

[00:18:45] Ben: Mm. Ha ha ha.

[00:18:49] Carol: no. I I want to do better. Like I feel like sometimes things get in front of you that you don't have an opportunity to change and it just, you know, it opens a door

[00:18:58] Adam: Yeah.

[00:18:58] Carol: to something you didn't see before. And I enjoy that now

[00:19:00] Adam: Self reflection.

[00:19:01] Carol: I yeah, I get to look back and go, how can I be better for my team? All

[00:19:05] Adam: For sure.

[00:19:06] Carol: right. So that's me, I'm

[00:19:07] Adam: Good on ya.

[00:19:07] Carol: winning, pretty winning, anyways. Tim's out, we'll mention that. So that you Adam, what you got going on?

[00:19:11] Adam's Triumph

[00:19:11] Adam: Mm-hmm. I'm gonna go with a triumph, and that is that um partly on their own and partly at my encouragement, my team is leaning really hard into AI code reviews. So I talked

[00:19:24] Carol: Nice.

[00:19:24] Adam: in the last show about um some of the tooling I was building up for trying to like sort of build our own manual AI code reviewer using like GitHub actions and and stuff. And uh I think I got some pushback from you guys, rightly so, that you know, there's there's ways around this that are uh less work and you know, why why reinvent the wheel? and so I I took a breath and I I looked at what was available and um I went ahead and uh started two different free trials of of AI code review products, and I'm not gonna mention any names because I'm you know, you got pay for that. If you want to sponsor us, then that's fine.

[00:20:00] Ben: Yeah.

[00:20:01] Carol: Ha ha ha

[00:20:03] Adam: Sponsor us. but we started two free trials, and you know, they both have their pros and cons. It's interesting to see them kind of like competing in our pull requests, who does the better code review and who's faster and all that kind of stuff. so that's been interesting. but

[00:20:21] Ben: so you have like both of these things are running on push and then giving responses? that's interesting. Do they respond to each other at all?

[00:20:27] Adam: Yeah. Right. So it's it's it's uh

[00:20:29] Carol: Nice.

[00:20:33] Adam: don't think so. I haven't seen it happen yet. We only added the second one today, like at the very end of the day.

[00:20:37] Ben: Gotcha, gotcha.

[00:20:38] Adam: Um but um yeah, I don't think that they respond to each other, but they do definitely respond to comments you leave. So like Steve did a pull request today, my CEO, who is doing less and less code these days, and he it was some CFML code and he it was a fairly small pull request, and one of the AI code review comments that he got back is you know, length is not a built in CFML function and it's not defined in this file. I think you meant len or whatever. and he left a reply comment on that, just as a sarcastic thing that he would do if I had left that that comment. he's like, what idiot wrote this? And the

[00:21:18] Ben: Ha ha ha.

[00:21:20] Adam: and the The review bot left a reply comment to him, it's okay. It's not, it's the easy typo to make and this and that. And I'm like, that's kind of funny. You're like, you're getting your code review and therapy. Um

[00:21:32] Carol: Nice.

[00:21:33] Adam: so yeah, so we're we're doing these two things. And basically, yeah, you create a pull request, and it will automatically start reviewing your code and leave comments and and do like request changes, and it'll leave comments on specific lines or or sections, like just like a human would, that sort of thing. And you can you can reply and say, like, well, actually no, because of this, this we're or you can say like that's a dumb edge case, that's never gonna happen. We're not I'm not gonna do that. It's like it'll reply okay, cool. I'm not gonna mark this as a deficiency or whatever language it uses. and so that's been good. And and sort of to facilitate all of this, I mean, being the guy who writes the policies is I guess a benefit, but we changed our policy. Um To not require so our previous code review policy or change management policy was that you know probably what you would expect, like if Carol creates a pull request, then Carol cannot approve the pull request to go to production, right? Has to be somebody else. Well, in this brave new world of agentic coding, is Carol still writing the code? Like philosophically speaking, right? Or is Claude writing the code?

[00:22:36] Carol: I know.

[00:22:38] Adam: and so Carol looking at it, like maybe there's a an element of like Carol should be able to approve it, right? And so what we've done uh so far, at least, and we have like less than a week of this under our belts, but as a baseline, we still require an approval, but AI bots can give an approval. and then in addition to that. We have we use GitHub as a has a feature called code owners where you can say that you have a code owners file, and then in that file you can say this file, if it's changed, requires a sign-off from and you can list a person or a team or a group of people, that sort of thing. And so if I change payment gateway.cfc, then it has to be somebody on this team has to review and approve the pull request as well, right? That way we can we can say, okay, like general rule of thumb, AI can write the code. I can, it'll push it under my account and I can approve the pull request and merge it as long as it doesn't hit one of those codonors um

[00:23:37] Ben: Yeah.

[00:23:37] Adam: blocks. And if it does, then I have to like get somebody else on the team to look at it and approve it. and there's some things that we're doing to help that go smoother. And I'm I'm real as I'm saying this, I'm realizing that this is kind of getting into the show topic. So I won't I won't mention this as as one of my things in the show topic, but so. like things that we're doing that are helping in increase our confidence in doing this. the instructions that we're giving the code review tools, one of the things is specifically one of your jobs is to verify that the pull request description matches what the code changes are actually doing, right? So that somebody can't sneak in a a cryptocurrency miner and and just not put that in the pull request body, right? That would be in there, right? so it's it's verifying stuff like that. and then we also, in addition to the code owners thing, we have like a list of quote unquote sensitive areas of the site, like payments, I don't know, privacy, personal information, leaking, that sort of thing. We've got this list of like 10 things. and it's says like, if you notice changes in any of these areas of the site, make a comment that says I I've noticed that there's changes to a sensitive area site and so make sure that you get a human review. Just just call that out one time. Just make sure that that it's there, that it's at least recorded. So those are helping. Uh and I think our velocity has definitely gone up this week as a result. So I am quite happy with this.

[00:25:03] Ben: nice. It'll be it'll be very interesting. I don't deal with anything like SOC compliance anymore. I don't know if Carol does. Tim probably might have something to weigh in on. But it'll be interesting to see how SOC compliance evolves in the age of AI. 'Cause I think you're the you're the one who's most steeped in it. And uh

[00:25:25] Adam: I just I just had meetings earlier this week with our auditors and they were talking about how specifically SOC two is going to be changing in the near future, specifically because of agentic coding and like this the

[00:25:36] Ben: Hm, interesting.

[00:25:37] Adam: new controls around that. So they were asking like how you know, what are we doing with AI and how are we using it and how might it affect stuff?

[00:25:43] Ben: Well then I look forward to your next compliance review so that you can tell us all about it.

[00:25:48] Adam: Well, apparently they give a lot of notice and then like once they change the requirements, they like give you a full year cycle to to to

[00:25:58] Ben: Gotcha.

[00:25:58] Adam: evolve your controls and stuff. So you get a lot of opportunity

[00:26:01] Ben: Heads up.

[00:26:02] Adam: to to catch up. But yeah. Yeah, we'll see. In

[00:26:05] Ben: Very cool.

[00:26:06] Adam: the next year or two I should have some hopefully interesting uh compliance notes on AI. So

[00:26:13] Ben: Fair.

[00:26:13] Carol: So for the the PR review tools, like we use GitHub Copilot, it's just kind of built in. Everyone is turned on it on every repo. You pretty much can't turn it off. I think some people can turn it off, but most people just leave it on because it's nice. You know, it's included in our pricing for what we do. I will say the one thing that gets us is we have a weird branching strategy where our main branch is development, and then we branch

[00:26:38] Adam: Mm.

[00:26:38] Carol: off of that for release branches. And because we've added another testing environment in between there called UAT, UAT and main aren't really in sync. So

[00:26:48] Adam: Mm-hmm.

[00:26:49] Carol: our developers get very angry when they do a PR into main It looks pretty good, maybe a comment, maybe something small. They merge it, they get ready to go to UAT and they cherry pick that change, getting like prepped for release. And now there's a whole new set of findings because

[00:27:08] Adam: Mm.

[00:27:09] Carol: the base of the branch isn't the same, things have changed, it s gets to decide on what it wants to pick at during each moment when it runs.

[00:27:16] Adam: Yeah.

[00:27:17] Carol: So that's I've been my only complaint with any of the PR tools is because of our branching strategy, we end up higher levels with comments that should have been found at the first level of the PR instead of like when we're going into release time. But we're changing that,

[00:27:34] Adam: Yeah.

[00:27:34] Carol: so it won't be a thing anymore. But that's been my only complaint.

[00:27:39] Adam: So there's a couple things. I don't know what model Copilot would be using under the hood to do that. but you know, like benchmarks SWE bench and the these other things that they're using to sort of measure the intelligence and the ability of these models. I started to see a couple of people doing sort of their own like personal benchmarks, and they're like they give it the same pull requests or code base or whatever to do like a code review on, and they uh measure How many findings does it have? And then they take the findings and have other models determine was this actually accurate? Was it a hallucination?

[00:28:14] Carol: Yep.

[00:28:15] Adam: how s was it as severe as it actually says it was, that sort of thing. And then so the I would be curious. I don't think I've seen anybody specifically review Copilot in that context. but yeah, I'd be curious to see how it ranks up against other models. So

[00:28:29] Carol: Maybe I could do that next week and let you know.

[00:28:32] Adam: Okay, yeah, do it.

[00:28:33] Ben: Yo, just on the topic of AI making me feel stupid, people hearing people talk about Git and pull requests in the age of AI, and people just saying flat out, like, Git doesn't make sense anymore. Or like pull requests don't make sense anymore, or like branching just doesn't make sense anymore. Like no part of me connects with that. I'm because I'm like, if you don't have that, then how do you work? Like, cause

[00:28:59] Adam: Yeah, hard disagree

[00:28:59] Carol: Yes.

[00:29:00] Adam: all of the above.

[00:29:00] Ben: Yeah. And people say it with such confidence. And and again, it's like one of these things. I I'm I must just be dumb to not understand what you're saying because it's so clear to you. And I literally can't imagine living in that world.

[00:29:15] Adam: Well, I think you have to ask yourself, what are these people actually working on? Right? Are they working on a big team with a bunch of people and and shipping to a product that uh has thousands of real production users? Or are they working on one thing by themselves with no users, right? Like are these the same people that didn't like git before AI?

[00:29:33] Ben: I don't know what anybody works on.

[00:29:36] Debugging the AWS CloudWatch Console

[00:29:36] Adam: So, yeah. All right, let's let's pivot into today's show topic, which is some fun we did with AI. Who wants I I wanna start if that's okay.

[00:29:45] Ben: Do it.

[00:29:46] Adam: I got I got a a fun one. So

[00:29:46] Carol: Good, do it. Yeah, yeah.

[00:29:49] Adam: know if I mentioned it on the show. I definitely mentioned it in our Discord a couple of weeks ago. Um that I was not able to view any logs in AWS Cloud Watch uh dashboard, like in their website, right? I could I I can get into the console, I can browse around, I can do stuff in IAM, I can, do ECS, run Lambda functions, whatever. Like it uh everything I tried in a WS console was working fine except browsing logs in CloudWatch. And I'd pull up the page, uh, and sometimes I would see like a little flash of content in the main content area of the page and then it would just disappear. Go to like blank screen and but the the rest of the skeleton of the page, right? The the sidebar was there, the header was there, it's just like the main content area was blank. And you know, being curious, uh, developer tools. I was like, are there any error messages in here that might give me some hints or whatever? Nothing super helpful in there. And I like I said, I posted in our Discord. I was my first thought was like, I wonder if this is an AWS issue, like uh, you know, is there an outage that and is anybody else experiencing this? That sort of thing. Checked with my coworkers. None of them were having any issues. This is very weird. So I was like, okay, well, you know, I'm not blocked by it. I'll you know, I the one thing I needed to look up in logs, like by that time somebody else on my team had already found it for me and and pulled it out of logs and gave it to me. So not that big of a deal. I didn't think too much about it. Yeah. Correct.

[00:31:10] Carol: So they could view the logs? They could view the logs, but you couldn't. Okay. Okay.

[00:31:14] Ben: That's weird.

[00:31:15] Adam: So I didn't think too much about it. And then a couple days later, or maybe even like a week later. I I've had to go back in there and it I ran into the same problem again. It's like that's weird. You know, if this was an AWS thing, you'd think they'd have it fixed by now. eventually I got the idea of like, let me sick AI on this and see if it can figure out what the heck is wrong. So I was just like, I don't remember the exact prompt that I used, but I do know I have I pulled open the browser and I opened the dev tools and I refreshed the page so it's all fresh and ready to go. And I popped open my agent and I said, uh, you can use computer use, use this browser. I have AWS console open. The only thing you are allowed to mess with is CloudWatch. Don't you go in my IAM. Don't run the fucking

[00:31:54] Ben: Ha ha ha.

[00:31:56] Adam: nothing. CloudWatch and CloudWatch alone, right? Like you are in me and I'm an administrator, so I'm watching you. Um and I was like

[00:32:03] Carol: I'm watching you. I love it.

[00:32:06] Adam: and I described the problem to it. Uh can figure out what's going on, maybe you can help me file a bug report or something. And then I saw it poking around in the dev tools. I was like, that's weird. And and and I'm watching the you know, they don't give you the full like thinking output anymore like they used to. So you kind of only get little bits and pieces of what it's thinking as it goes along. And I'm watching it poke around at the the um with like JavaScript stuff in the console, right? So you're like, you know, sometimes you could do like jQuery dot this or whatever and and interact with the page, right? Well, sometimes I guess you can have um like your SDK stuff available on the page. So it was doing like AWS.cloud watch dot this dot that, right? And like interacting with the page. And eventually it comes back and it's like, okay, I see the problem. your profile has saved preferences for like how to do I forget exactly how it phrased it, but basically it was saying that. I had save preferences for displaying CloudWatch logs. Like maybe I had moved some columns and like hide these columns, show these columns and and maybe you know sorting or whatever. And whatever I had saved was conflicting with like a UI refresh that they had done and it was like loading my preferences was causing the UI to crash. I was like, oh, that's interesting. Okay. And I was like, uh I don't feel like filing a bug report for this yet. Can you just fix it for me? Can you like use the SDK and like clear my CloudWatch display preferences? And it goes off and it does it. And now I can view CloudWatch again. I'm like, yes. That was so

[00:33:35] Carol: Mm.

[00:33:35] Adam: fun.

[00:33:37] Ben: wild. It's it really is just unrelenting in its desire to find things.

[00:33:40] Adam: Yeah. And I don't think that it's I don't think that it's like mean, I could be totally wrong, but I have no idea if it's documented anywhere. Like these the it's just like going on the dev, you know, into CloudWatch, open the browser dev tools and start poking at JavaScript objects on the page. I don't think AWS like gives you documentation on how to use what's available there. I think it had to like introspect and just figure it out, which was wild.

[00:34:07] Ben: That's pretty amazing.

[00:34:09] Carol: um

[00:34:08] Adam: So all right.

[00:34:10] Carol: maybe like two months ago or so, Ada no, we don't use AWS, we use Azure. Sorry. Azure had a thing going on where they had changed their UI. And if you were in dark mode, every tech like all the text was just gone. Like you couldn't see it anymore. So I would load pages

[00:34:25] Adam: Ooh. Very dark mode.

[00:34:28] Carol: to try to look at settings, and all I could see were like white drop downs in the middle of pages. And I was like, what the heck is going on? And Of course, like I can't hook up AI to any of our like we are not allowed to use chromium like We can't do any extensions, like can't have system access. So it's it's locked down with what we can actually give it access to. But I did start doing the whole like, Do you know about this? Like what's going on with it? And sure enough, same thing. It was like, Oh yeah, it's known. They've messed up like their dark mode settings. Just go to your preferences and put it back in light mode until it's resolved. And I

[00:35:02] Adam: Whoopsie.

[00:35:03] Carol: was like, So sure enough I put it in light mode and everything reappeared again. But then

[00:35:07] Adam: Yeah.

[00:35:08] Carol: I have one

[00:35:08] Ben: Yeah.

[00:35:08] Carol: tab in everything that is light mode and it made me mad for weeks until

[00:35:13] Adam: Yeah.

[00:35:14] Carol: it got fixed.

[00:35:16] Adam: That sucks.

[00:35:17] Recovering Ben's Old Flash Recordings

[00:35:17] Ben: I go. All

[00:35:18] Adam: Yeah. Do it up.

[00:35:20] Ben: right. So I've had my website for a really long time. And in the early days, I would do little screen recordings and I used a a Tech Smith product called Jing, J I N G, and it re

[00:35:32] Adam: yeah, I remember that.

[00:35:34] Ben: and it recorded them as Flash movies. And which was fine because the web was very friendly towards Flash. In fact, you know, early uh phone

[00:35:42] Adam: In the before times.

[00:35:43] Ben: gap Cordova kind of stuff, you could actually or Or no, it wasn't. It was what was the flex. Flex was the one where it would like

[00:35:48] Adam: Flex. Yeah.

[00:35:49] Ben: compile UIs into Flash for mobile devices.

[00:35:51] Carol: Cheese.

[00:35:52] Adam: Yeah, did some flex, yeah.

[00:35:53] Ben: Yeah, f Flash was a big thing. And then the and then at some point Steve Jobs said, Hey, Flash is gonna go away because it's not great for the battery. And then the web just basically turned off Flash. And at that point I had like 300 plus screen recordings recorded in Flash and on and being rendered on my site. And literally for the last 15 years, they've just had big white empty spaces where there was a flash video. And this whole time, I have tried various products, like professional products to convert flash to movies. I've tried free things. I've tried uploading to sites. tried like VLC video viewer apparently was supposed to be able to do it. All kinds of stuff and nothing worked. There was no way for me to convert these flash videos to like MP4s. And when Claude Fable came out a couple weeks back, or whenever that was, I said to Claude Fable, Hey, I have these Flash videos. I've never been able to convert them. I would love to get them to an MP4 format. Previous versions of Claude could not figure it out. Give it a go. And Just the unrelenting exploration. It started looking at the files and poking at them. And they're like, a lot of flash videos are just thin wrappers around video. Is that what this is? Let me decompile it. no, it's not a video. Like actually has little screen assets that it animates around. And it goes,

[00:37:24] Adam: wow.

[00:37:24] Ben: oh, and like it starts talking, you know, it's it's telling me what these files are made of.

[00:37:29] Adam: Mm-hmm.

[00:37:30] Ben: And then it's like, can I calculate where these things are on the screen? And does some math. He's yes, I can calculate where these things are. Let me do some sampling across a bunch of videos. And it's like, I sampled across 30 different videos, and I was able to do the math and looking at those timelines in the underlying video, everything seemed to check out. I think what I can probably do is just generate image files per frame and then use FFmpeg to weave them back together into a video. like Literally like 20 minutes later, had converted like 350 flash videos into MP4s.

[00:38:04] Adam: That is wild.

[00:38:06] Ben: Wild. And it had generated

[00:38:06] Carol: That's nuts.

[00:38:07] Ben: a JSON document of all of the different video file names and the dimensions. And I'm like, can you just upload those to Vimeo for me and give

[00:38:18] Adam: Ha ha.

[00:38:18] Ben: me the list of the Vimeo IDs so that I can fix

[00:38:21] Adam: Yeah.

[00:38:21] Ben: the the layout? It's like, yeah, no problem. Bleep bleep bleep. And like again. 20 minutes later, 300 videos had been uploaded to Vimeo. I had a mapping of old swf file names to new Vimeo IDs that I could then render with an iframe, I think is how Vimeo does it. And like this 15-year-long saga of trying to get these

[00:38:43] Carol: Mm-hmm.

[00:38:44] Ben: videos converted um is done, more or less. Now I see Adam highlighted the thing in the document here.

[00:38:50] Adam: I know what you're talking about, sir.

[00:38:52] Ben: So this this is the interesting thing about having um content that's authored in markdown and then converted to HTML. so I don't really have a great way of updating markdown. the Java library that I use to do the markdown to HTML is is this Java library called Flexmark. And Flexmark technically generates an abstract. syntax tree. It's kind of document object model of the markdown. And you can theoretically query it is not the right word, but you can like you can do like you can like iterate over the nodes in the syntax tree and look for certain types of things. And in theory, I could say, well, if this is a code block and this code block represents an object tag. And the object tag has a param tag and that param tag has a a value that matches this SWF file name, then I can just replace this whole thing with an iframe. I've I could have probably had Claude do that. I wasn't ready to go that far. So I'm like now just in time. Right before I render a blog post, I'm saying, does this because I can use JSoup at that point, which is like a Java friendly, it's like a Java version of jQuery. I can just mutate the JSoup DOM and swap.

[00:40:04] Adam: So instead of running one more prompt, you're running the same conversion.

[00:40:07] Ben: Yeah, yeah, yeah, yeah. Well, because I'm so this this is where like the trust factor comes in. It's like I don't

[00:40:12] Adam: Yeah.

[00:40:13] Ben: trust it to start running updates against the database.

[00:40:16] Adam: Dude, you can have Claude figure that out too. yeah. I mean the the the

[00:40:20] Carol: Mm.

[00:40:21] Adam: the I thought you that's not where I thought you were gonna go. I started talking before you finished your sentence, but uh

[00:40:25] Ben: No, no, no.

[00:40:26] Adam: you you know, you could have like have it pull, you write something that pulls out a sampling of thirty of the documents that have videos or whatever, have Claude do the conversion and then say and then have something else, another session, check and make sure it got it right.

[00:40:40] Ben: Yeah, that's I'm I'm not yet in that type of mindset where I have things verify things. It's more like I'm the verifier when it does stuff. So that's why I'm like not ready to pull that trigger. But that that kind of step probably would make more sense. But anyway, I

[00:40:56] Carol: Mm.

[00:40:57] Ben: like that that was just a huge win. Like I've had these hundreds of SWF files on my computer for this point it's going on almost 20 years, and finally Freaking Claude Fable did it. Mind blowing that it can do that kind of stuff.

[00:41:13] Adam: I'm impressed.

[00:41:14] Carol: That's awesome. Yeah.

[00:41:15] Ben: Yeah.

[00:41:17] Carol: I'm just surprised you wanted to go do it still. Like it's one of those things where after it's been wrong for so long, you kind of just give up.

[00:41:25] Ben: Carol, I still have

[00:41:26] Carol: Yeah.

[00:41:27] Ben: on my list of things to do to change Carol Hamilton. Well, what do you write now? yeah. I

[00:41:34] Carol: Weiler.

[00:41:34] Ben: still I still I am in the way that I am just in time changing

[00:41:39] Carol: Ha.

[00:41:40] Ben: SWF files to Vimeo iframes, I am just in time changing all of my Carol Hamilton references

[00:41:46] Carol: Ha ha ha.

[00:41:47] Ben: to Carol Weiler. So uh

[00:41:49] Carol: I love it. I love it. Okay, I'll take it then.

[00:41:51] Adam: Do you just have like a a a step

[00:41:52] Ben: This is another

[00:41:53] Adam: that's like here are all the just in time replacements we need you dirty bastard?

[00:41:56] Ben: Yes, I have a component in my

[00:41:57] Carol: Mm-hmm.

[00:41:59] Ben: Cold Fusion called blog post normalizer.

[00:42:00] Carol: Claude could do that for you. Send me your code. I'll have Claude do it for you. I'll change my name. That's

[00:42:06] Ben: Mm.

[00:42:07] Carol's Cursor Pilot

[00:42:07] Carol: awesome. All right, I'll go next if you're good.

[00:42:10] Ben: Go for it.

[00:42:10] Adam: Yeah, do

[00:42:11] Carol: So um you guys remember like six months ago when AI started and we all were like, Let's use Cursor. And then Cursor died, and

[00:42:21] Adam: Yeah.

[00:42:22] Carol: everyone went to Claude and everyone went to ChatGPT projects. I'm using cursor. So

[00:42:28] Adam: Say more.

[00:42:29] Carol: um yeah, so we just got the pilot into cursor for I think we're doing a 30 day pilot with them because you know the good thing about being the government is that they want to try everything, like every tool, like we to see if it's usable, we want to see if it's something we should purchase. And so, you know, we get these like free trials to s to like to see if the software works for us. My opinion is cursor isn't working for me. So

[00:42:52] Adam: Ha ha ha.

[00:42:54] Ben: Girl.

[00:42:54] Carol: I I've gotten really used to the responses from Claude and just how I interact with Claude day-to-day client, like the way it me back information, kind of I feel like I've trained it to be my companion. Now going in the cursor, it is just giving me wrong answers. It doesn't understand,

[00:43:14] Ben: Yeah.

[00:43:15] Carol: like though there is an agent file in there, like doesn't understand like this project's doing. And because it's a monorepo, I feel like I'm having to babyset it the whole time and direct it into the exact path of where I want it to go. So some of the stuff I was working on, I just ask it to give me a plan for what it wanted to do. After seven minutes of planning, I stopped it. and then ask it to just tell me like go make this work and it just immediately started changing things and I was like wait wait wait we we aren't really like doing it like hold on and yeah anyway so cursor isn't working for me spacex product not my not my jam I'm

[00:43:54] Adam: No, yeah, I forgot about that.

[00:43:55] Carol: a claude I'm a Claude

[00:43:56] Ben: that's right.

[00:43:57] Carol: fanboy yeah SpaceX bought it out so I mean they are they are huge on it like SpaceX is big on this is the best thing here's why you should use it I'm just not seeing it yet. after piloting all these things and having I have Chat GPT, so I have Codex, I have Claude, I have others, but I'm not using them. And then now using Cursor, Cursor isn't holding up. The one thing I do

[00:44:21] Adam: Yeah.

[00:44:22] Carol: like about Cursor, Cursor makes model selection seamless. It makes the determination of what model you should be using based on like how it's currently operating and what you're trying to accomplish. So for a user that doesn't want to determine what model to go grab, because it has the Claude models and it has um ChatGPT models, I don't think it has the newest soul models. But anyways, it has the newest Claude models. So I don't have to decide what I want it to do. I just hit balanced or high performance and it figures out what to do. So for a user that just wants to go write like a vibe coded project, it's nice because it takes some of that questioning, like the setup. out of out of the picture.

[00:45:09] Ben: Interesting. And I think I feel like I remember that Cursor trained their own models as well, but I don't know if that's

[00:45:16] Carol: They do, yep, they

[00:45:17] Adam: Yeah.

[00:45:18] Carol: do, but they also use the other models.

[00:45:20] Adam: Well yeah, I'm curious. So XAI had their own model that's where Grok is. There's a also there's two different models that are called Grok for whatever reason, but

[00:45:25] Carol: Yep, grok Mm-hmm.

[00:45:28] Adam: uh yeah, XAI had their own Grok. that's the one that makes child porn on X for you, if you want.

[00:45:34] Ben: Yeah.

[00:45:35] Adam: The so it's been years. It feels like it's been years. It's probably been like three weeks uh since I used Cursor. Uh yeah.

[00:45:39] Carol: Years. I know, right? That's why I said six months, right?

[00:45:44] Adam: uh I I remember being very happy with their composer one uh model what that was like built into Cursor back in the day. I'm sure you if you go back uh

[00:45:52] Carol: Yeah.

[00:45:53] Adam: on our older episodes you'll find it. But um yeah, I mean it's uh which if I remember correctly Composer one was uh a version of Kimi that they like did a lot of reinforcement learning on. So I think what they did was like they uh had people using like Opus, whatever Claude and and Chat GPT through the cursor UI, and then they had access to all this data of like these are your prompt and this is the context that fed in, and this was the result, and they used that to train their own model. That's what Composer was. And on top of that, however they managed to do it, it was really freaking fast. Um, and so it just it was a fun and and I don't know, it felt good to use it. It felt good that it was fast and stuff, but I I don't think it's Frontier anymore for sure. So

[00:46:43] Carol: I will say the fast statement is out the window. It is the slower

[00:46:46] Adam: it's not fast anymore?

[00:46:47] Ben: Ha ha ha.

[00:46:48] Carol: yeah, it is the slower of the three. Just starting the desktop app, it just spins. It just spins.

[00:46:54] Adam: Well yeah, I'm I was only referring to the the time from you know, you send your prompt until the time it starts generating and and the speed at which it generates. Not ev not yeah.

[00:47:03] Carol: yeah, no, it's all of it. It's yeah, it's not even just

[00:47:05] Ben: Ha ha ha.

[00:47:06] Carol: like the prompting. It's also just startup, getting into things, clicking around. It is very slow. And I don't know if it's because we use Windows. So, you know, it's a Windows app and almost every AI client was built to be Mac based or Linux based, not really um Windows based. So the Windows is the m the fastest. Even with Claude, it's it's not as good as on other um other machines, I guess, but Cursor is very slow. The only thing I well, the second thing I like about Cursor though is the built in IDE and it has the VS code IDE built in. So

[00:47:43] Adam: Mm-hmm.

[00:47:44] Carol: while it's supporting that, you know, things change all the time. I do enjoy just being able to hit the IDE button and then my codes popped up and I can just look at it, close it. I don't have to like actually bounce between products. So

[00:47:56] Adam's Photo Frame Dashboard

[00:47:56] Adam: Yeah. Okay. Should I go?

[00:47:59] Carol: Yeah, go ahead.

[00:48:00] Adam: Okay, cool. Yeah, everybody knows who the syntax podcast and Wes Boss are. you may or may not have seen he did a YouTube video not that long ago talking about how I think he said it was like his sister or sister-in-law got one of those um I think they're called Skylight uh did Sky Frame or whatever. It's a it's a device, you hang it on your wall, it shows

[00:48:17] Carol: Skyframe. Mm-hmm.

[00:48:20] Adam: like what's up with your family at all times, right? Your your shopping list and this is the week's dinner menu and here's your calendar and all that stuff, right? So uh a family member of his got one, it was busted, they gave it to him, and he eventually figured out he could have AI, have him try to fix it, help him fix it. And he ended up having to do hardware hacking on it, whatever. But it gave him this idea of like, this thing costs, I think it's like twelve hundred bucks, fifteen hundred bucks, something like that. It's crazy expensive, right? So effectively be a touch screen where it's like got a little dashboard on it, right? And he's like, I wonder. Can we just like hack this together with like a cheap Chinese photo frame? and and AI. And so he just bought a cheap, like a fifty dollar frame off of Amazon, plugged it in. He had like work with AI to figure out how to root it, 'cause it's running Android. so he had to be able to like take off the um the default app that launches to display pictures or whatever. Um, and then put on his own software to display like your own dashboard and it's just it's built with HTML, CSS and JavaScript, right? Um so he he did it. He made this video and he's like, and this is just this is not like a here's a thing I made and you can do the exact same thing as me, right? Buy this frame, run this software. He was like, this is so easy. You know, probably if you get a super cheap Chinese photo frame, it's gonna be running Android. Um and if you just plug it in and point your AI at it and and like say go, then it'll figure it out. And he like he's you can you know, I'm putting this up on a GitHub repo, point it, point your AI at my repo as well and say, I want to do this. My frame is plugged in. See if you can figure it out. That's like literally what I did, right? So I'm gonna see if I can reach over here and pick it up. It's I bought a big frame. Um, it's like compared to the size of my head, it's huge, right?

[00:50:09] Carol: that's cool. Yeah.

[00:50:11] Adam: So it's like a almost a 16-inch screen. I paid a hundred bucks for it because I wanted a big one. but I mean it's still a hundred bucks for this, it's a lot less than thirteen hundred.

[00:50:20] Ben: Yeah, that's crazy.

[00:50:21] Adam: Um and so this is I mean it's doing pictures right now. This is a screensaver, but um it

[00:50:25] Ben: And it's a touch screen.

[00:50:26] Adam: is a touch screen. so basically, and I've showed I think I've showed a bunch of of screenshots of this in our Discord. So if you're not getting the visual component of this, just jump into our Discord. Um but basically, you

[00:50:36] Carol: Mm-hmm.

[00:50:37] Adam: know, it's like uh I'll try to describe what I see on the screen here. So like up in the the top left corner, it's got the current time, and then next to that is the like my current local weather. And then in the top corner, this is hyper specific to me, it is Doing weather forecasts, information specific to skydiving. So it looks at the

[00:50:54] Ben: Mm.

[00:50:54] Adam: upcoming Saturday and Sunday, and it looks at very specific things like what is the cloud cover? Is it gonna be clear, partly cloudy, mostly cloudy, completely cloudy? What is the the and it's looking at aviation forecasts. So like is it VFR or IFR? And if you're a pilot, then you know what that means. You know, what's the what's the chance of rain? What's the what's the

[00:51:12] Ben: He's um

[00:51:14] Adam: steady wind gonna be and what's the wind gust gonna be? What's the gust differential gonna be? Because that's important for skydiving, right? So it's looking at all these things and and using that to to try and like make a decision on uh how likely is skydiving this weekend. And it's it's generating like colored bars. and so like green is good, yellow is maybe, maybe not, and red is bad. Um, and so it's kind of giving me a a an early glimpse into the weekend. Like how likely is skydiving this weekend? That's kind of fun. The the vast majority of the screen here on the main screen is like calendar stuff, right? It's like there's a big

[00:51:46] Ben: Yeah.

[00:51:47] Adam: chunk dedicated to like the next thing that my family is doing. Right now it says nothing else is scheduled. Enjoy the quiet. This so a lot of this and I will say a lot of this software is like a copy modify job from what Wes had in his repo. Right. So I I I pointed it at his repo and I said, My frame is plugged in, figure it out, go. It did it and it installed his app on my frame. Uh, and his app comes with like demo data, right? It's got calendar placeholders, right? It's got a bunch of events on the calendar, and you can like pick what time of day you want it to be so you can see what the calendar looks like in different scenarios, that sort of thing. So I I took that and I was like, okay, we wanna hook that up to my Google Calendar and I set it up so I'll give you the private iCal feeds from Google Calendar, and you can just pull in events. And then down here in the corner, it's showing a a specific album from my Google Photos. That's like my highlights. This is a private shared link. So, you know, it's not uh I it would be open to the public if you knew the URL, but the URL is not even in the Git repo. So

[00:52:45] Ben: Right, right.

[00:52:46] Adam: whatever. Um, so like that. And

[00:52:48] Carol: Mm-hmm.

[00:52:48] Adam: then so that's like the the main screen, right? So kind of that's the up here in the corner underneath the clock. I don't know if you guys can make that out. It's like it's saying there's fourteen days until Thanksgiving. Um right. It it's it's brown because it fall colors. And I I'm gonna pull it

[00:52:59] Ben: okay, yeah. I saw a red bar. Or a maroon bar, I guess. Brown. Hmm.

[00:53:05] Adam: closer so you guys can see. It's got like like uh subtle background effects in there, right? It's not just a brown bar, it's like leaves and stuff. Um acorns.

[00:53:13] Ben: okay, yeah, yeah.

[00:53:13] Carol: That's cute. Seasonal.

[00:53:15] Adam: Yeah. and and you'll see that motif repeated throughout

[00:53:16] Ben: I thought that was fingerprints.

[00:53:18] Adam: the whole thing. But like and then if I tap on that it goes to a screen that has a whole bunch of okay, there we go. I was gonna say it's now it's being slow. a whole bunch of like countdown things, right? So showing like these added a a new Google Calendar because I apparently something I didn't have in my life that I needed was another Google Calendar. Um, but specifically to fill up this screen, right? So this is like uh big events that I want countdowns to, right? Thanksgiving's on there, people that are important to me, their birthday is on there, my last day working this year is on there, Christmas and what is that, New Year's Eve,

[00:53:51] Carol: I love that.

[00:53:52] Adam: uh, holidays that I get off from work. Valentine's Day, St. Patrick's Day, Easter Halloween. Star Wars Day is on

[00:53:58] Carol: I didn't see my birthday. Did you forget it?

[00:54:01] Adam: there.

[00:54:01] Ben: Ha ha ha.

[00:54:03] Adam: Gonna go skip right past that.

[00:54:06] Carol: Mm-hmm.

[00:54:06] Adam: Mother's Day. Anyway, you get the idea. So that, and then

[00:54:09] Ben: Very cool.

[00:54:10] Adam: so like I don't I don't think I've made I don't think I it on the frame yet because it's not done. I'm just doing placeholder data. But so like some other things that I'm working on actively is package tracking, right? So it

[00:54:21] Carol: I love that.

[00:54:22] Ben: That's cool.

[00:54:22] Adam: In the year of our Lord 2026, we always have packages in the mail, right? So it's like where where are my packages?

[00:54:27] Carol: Mm.

[00:54:28] Adam: So I'm trying to come up with a good way to get the tracking numbers onto the device. I think I have a decent idea, and I'll talk about it more um maybe after I get it done. Um, but so I have a good way to copy and paste the tracking numbers onto there. I don't know if I'm gonna be able to do Amazon, which is disappointing because they like lock their data down pretty good, but maybe I can figure something

[00:54:46] Ben: Yeah.

[00:54:46] Adam: out. Maybe, you know, maybe I can just say like,

[00:54:47] Carol: I yeah.

[00:54:48] Adam: this is the thing I'm getting and this is the day it should be delivered. And then

[00:54:52] Carol: Or you

[00:54:52] Adam: Track

[00:54:53] Carol: could create a Thoraway Gmail account and set up your inbox rules to always forward

[00:54:57] Adam: Mm.

[00:54:58] Carol: your tracking information to that throwaway inbox which then it reads from.

[00:55:02] Adam: Yeah, could do that. Yeah.

[00:55:03] Ben: Carol's like theoretically.

[00:55:05] Carol: Just said.

[00:55:06] Adam: Not saying I've broken the terms of service to do that, but that's

[00:55:10] Carol: do you gotta do.

[00:55:10] Ben: Yeah.

[00:55:11] Adam: yeah, yeah, yeah. Uh so

[00:55:13] Ben: So is that an is that an Android app? or or did he did he completely like rebuild

[00:55:16] Adam: Uh the So

[00:55:20] Ben: an app from scratch?

[00:55:21] Adam: that is a very good question. Um, and I'll tell you my answer is uh. do you know how many lines of code I have written for this? Literally zero. I I've written zero

[00:55:28] Ben: That's bright. That's between uh Claude and God.

[00:55:33] Adam: lines of code. I have written zero command

[00:55:35] Carol: Hm.

[00:55:35] Adam: line instructions. It's all been prompts. I have that's the only

[00:55:39] Ben: This is this is like

[00:55:40] Carol: Wha what w what was your exact question, Ben? What was your question though? Oh yeah. Yeah. Yeah.

[00:55:41] Ben: full vibe coding. Is it an Android app? Like I didn't know if

[00:55:45] Adam: To to the the dashboard itself, is it an Android app? And so I can tell you

[00:55:48] Ben: Yeah.

[00:55:49] Adam: I there is an Android app involved. I believe it's called Home HQ. Now, is that something Wes made? I don't know. Is that something that was already available that just Wes made use of? I don't know. I can tell you, because I've asked, because I was curious too, that under the hood, all of this stuff that it's doing is HTML, CSS, and JavaScript. Um

[00:56:08] Ben: Oh, cool.

[00:56:09] Adam: but

[00:56:09] Carol: Nice.

[00:56:10] Adam: That's as far as my knowledge goes, right? Like I like I said, I haven't read the code. I

[00:56:14] Ben: Yeah, yeah.

[00:56:14] Adam: haven't looked at the code. It's like literally just pointed it at Wes's repo. And I said, I won't do this. Um now I I

[00:56:21] Ben: Yeah, that's cool though.

[00:56:22] Adam: have been I I do intend to make my repo public so that if anybody wants to use mine as inspiration or look at how I did stuff, whatever, that's fine. Um this as I've been doing this, I've been Keeping that, I wanna make this public in mind. And so like the Google Photos album URL and like my family's calendar URLs, those are all like in a config file that's not in the repo type thing. So

[00:56:50] Carol: Mm.

[00:56:51] Adam: But anyway, so this this has been fun. It's been a new obsession.

[00:56:53] Carol: Interesting.

[00:56:54] Adam: I've been working on it like every night while we're watching T V at the end of the night. So

[00:56:59] Ben: Nice.

[00:56:59] Carol: Fun. I just emailed my husband and was like, Hey, I wanna do this.

[00:57:04] Ben: Go find me a touch screen.

[00:57:05] Adam: Yeah. I I mean I was talking to one of my coworkers. I gave him the same demo I just gave you guys, uh

[00:57:06] Carol: Yeah. Mm-hmm.

[00:57:10] Adam: early midday today. And then uh totally unrelated, like sent him a question about work later in the day, uh like two hours later, and he replied, I've already bought the frame. Like

[00:57:21] Ben: Yeah.

[00:57:21] Carol: Yeah, I was just about to ask. You should link the frame you you got. 'Cause I want a bigger one, 'cause I'm old and my uh eyes are not

[00:57:24] Adam: Okay, well let me get an a affiliate link here ready so I can make seven cents off of your

[00:57:29] Carol: what they used to be.

[00:57:30] Adam: no. You know what I did uh also for my frame selection was I pointed AI at Wes's repo and I said, I want to do this. I want you to help me pick out the right frame for it. Cause in his repo, I guess a lot of people have done pull requests to like document, okay, this is the frame I bought, and this was the process to root it, or it was very easy, or it was difficult, or whatever. Right. So there's some documentation about which frames are known working, that sort of thing.

[00:57:54] Ben: Mm.

[00:57:55] Adam: So I pointed it at his repo

[00:57:55] Carol: I just

[00:57:56] Adam: and I said, This is what I'm I'm I want to do this project and these are my criteria, right? I want a large frame, I want a high pixel density, I don't want to spend an arm and a leg, and I don't want it to be underpowered, right? Give me decent CPU and and memory. I think that was pretty much

[00:58:11] Carol: Nice.

[00:58:12] Adam: the criteria that I gave it. And it came back with this one and it was like a hundred bucks. So

[00:58:17] Ben: Nice.

[00:58:17] Carol: Nice. I will say I just put a link in for show notes to the device markdown file in the repo. So if anyone wants to

[00:58:24] Adam: Cool cool cool

[00:58:25] Carol: quickly see the devices, they're in there.

[00:58:27] Adam: Yeah, we'll I'll definitely we'll just link the whole uh Wes's whole uh dashboard repo. But I think we have time for one more if you guys want to squeeze one in.

[00:58:38] Turning Noisy Emails Into Teams Alerts

[00:58:38] Carol: one cool thing I've done with a with AI in the past week is that I've realized we have a lot of noise in our alerting. So we have these nightly emails that go out that let us know the status of things. However, over time those emails just increase with no reporting of just problems. So now I get ten nightly emails for processes just to say

[00:59:03] Ben: Mm.

[00:59:03] Carol: It ran. Are there problems? And then just has big bold word that says no. And I've gotten

[00:59:08] Adam: Mm.

[00:59:09] Carol: tired of it. So a while back I kind of just created inbox rules and was like just everything should go to trash if it comes from our email address and you find

[00:59:19] Ben: Ha ha ha ha.

[00:59:20] Carol: these things, right? Like maybe here's a couple we should look at, otherwise they go to trash. I realize that's not healthy. the other thing is that we have our deployment notifications that go out. Well there's They were set up to be notifications per project. I realized that our deployment tool that we have has the ability to create subscriptions. I pointed AI at our store procedures that we use to generate these emails because they all come from SQL. And I also pointed them to Octopus Deploy and said, Hey, I really want to make oh our notifications mean something. I only want to get notified if it matters. I should never get an email to tell me everything's good with the exception of production releases. I just want to be notified that someone clicked the button and it went out. That way if someone did it accidentally, I know something just went out to prod, right? So I'm just alerted. That doesn't go to everyone. That's just a a me setting.

[01:00:13] Adam: Mm-hmm.

[01:00:14] Carol: So I gave it a key and was like, okay, I'm gonna set up my API key for a short-lived day, and you're basically gonna hit our deployment API and you're just gonna pull back everything that we can change. So within a day, I had all three of our deployment spaces updated. I had all of our lists like consolidated, I had all the steps deleted. Now everything is just subscription. So if any failures happen on runbooks or if they happen in tasks or if they happen in deployments, then we know about them. And instead of going to emails where everyone has already created these rules, they just send them the trash like me. I created a new team's channel and put in a webhook that just says anything going here, it's gonna be an alerting post. So now all problems get reported and now we have the ability to see it. This isn't a task I would have quickly been able to do without AI. I would have had to go like research all the variable sets, all the steps in the deployment. Like it would have taken time to just build out a list of what needs to change. But using AI allowed me to get it done quickly, very quickly. And in, you know, like a two day window, all of our notifications have changed and now we only know when we need to look at something instead of

[01:01:28] Adam: Nice.

[01:01:29] Carol: getting noise that everyone ignores. So

[01:01:31] Adam: Yeah.

[01:01:31] Carol: those are things that made me happy to use AI because it's stuff we should do. But you don't wanna take it on 'cause it's so tedious and you know

[01:01:39] Adam: Yeah.

[01:01:40] Carol: you're gonna have to dig

[01:01:40] Ben: Yeah.

[01:01:41] Carol: through so many steps and so many places and it's gonna never end. And it it just kinda motivates me to take on those those efforts. So that's one cool little thing I do with AI.

[01:01:52] Adam: Yeah. Nice.

[01:01:52] Ben: Nice, very cool.

[01:01:53] Adam: Yeah. So I will say, because I always mention it here on the show, and this is a free plug I am pro throwing out. I'm not getting paid for it or anything. It's just a it's a small business I like supporting, but um it's probably not a perfect fit for everything that you mentioned there, but might maybe be a good fit for some of the things you mentioned is so it's Dead Man's Snitch, S N I T

[01:02:11] Ben: Mm.

[01:02:12] Adam: C H,

[01:02:12] Carol: Yes.

[01:02:13] Adam: um, instead of switch. which is i i kind of exactly what you're talking about, right? Instead of like constantly flooding me with information that everything's fine. I want stuff to be checking in, but only get my attention when stuff is not fine.

[01:02:27] Ben: Mm-hmm.

[01:02:28] Adam: that's that's nice. I like that service. So Uh

[01:02:32] Carol: I will go see if it's approved.

[01:02:33] Patreon

[01:02:33] Adam: it. Good luck. so this episode of Working Coad was brought to you by hearing Ben say that he does unit testing and we don't even react anymore.

[01:02:39] Ben: Ha ha ha.

[01:02:43] Adam: Uh and listeners like you,

[01:02:44] Carol: Mm-hmm.

[01:02:44] Adam: if you're enjoying the show and you want to make sure that we can keep putting more of whatever this slop is out into the universe, then you should consider supporting us on Patreon. Our patrons cover our recording, editing, and transcription costs, and we couldn't do this every week without them. Special thanks to our top patrons, Monty, Giancarlo and Peter. We really appreciate you guys. We're gonna go uh record the after show, which uh if you're not acquainted, means that we're gonna the outro music's gonna play and we're gonna keep talking. Uh and the These two slackers haven't put anything in the document that doesn't exist. So I'm just gonna tell you what I'm gonna talk about in the after show, which is there's a TV show that I'm really into. We just started it and I'm like one and a third or one and a half episodes into it, and I'm I'm totally digging it. Way more than I thought I would. Uh and also there's an Easter egg that I put into my photo frame that I want to show these guys. Um so yeah, we'll we'll get into that. that's gonna do it for us this I should before I do that part, I tell you this part. so if you want to help us out and get access to the after show, then the

[01:03:42] Carol: Yeah, see, see?

[01:03:43] Adam: way to do that is to go to patreon.com slash working code pod. You can tell we're total professionals over here. Uh us a little bit of support. We will throw you a lot of bit of after show and our eternal gratitude. Now that's the part where I say that's gonna do it for us this week. We'll catch you again next week. And until then.

[01:03:58] Thanks For Listening!

[01:03:58] Carol: Remember you guys who are still wearing your sweatpants from a week ago, your heart's matter

[01:04:03] Adam: Hey, they're jeans.