The Headless Lifestyle

Coté (00:00.63) I'm thinking of starting a new lifestyle and I don't have a name for it yet, so I'm gonna go with the headless lifestyle. We've talked about the Apple lifestyle. We've got the command line lifestyle, which I think we've all been command-line-lifestyle-adjacent for a while now. It's more like we're the landed gentry, where we've got our servants who go live the command line lifestyle for us and we profit from it. We just kind of hang out in our parlors,

Coté (00:30.762) as they go and command line it out there. But I've been doing some tracing stuff. Since I worked at BMC Software so long, I just can't get it out of me to want to go monitor or observe stuff. So I figured out how to hook it up to Claude Code, and it'll go talk to our Tanzu platform thing and go through this MCP gateway. And then I wrote this MCP server, of course, because I work at Tanzu in Spring.

And all I wanted to do, Brandon, was just see every single thing that went across the wire, and all the conversation that happened. And, you know, that's what you want to do, right? Yeah, yeah.

Brandon (01:08.529) Okay, wow.

Like a Wireshark kind of thing? Let me see — like, all the requests and responses, that thing. Okay.

Coté (01:17.28) Right, right. But then even more — you go into each component and you get everything you can out of the component. Spring Boot apps have this thing called an actuator, which is basically — I don't know — it's like JMX, but not JMX. Same idea: you've got a bunch of metrics. So you can go into a Spring Boot app and say "give it all to me," and it'll just dump it all to you. Now, one observation before I finally get to my headless thing —

Brandon (01:20.509) Okay.

Coté (01:46.253) So, doing this — eventually what you end up with at the end is a gigantic JSON file. You've got the beginning, of me typing something into Claude, and the end, where Claude displays it to me. And then I press the tab key — there's a new layer of JSON, right, going to the initial platform. Press the tab key, and then it goes there. And it goes all the way down and comes all the way back. Just —

I don't know — a trace of stuff. Now, one interesting thing there: I think I had to tell Claude every third prompt, "I want everything." I kept feeling like — is it Gary Sinise? No, wait — which one's the one who wears scarves, plays the burping-and-farting MI5 person? Gary Oldman. I felt like Gary Oldman in The Professional, where he keeps telling people, "I want you to bring everyone!"

And then eventually he yells at them. So you've got to tell it over and over again, otherwise it just keeps clipping things. So then I got this giant JSON file, and I'm looking through it, and — I've looked around for JSON viewers before, and they all seem to be written by people who didn't actually want to look at JSON. It's like trying to find a log viewer and thinking, "I don't think these people actually want to look at logs. They're not into it."

Brandon (03:04.112) Yeah. Okay.

Brandon (03:11.762) That's funny.

Coté (03:12.918) And what I realized is I've got Claude Code. So I just had it write me an HTML page that did exactly what I wanted. And of course it does exactly, exactly what you want — you're like, "no, I want the arrows to point this way," or "how about this thing, how about that." And it got me thinking: I don't need a UI for anything anymore. I just need a command line interface, or if they don't have one, an MCP server.

And then every time I use it, I'll just tell Claude to make me an HTML page of what I want. I don't need a UI — that's ridiculous. I've been thinking through this for over the past week: could I put in place this kind of headless lifestyle, where every time I do something — maybe I've got reusable UIs, but I make whatever UI I want. I don't need to worry about logging into a console. And the other thing I was thinking about —

I guess someone got fired from Google for making usable MCP servers. Exactly. And I have tried it — once I heard about those Google MCP servers, it was at Google Cloud Next, whichever one it was — they said, "we've got MCP servers for everything!" And I was like, "Yay, that's great." But then, not really,

Brandon (04:15.612) Command line interface for Google Workspace.

Brandon (04:29.404) Yeah.

Coté (04:38.904) 'cause you've got to go into Google Cloud and make a Google Cloud organization, and then create an account, and if you want to do this — all I want to do is check my Google Analytics — and next thing you know I have to do this, and then you finally get it all set up and a few days later your token expires, and it's like, "no, don't click on that. Okay, wait, there's that invisible pixel I have to click on that lets me actually click on the thing"

to do it. And it's kind of ridiculous. I've seen the hyperscaler consoles now several times, 'cause I hooked up an Azure thing to make my dashboards, and I don't know if you know this, Brandon, but it's a shit show in there. I wouldn't even — that's the wrong word — it's just, I don't know what it is. It's like a post-apocalyptic

Brandon (05:27.449) Really, this just in? Huh.

Coté (05:36.172) wasteland of UI negligence in there.

Brandon (05:40.541) What is it — is it Conway's Law? Like, the organization ships whatever the org chart becomes. So every one of these consoles — not even picking on the hyperscalers specifically — at the end you're just like, "oh, that's the API team, oh, that's the infrastructure team." You can see each one of these things you have to click into is basically another organization. But go on, I interrupted you.

Coté (05:43.659) I think — probably. Yeah.

Coté (05:55.006) Absolutely.

Coté (06:02.346) Yeah, no, it's terrible. Those UIs are awful. So if you go back to this headless thing — you don't have to care. You've got your little buddy and you're like, "hey, sorry about this, go figure it out. I just want to see this, give me one button to click on," and it'll give you that. It'd be great. So I don't know, I'm thinking about how far I can embrace this way of interacting with stuff. But of course I'm probably gonna hit the barrier

with all those Google MCP things, where I've got to go into a UI. Although, once driving Chrome gets good enough, maybe I won't have to. I did try this — this would be the ultimate use case, maybe we'll actually get enterprise AI ROI and trillion-dollar valuations if I can finally pull this off. I could kind of do it, but it was really slow. I set up Claude and said, "I want you to load up Chrome,

my LinkedIn follows, and I want you to go unfollow everyone, and only add back people you think I want to pay attention to." It did that for maybe fifty people before it got tired and things stopped working. But —

Brandon (07:17.148) I think what you're hitting on — and maybe this will be a theme of this show — is: where we are today, everything that exists for the most part was created with users in mind, and now we're moving into a world where many of us would prefer an agent-based — which is really just a way of saying command-line-only way of doing things. And when we say command line, we mean command line for everything, because I've done something similar. I've successfully created a couple of different

Coté (07:34.264) That's right.

Brandon (07:46.99) things, with the help of Claude, some scripts. I have an alias now on my Mac — I just type "analytics" and it does exactly what you said. It's programmatic: it goes out, authenticates, gets all the data, and creates a simple local HTML file of exactly what I want to see. It's my own little dashboard. And of course, if I want

Coté (08:10.862) That's right.

Brandon (08:12.635) the colors red or blue, I can just infinitely tailor it — make it look exactly how I want. So I think everyone listening who's gotten to this level probably has some of that working. It was a little too complicated to get there, though. And I think where I was going with that is: there's just a lot of stuff missing today. Like, Cloudflare is the same thing — most of the Wrangler API works,

except for a few things, and then Cloudflare will even tell you, "can't do this for you, you need to log in," and it'll tell you where to go, here, here, and here. And that's actually pretty good. I think I read the other day that Cloudflare bought a company — VoidZero, or something like that — and they're building a new command line. Again, the theory is the goal is to do everything. So I think Google Cloud too — and I think maybe this is where that person, for whatever reason, got himself in trouble —

Coté (08:43.329) Mm.

Coté (08:49.23) Yeah.

Brandon (09:11.941) coming at it from a fresh view of "no, the command line is the interface, there's no GUI to log into, we can never ask someone to log in." That design pattern is completely new — most things haven't thought about that. It's always, "we'll just have them make an API key in the UI." And it's like, no, that's not it — the person doesn't know the UI. Like you and I, as we move between services, we

don't know what we're doing in there. It would be so much better if you could just tell the agent, and the agent goes through a well-documented API, and the API can be very protective. I think that's where we are — we're in the transition. Maybe as new services are built, they'll be API-for-command-line-first, and that'll be great for all of us.

Coté (09:41.014) Exactly.

Coté (09:59.671) Yeah — and so, you mentioned analytics. Someone who does well with analytics is our podcast host, Fireside. You do have to go find an API key and make one, but then you give it the API key and it gives you the analytics. Very straightforward. Well — it was you who requested that, wasn't it?

Brandon (10:16.859) But you know what, little history — you know why they created that? You requested it. It's so funny, because they added it, and it's just — okay, though, right? They have something, but clearly not that many other people are using it. Because I think of Fireside as something I use all the time — yeah, I want a full command line interface with everything I have to do inside it. Probably everybody who deals with any content management system —

Coté (10:32.995) Yeah.

Brandon (10:44.891) if you do things a lot in it, that's what you want. And the content management system almost actively fights against API access. It's like, "we don't want you doing this through an API, we want you clicking through things and uploading things." It's weird.

Coté (10:51.341) Yeah, yeah.

Coté (10:58.188) Yeah. Well, I mentioned the WordPress thing last week, right? Their command line tool is great — once you SSH into their servers, I think it does everything. I'm sure there are things it doesn't do, but I'd sit there and be like, "can you make that third nav bar in the footer move over a little bit?" And it would just do it. And then —

Brandon (11:01.051) Mm-hmm, okay. So they're getting there — if everything was hands-off, you'd never have to go in and see the commands.

Brandon (11:22.691) Mm-hmm. That's awesome.

Coté (11:25.25) You can — and I think it's also because of the WordPress architecture — most of WordPress is just stored in a MySQL database. So if you know the schema, you can just go — this is the kind of thing that would make a human programmer keel over in fear — but you can just go to your production database and change stuff. Now, wisely, they have a whole staging environment at that tier that you can do this in first. So that's nice. But yeah.

Brandon (11:37.231) It can just do anything.

Brandon (11:45.979) Change it.

Brandon (11:54.204) But wait a minute, before we get off it — is this the miracle? Should we thank our forefathers at Bell Labs? Somehow they figured all of this out with the Unix command line, or POSIX, P-O-S-I-X, right? They were like, "no, everything inside this operating system is driven by a command." Maybe they didn't even conceive of UIs at the time, so maybe they weren't burdened by that — but because of it, really everything in a Unix-based operating system

Coté (12:06.862) Right, right, right.

Brandon (12:24.227) can pretty much be done from the command line, which in some ways has literally circled back around, where you're like, actually, you know what — they had it right at the beginning, and maybe they didn't think it all the way through, but they still figured it out.

Coté (12:27.276) Yeah, yeah, yeah.

Coté (12:35.438) I think —

I think, from talking with people who do hardcore configuration-management stuff, there's even another level of enlightenment, which is: everything's little commands, and also everything is files. And it feels like — once you get those two ideas in your head, everything makes sense.

Brandon (12:52.016) Yeah.

Coté (13:06.368) It's files and little commands.

Brandon (13:06.427) Right, which in some ways is a throwback to Unix — everything in Unix looks like a file, even a network interface, and everything in Unix is designed to be piped into something else. In some ways the config-management people kind of rediscovered this: "you know what, let's just do that." And all along the way people were saying "I need a UI, it's too hard for me." But now, look at us — as end users we're like, "this UI is too hard for me, get it out of my way. I just want to talk to the

Coté (13:17.806) Mm.

Brandon (13:34.511) robot, the agent, and have it do everything for me. And if the agent can't do it, then I can't do it — I don't want a UI, I want you to make the agent do it."

Coté (13:40.449) Exactly. Yeah, and we could just go to all these software companies and say, "I appreciate it, but I don't want you making me an AI UI anymore. I don't need it, don't do it — just give me a better API." So, the final thing on this: now that I've done all this tracing and understand what — let's see if I get this right — OIDC is, and I know what a JWT is and I've pulled them apart and looked at them —

Brandon (14:01.976) Yeah.

Coté (14:06.264) there's no excuse for the Google Analytics MCP server to be so bad. Right? You could do everything you want to do. If I can log in as a normal human through a web browser to look at my Google Analytics, there's no reason my robot can't do it. You can wrap up whatever funny business you want, but there's some weird decision they're making that's just —

it's the opposite of an Apple decision. An Apple decision would be, "well, there's really only one way to do this thing, and there's no way to get in or out of it except through this tiny little hole of what we want you to do." Good example: if you ever use the Freeform app, it's just a big whiteboard you put things on, and all you can do is export one giant PDF of it. That's it. So —

but whereas I feel like the Google people are probably thinking, "well, they might want to configure this for some different situation, so we're gonna need this setting and that metering." And it's like, no — I just want to see my Google Analytics, I don't want to be a sysadmin to look at this. So maybe they should figure that out.

Brandon (15:14.81) Maybe we should segue a little here — this was one of the stories we wanted to talk about briefly. This person — I'm gonna try his name, I think it's Justin Poehnelt, not sure if we've got that right — was a Google engineer who was fired for creating the Google Workspace CLI. The short version: it's a CLI that lets you do pretty much everything in Workspace, and he made it public on GitHub, and

I guess the people at Google decided he didn't do it the right way, and asked him to leave. It's pretty interesting — I looked into it a little, and it basically constructs the entire command line interface on the fly. It looks at everything you can do, which seems incredibly resilient — "change as much as you want, my command line will always figure it out." So —

Coté (15:56.654) Mm, yeah, yeah.

Brandon (16:09.87) I think he was in DevRel, and I think he's gone on to find another job — maybe the story ends well, I don't know, you'd have to ask the lawyers about that. But I guess the question is — kind of your original question — there are really smart people at all these companies, and a lot of times they know what could be done, and what a lot of people would want done. In this case, he actually did it and put it out, but it ended poorly for him. So is that ultimately a question of —

what's going on here? You can argue it either way. You could say — what do all companies want? They want you, Coté, you know this — they want you to take initiative, they want you to be an owner, they want you to go out and get things done, ask for forgiveness, deliver for customers. We've heard this stuff a million times, right? But then there's the other thing, where they don't want you introducing some kind of legal liability. That's usually where a lot of this comes from. And if you do that,

Coté (16:48.141) Yes.

Brandon (17:06.074) they're usually like, "you're a renegade," and they take back everything they told you before. "Why can't you work the system? Why can't you fit in? Why didn't you ask me?" So you take the same questions and reverse them. This is the dilemma of white-collar work, right? What should I do? We won't get into the long discussion today, but this is that dilemma. I don't know anything about what happened behind the scenes other than that he said he had a supportive manager

who knew this work was going on — so it doesn't sound like he was doing it alone in a room and sprung it on people. It sounds like people knew about it. So ultimately, what would you advise the person sitting at a large company today, who just heard the people on Software Defined Talk say, "I'm gonna build a great command line interface so Coté can achieve his headless lifestyle, and I'm gonna put it out on GitHub." Should you tell that person, "please do that, that'd be really beneficial to

you and the community," or should you say, "maybe check with corporate before you commit and push that"?

Coté (18:07.634) Well — that would be my default answer. However, there's not enough about this story for me to get past the confusing parts of it. One — and I'm just going to assume, I could be wrong about this — but if someone is in DevRel, what you just described is their job: to get stuff into

Brandon (18:32.474) Seems like it, right?

Coté (18:35.64) GitHub — that's part of their job, and it should be fine. Now, on the other hand, a DevRel person should know not to cannibalize the business by putting stuff on GitHub for free — you wouldn't make a free version of something your company's trying to sell. However, on top of that confusion —

I don't think Google's business is command line interfaces. And in fact, with any platform, the more inputs you have into it, the better things generally are — your existing customers get more options for how to use something, and new customers who for some reason weren't using you can now use you too.

If you make it easier to use Google Workspace on the command line, I don't think people are going to pay Google less. Now, I don't know all the Workspace tooling — maybe they do sell tools that a command line interface would make redundant, so they'd lose some line item on a "call us for pricing" table somewhere. But maybe not.

Brandon (20:00.473) Well, just to give you a little more context — the story gets a bit weirder. After he was let go, Google, through the proper channels, officially released the Google Workspace CLI. So they released their own version, built through the proper channels. I don't know — maybe that's part of the story we don't know. Maybe Justin knew this was coming and said,

"you're not doing it right, I'm gonna show you how to do it" — I could see that creating tension, maybe that's why someone had to leave. Or maybe he was just like, "I know you're working on it, but it's not available yet, I'm in DevRel, let me kick something out because people want it." And it did — it hit number one on Hacker News and got a lot of GitHub love. That in itself is hard — plenty of people release products and no one ever notices. So you wonder why —

Coté (20:39.414) Right, right, right.

Brandon (20:59.801) why not have some middle ground, where it looks like Justin has some good ideas, and you have him work with the team and get his ideas incorporated? It's not hard to deprecate something on GitHub and say, "hey, this isn't maintained anymore, the new version is over here — we've graduated from this open source thing to the officially supported Google version, here it is." That seems like it would've been a good outcome. But ultimately, I don't know — he was told he needed to leave.

Coté (21:23.086) Yeah, yeah.

It all leads to — and I don't like to speculate on this kind of stuff out loud, because I don't know — but it all leads toward there being some other problem that's existed for a while, and this was just a convenient opportunity to let someone go. Unless there's some product Google is selling that this would damage, which, as you said, there isn't, because they released MCP servers for it all, for free. That negates that. It's the kind of thing where — I think officially it was because he was using their trademarked colors on the repo. And I've been in DevRel a while, and you've been around it — the way you solve that is you say, "hey,

did you take those colors off your GitHub repo? Thanks for your work."

Brandon (22:27.693) Right — you say, "change the colors," and you agree. Well, as we wrap up this subject, let me just read his one paragraph, so we have it on the record. This is what Justin says — you can read it in his post. Quote: "I think the cause was that Workspace, and certain leaders and projects, were afraid of being disrupted, but the fear wasn't specific to my CLI — it was a broader fear about what agents meant for Workspace. Either way, the irony of my termination was that the announcement of the official Workspace CLI came at Google Cloud Next,

two days before I was fired." So, you can read into that — he's attributing a bit more to corporate politics, saying, "people above me did not like this." And people don't like when their fiefdoms are touched — we've all been in corporations, we've seen that. So this kind of rings true. That said,

Coté (23:11.214) Hmm.

Brandon (23:26.157) maybe Justin was told, "you just need to wait, it's going to be out in a few days," and they really wanted him to do DevRel on the new thing they were creating. You can see the product management side saying, "we can't have everyone creating different APIs all over the company, we're not going to be able to maintain it — we already have that problem, we're going to get a lot more 'killed by Google' posts and people are going to be really mad, so you just need to wait." So maybe there's another side of it, where it's, "you're not making us look bad, but you're making it harder for customers, because now they've got two things to figure out." I don't know where this one nets out. I don't know Justin, and — of course, I always want people to land good jobs. He does wrap it up nicely at the bottom: "nearly seven years at Google was an incredible opportunity for me, and I was fortunate to have wonderful teammates and a manager who fully supported me through these last few months." So, at least in writing, he's going out on a high note.

Coté (23:57.058) Yeah, yeah. Yeah.

Coté (24:23.116) Yeah, well — weird stuff going on there.

Brandon (24:25.785) But I'll just say this — since I don't work for anyone, and for most of you listening: I just want more CLIs. So put them out there. That's better for me. That's what I care about.

Coté (24:35.382) Exactly. There should be as many as possible. No threat, it's fine. Well, I think we've been covering — I think we're almost in week three, maybe this is the end of it — of the Fable nonsense. It looks like, just yesterday or this morning, everything's fine.

Brandon (25:00.569) So give people some dates. That was Tuesday, June 30th — today being July 1st. Probably by the time you hear this it'll be completely resolved, but go ahead, set the dates.

Coté (25:08.632) That's right. So — Anthropic, as always, is like, "this latest thing we did is gonna destroy civilization, here you go, Max subscribers get early access." That was Fable, which was a step down from the previous civilization-destroyer, Mythos or whatever. Then, three or four days later, the US federal government says, "this is too dangerous, you have to shut it down."

Brandon (25:16.302) Mm-hmm.

Brandon (25:26.861) With us.

Brandon (25:37.155) Yeah, export controls.

Coté (25:38.934) Right, right. And then now it looks like it's coming back — I haven't seen it yet, I'm not calling it yet because I'm a registered American, as it were, all my stuff is American, but I'm in the Netherlands and haven't seen it appear yet, so usually —

Brandon (25:50.329) Hehe.

Brandon (25:56.217) Well, I'll confirm — I got up today, July 1st, and I thought they said it'd be back today, so I was going to try it. I just missed it, I guess. As of today, July 1st, I have not seen it — it's not available in my Claude Code, wasn't available there — but the indications are that it's coming. Of course, anything can happen, Coté.

Coté (26:20.8) And I think the main point of all of this is that there's been no explanation of anything, other than just "booga booga booga booga."

Brandon (26:34.723) Well, to be fair, factually — it seems like eighteen days have passed since this started, and from what we've seen, people allude to there having been a lot of meetings. I think we can confirm that. But whether there were actually any changes to the model — I'm not even sure —

I'm not even sure what you can change in eighteen days on one of these gigantic models. You can't retrain it that quickly. I guess all you can do is add more guardrails, which — I think you like to say, Coté — well, now we're just talking about markdown files, right? Anyone can write markdown pretty quick. So maybe —

Coté (27:16.302) That's my thought exactly.

You just put something in your .claude/rules directory and —

Brandon (27:25.655) Right, so maybe they wrote a lot of markdown in eighteen days, which is entirely possible, but they're not even saying that. And then it seems like the government's going to say okay. And in the middle of all this, OpenAI released their Mythos-class model — I don't even know if that's a real term, but you know what I mean — and it looks like they submitted it to the government for pre-review.

So it's almost like — Justin, with his thing, he didn't pre-meet before he released, he didn't play the game. Whereas OpenAI is like, "we pre-meet, we ship after the pre-meeting and the real meeting, we socialize everything." And it looks like —

Coté (27:58.799) I think that —

Coté (28:05.678) I think so.

Coté (28:15.394) You've gotta pre-wire, socialize, all that stuff, yeah.

Brandon (28:21.368) At first they were told to give it a few more days, but then it looks like they got permission. So maybe that's all this is — maybe it's just about having meetings, and that's all we're at.

Coté (28:31.342) I like this, because it matches up with an anonymous quote from last week when we were talking about it — that the Anthropic people just didn't know how to talk to the administration. Which is exactly what you just described — they didn't do the socializing, the pre-meet, the pre-wiring. They didn't realize they had to ask permission to do something they were going to do, and would just be told it was okay.

Brandon (28:42.85) Right.

Coté (28:57.72) But, after much consideration, I think we can safely say it was just a bunch of bullshit — that's probably what was going on there.

Brandon (29:07.032) I feel like — I'd have to go back and listen, but I think even when this first came out, we said something like, "it'll just be resolved by midsummer." We should check the transcripts, by the way — we started adding transcripts to the show now, so if you go to the recent episode you can click the transcript link and, sure enough, you'll see an AI-generated transcript. Don't know — sometimes good, sometimes bad, but —

Coté (29:17.87) For sure.

Brandon (29:34.487) nonetheless, it's like — we all kind of knew this was going to end this way, right? I don't think it took any genius observation to say a couple of weeks would go by and then we'd all have access to this again. Maybe that's bad on us, maybe we shouldn't even bother covering it, since we know how it's going to end. And again, nothing about the technology actually seems to have changed fundamentally — it's just people having meetings.

And going back to the original story, it's still unclear to me — based on reports, it was Amazon who called and said something was wrong. What did they say? And why now — have the people at Amazon who reported the problem been satisfied it's fixed, or did they just have another meeting?

Right? None of this — it's just that the only thing that changed here were people. Nothing about the technology itself seems to have changed, which makes the whole thing kind of comical.

Coté (30:36.28) Yeah, yeah. Meetings. I wrote this — if you want to see some great discussion between episodes, you can go to softwaredefinedtalk.com and find the Slack channel. I wasn't on it, but I was listening to the monthly Enterprise AI Show — it used to be called the Reasoning Show — on TheCloudcast.net.

Brandon (30:58.52) You're of course referring to the Enterprise AI Show, where they did their cloud news of the month, but go on.

Coté (31:04.462) I always look forward to it. I was listening to it, and Brian and Aaron were going through it, and I was thinking — I think we're almost at the plateau of productivity with AI stuff. They even said it later in the episode. And the way you know you've reached the plateau of productivity is it just becomes the same boring enterprise stuff all over again. Not even

beyond SSO and identity stuff, but just — people are concerned about controlling costs, concerned about controlling access to things, starting to wonder about lock-in and how they'll manage that, there's multiple providers you've got to select between. And then there was the announcement from Amazon that they're going to spend a billion dollars on what I'd call post-sales engineers, or consultants, or forward-deployed engineers —

Brandon (31:58.84) Right.

Coté (31:58.848) FDEs — essentially saying that the AI stuff they're selling — which is neither good nor bad — has this very important attribute of enterprise software: you need a lot of people to get it set up and make it work. It's not like — it's just like trying to get programmatic access to Google Analytics, you can't just do it, you've gotta

do a bunch of stuff. So, too bad some of those tens of thousands of people Amazon laid off couldn't just be repurposed for that. But such is life — they didn't have the right meeting. But —

Brandon (32:39.595) Well, I think the story making its way around the internet this week is that Ford — the American automotive manufacturer — is rehiring, quote unquote, "graybeard" engineers after AI fell short. Basically, they hired people back onto their technical team after finding they were still having problems. Again,

I don't want to keep coming back to our previous episode and sound like Ben Thompson, quoting ourselves, but I think we kind of said this — that people are probably moving too fast and making huge assumptions, that they can fire lots of people and everything's just going to work. What was the phrase the executives used — "measures"? Like, "we have all these measures, so we have to get rid of all these people." And it's like, no, you're probably going to learn the hard way that —

Coté (33:25.162) Yeah, yeah.

Brandon (33:35.778) you don't really know what's going on. I just don't know why, especially if you're at a company in a good financial position, there's such a sprint to say "we've got to let a lot of people go," and then it turns out you let too many people go, and now you have to bring them back. I think the Amazon thing is a good example — and if they were here, they'd probably say something like, "well, those weren't the right people." But I have a hard time believing that none of them

were positioned to be forward-deployed engineers, or had skills close to that. It's just frustrating — the way we churn through people's lives. If we could just be a little more pragmatic and think things through a bit — because that's what we always talk about when we talk about strategy, thinking things through — and yet it seems like it just never happens

at all. And this kind of ties back to the previous story — this Fable thing, making these sweeping generalizations, like nothing can be used. Could we just think through what we're trying to accomplish, and actually come up with a plan, with at least a couple hours of thinking, before making all these massive changes? It seems like, consistently, the one thing that's never been true in Software Defined Talk's lifetime is that anyone actually thinks anything through. Even the most obvious decisions are

ignored — even the simplest things, like, "wait a minute, before we do this, let's consider these five obvious questions." Rarely is it actually done. So clearly I'm missing something, Coté. Clearly people think I'm wrong, that thinking through these decisions is just too much work. What are your thoughts?

Coté (35:17.165) Well, it reminds me of a conversation I had with someone at a big company recently. They were saying they're doing a lot of AI there, and — this is what I gathered from the conversation — most of the workers using AI find it great, they're productive. We had the whole, "I'm doing more work

than I used to," just because all the stupid work is now automated — paperwork and things like that — which means there's even more work I can do. There was another conversation I had with people about product management, and they made the point that because their engineers are working so quickly, they're starting to come back to product management asking for more things to do, which —

your product managers must be thinking, "what a world, to have developers coming to ask me for something to do."

Brandon (36:24.331) I know — well, it is, it's great for everybody. To your point, it means one, hopefully if we have a big backlog we were going to get to anyway, we can get through it faster. And if we don't have a big backlog, we can go try a bunch of stuff. Because that's the whole point.

Coté (36:36.937) Exactly. So there was that level, let's say, of labor, of the workers, operating that way. And then the higher-level bosses come in, and they don't really know what to do with that, but they do know they can fire people and spend less money. It's kind of like that's the only button they have — all they can decide is do I press the button or not press the button.

Brandon (37:01.494) Right.

Coté (37:06.197) Right. So the only action that really results from all the meetings is: do I press the button or not? From an executive perspective, it's just getting rid of people or not getting rid of people — that's all they want to know. I think figuring out that bridge — for individuals, it'd be good to figure out. But then I think it results in what you're saying:

it seems like I should be pressing the button right now to eject people, since people are telling me things are easier and they're getting more done, which means I need fewer people, so I'm gonna push the button. That's all they know. There was a related conversation — I forget who this was with — where they said some executive would go off and vibe-code something, and then

they'd say, "I need a place to run this now," and get frustrated because it's not easy in a large enterprise to just run an application anywhere. And they'd start to understand all this platform, DevOps, cloud stuff that's been going on for twenty years, and encounter the opposite of what they expected, where it's just —

Brandon (38:18.73) Right.

Coté (38:32.715) you can get this application running on its own, but now you want it running properly, at scale. I think that also drives the misperception, kind of like the Ford thing — people are more productive, so I should fire them, because I don't really understand what it is they do. All I know is I asked Claude this weekend to make me an SVG diagram of a car, and it did. So —

what do we need that whole building for?

Brandon (39:03.498) Yeah, like, what do we need the graphic artists for? Well — there is a bit of hope, I'll say. I don't have the original link, but someone posted a nice screenshot in the Software Defined Talk Slack, from the Financial Times, and it kind of shows what we've been talking about. It says that companies that have been, quote unquote, high in AI adoption have actually been hiring more. So maybe that's a bit of a counter-trend, and

it's not as popular to talk about — it's not bad news, and bad news sells newspapers. The fact that some companies are using AI and hiring — not tons of people, a few percentage points — maybe that is the real story: that there's some of this normal behavior happening, it's just not very exciting to talk about.

Coté (39:55.639) Can't hear you.

Coté (40:01.772) What happened? There you are — now I hear you, I'm back. All right, those are good edit points. So — what did I say, I'm getting an echo. You've gotta change something off. Ugh.

Brandon (40:03.528) Okay, I'm back. All right, those are good edit points. What did I say — I'm getting an echo, you've gotta change it off.

Brandon (40:17.494) Always good podcasting. I still have the echo. Hold on, let me talk for a second. Nope, still echo.

Coté (40:19.5) What happened, there's still an echo? Okay, hold on, let me talk for a second. Nope, still echo. What do I turn off here — is it the mix-minus thing? If I turn that off, is that good? No? Hold on, let me talk for a second, I'm still hearing the echo. Okay, hold on, I'll switch — it won't let me switch while I'm here. Okay, all right, let's see.

Brandon (40:28.316) It's a mix-minus thing, I'm sure.

Okay, hold on, let me hear — let me talk for a second. I'm still hearing that echo.

Brandon (40:45.163) Hold on.

Coté (40:49.994) Anything? Yeah, I'm still hearing the echo. Okay, if I mute that — well now I can't hear you. No, I can hear you. Do you hear an echo?

Brandon (40:51.024) Yeah, I'm still hearing the echo.

You can't hear me now? No, you can hear me. Let me talk for a second — I no longer hear an echo.

Coté (41:02.806) Okay, weird. Okay.

Brandon (41:06.041) Okay, so hold on — did you hear what I was saying? No? Okay, so I was going over this chart that's in the show notes, basically saying, hey — it's a counter-trend, so let me set you up. Here we go — three, two, one — I'm going to say something and then turn it over to you. You ready? Three, two, one — I think what we have here is a little bit of a counter-trend: some of the companies

Coté (41:09.775) No.

Coté (41:14.54) Yeah, I heard that. I heard that, yeah.

Coté (41:26.958) Uh-huh.

Brandon (41:33.042) are actually still hiring, and it's just not as popular to talk about. They're not hiring in huge numbers, but the people using AI are actually finding ways to hire people. So maybe it is actually happening — it's just not talked about as much as we'd think.

Coté (41:46.285) Now, is this — maybe there's another one, but I looked at a study based on Ramp and Revelio Labs data. It must be the same one, because it had similar findings. I think the charts are correct, but with a lot of context needed. Maybe it's not the Financial Times one, but whichever one I was looking at showed the same thing. First,

it's based on two sets of data. Ramp is like — it's like Concur, I guess, it does expense reporting, primarily business purchases — which, you know, Bank of America probably doesn't use. So it's used by smaller companies and startups, and a large portion of tech companies are startups, so a large portion of it is probably

Brandon (42:23.22) Yeah, it's online, business credit cards. Go ahead.

Brandon (42:29.942) Right, startups. Yeah.

Coté (42:41.962) organizations that were growing anyway. Not that — well, I mean, it is a little dismissive — but they're probably also companies that want to use AI to grow a tremendous amount. Now, I think there were about 21,000 organizations they looked at, which is a good set of data. And the other company they got data from — speaking of LinkedIn — screen-scrapes LinkedIn and other places to see new employees and things like that. So I think —

I think the report's conclusion is that organizations set up to take advantage of AI seem to be doing well with it, and they're actually hiring more people. Which is all to say — as far as large existing organizations go, that data just doesn't exist to know if that's the case there. Back to Ford and the button-pushers — all of that, over-contextualized, does kind of fit with what we were just saying. And I think what we — and probably lots of listeners — have experienced is: the more you use AI, the more stuff you do, the more you get done, and the faster the business moves. Like we were saying, even if developers free up and come back to product management, there's always an endless backlog. And if you have more people who can work on it, and you even hire them, things are probably going to go well. We'll see if it translates. I guess —

I guess the other thing I'd want to see is: assume you're not a high-growth company, and you're on the other end of growth, where you want to reduce costs — does AI help you reduce costs, or not? Based on the graybeard thing, it's probably hard to say, because I bet in a high-growth company, you probably

Brandon (44:17.814) Right, like, what are you doing?

Coté (44:41.196) don't exactly know why things are going well or badly, you're just figuring it out in the moment. But I bet in a large existing company, very few people, if anyone, actually know how anything works at a fundamental level. It just seems like magic in a big box down there, and it's hard to track who does what in the causality.

Brandon (45:03.734) So — I did find the official report. It's called "A New Look at AI's Impact on Jobs," and it's from Ramp. Everyone interested, click the show notes, which you can find on the web or in your podcast player, and view the chart yourself. But I think there was another piece I read before the show, in the New York Times — an opinion piece. To net it out: the person was basically saying that AI

can only, if you will, see all the text we've written and regurgitate it back. But the last piece — and maybe this will be something for the frontier models — is judgment. This idea that

judgment, while it's very easy for a human being to say, "should I delete all the production databases or not?" — it's very clear to us as humans, we'd never do that. Whereas that concept is much harder for an AI. It's this idea of judgment at the end that's very hard, and it's not something the technology is necessarily good at. It can do all the things faster and better than us,

Coté (46:15.982) Mm.

Coté (46:22.488) Yeah, yeah.

Brandon (46:25.695) but it doesn't ultimately distinguish between "this is a catastrophic mistake" and "this is a dev system, it doesn't matter." Maybe that's the final thing — maybe it's kind of boring, but the human being has to be the one with judgment at the end. At least with current technology, it's going to be hard to get around that.

Coté (46:53.932) Yeah, I like — another person I was talking to said something related, to put it in an absurd way. They said you have to be careful the first few times you use AI for development or product management or whatever, because you could tell it you want to make a device for transporting one to two humans, and what you'll end up with is a wheelbarrow. And

Brandon (47:22.293) Right.

Coté (47:22.658) because it'll work — but then you have to tell it, "we make airplanes." There's all this latent knowledge and stuff you intuit, and you realize you have to be very explicit about it. And the good news — depending on your position on this stuff — is that if you spend a lot of time telling it "you're an airplane company," it starts to get good at that.

Brandon (47:47.936) Yeah — and maybe that's the whole point: it's not so much that a lot of jobs get eliminated, it's that the rush to eliminate them is way ahead of what's actually been proven. And maybe that's just the nature of capitalism — you want to be able to say, on your earnings call, "I'm pushing the button" or "not pushing the button." But before we get out of here — since Matt's off this week, I wanted to hear your take: OpenAI's reportedly planning to push their IPO to next year, because it looks like they're not going to hit the trillion-dollar valuation this year. Of course, when you go public you have to make sure you're worth a trillion. The question is, is that a psychological number that actually matters, or are they

Coté (48:30.2) Mm.

Brandon (48:45.681) on the verge of missing maybe the best window ever to go public? What do you think?

Coté (48:51.82) Well, I've for years bought into your theory of share-price valuation, which is that it's story times revenue, or whatever. And yeah, you've gotta go for the trillion. I think that's a legit strategy. Yeah, yeah.

Brandon (49:14.729) You've gotta go for the story, basically. The story isn't as good if it's 780 billion. The story's a lot better if it's a trillion.

Coté (49:19.05) Right. And after all that buildup, when I was looking this over, I was thinking, like you, "why are you waiting? Who cares, money is money." But then I realized there are probably a lot of investors and other people who've been told it's going to be a trillion-dollar IPO. So it's back to the story — a bit circular when it comes to valuation. They were told this is going to be a trillion-dollar IPO,

and so the money they put in was based on valuing things — and doing the strategy — around a trillion dollars. So how else would you get that much money and valuation, right? And then if you IPO at less than that, it doesn't work for the investors, because they're all wrapped up in it. Maybe there should be at least one person in the room at Software Defined Talk Consultancy LLC to say,

"can we go back to when we sold C-A-72 and made thirty billion dollars?" I mean, depending on the size of the investor — this is the investor, and you just say, "hey everyone, remember five years ago when we were going to make

Brandon (50:32.788) They make even more than that. But yeah. Yeah, right, yeah. It's a lot.

Coté (50:44.15) eight hundred million dollars, and that was going to be our crowning achievement — maybe we should take the money and run." Right?

Brandon (50:45.747) Yeah.

Brandon (50:50.472) Yeah, no one's ever gone broke taking the money. Well, on that note — the thing I wanted to bring up, and maybe this goes back to our old friend at SoftBank, what's his name — Masayoshi Son, I can never say it right, but you know, the person who runs the whole thing — we've covered his slides in great depth. The last one I think we did was the unicorn slide, all these unicorns on it. He's released a new deck, and I think, kind of embracing what you're saying about stories —

Coté (51:12.899) Yeah.

Brandon (51:20.006) I think sometimes this is just the ultimate troll — he's trolling all of us. We've all been in business, plenty of people have fancy MBAs, and the slides he puts out are so comical they can only be on purpose. This week's slides were a series about a goose — a golden goose — and then a goose laying lots of golden eggs, with plenty of visuals to go with it. I think the

general theme of the presentation is something like: people are not valuing the goose — which I guess is him and his fund — because it creates all these golden eggs. And the slides are absurd. If you or I made these slides, we'd probably be asked to leave, or at least not given access to PowerPoint for a while. But he's done this so many times that it's like —

back to your thing — maybe his story, his motif, is like the idea of the "crazy executive," or the crazy political leader — who knows what they're going to do. Maybe this is a version of that: the crazy executive who creates crazy things only they could create, and somehow

Coté (52:30.392) Sure, sure.

Brandon (52:43.196) it gets written about. And of course we're talking about it, so we're part of the problem. But in the end, these slides are just nuts. I don't know — having said all that, what do you think of them?

Coté (52:53.794) Yeah, they did look pretty great. We should train some Nano Banana on it and get it to create slides like that. I don't know enough about SoftBank, but my impression is they just take big swings, as they say. And there's probably also this reinforcement loop, like with a lot of the "crazy billionaires" —

Brandon (53:00.852) Mm-hmm.

Coté (53:22.176) by the time you become a sustaining billionaire, you probably figure your entire life was led correctly up to that point, including the failures you learned from, so you should just keep repeating that. I don't think there's — well, there's a negative case, like — what was the Twitter guy, Jack — he was one way for a while, and then he got really weird, and then I don't think —

Brandon (53:47.39) Right.

Jack Dorsey we're talking about.

Coté (53:51.981) I guess Block is in okay shape, isn't that his company? But Twitter didn't really excel after that weirdness, and I don't know, it was strange. Whereas the other people just keep going on the trajectory they've been on the whole time. So maybe in this case, you've got the golden goose guy, and his investors would react poorly if he put out sane slides.

Brandon (53:55.08) Yeah, it's still going.

Coté (54:20.728) They'd be like, "what's wrong with him?" I'm sure there's a sports analogy here — some very polite sports figure, or some very off-the-wall one — and if they show up at the press conference and do the opposite of what people expect, everyone's like, "I don't think this is going to work out next game, something's up with this person." So they've gotta keep living that weirdness. And it's compounded by the fact that they probably don't get out much

Brandon (54:20.777) Right.

Brandon (54:38.281) Yeah.

Coté (54:48.758) and have normal people tell them, "these slides are fucking crazy, what are you doing?"

Brandon (54:56.574) Well — like you said, maybe if you invest in someone expecting them to be eccentric and crazy, they have to deliver that. And as long as it produces results down the road, you can be as eccentric and crazy as you want. Maybe that's the key to creating golden-goose slides. I don't know.

Coté (55:17.046) Yeah. And, of course, we talked about it because it's funny and we have stuff to talk about. But I think, as a normal human in life — and this ties back to the AI thing — I think you have to start calibrating on: I can just ignore all of that. I don't need to go seek out what the current slides are for things. Going all the way back, that's what I was thinking about after listening to the Reasoning —

Brandon (55:31.956) Mm-hmm.

Coté (55:46.467) sorry, Enterprise AI — podcast: I think we're almost at the point with AI stuff where we could just check in at a monthly level and see what's going on. Like with the Fable thing — just give it the Trumpian "two weeks" and it'll be fine, right?

Brandon (55:57.47) Mm-hmm.

Brandon (56:05.908) Seems like it works out. Yeah, maybe that is the whole thing — maybe as we conclude here, everyone should just build their CLIs, deploy them, it'll be fine. Didn't work out so well for the Google engineer, but it's probably fine for everyone else. Just deploy it. And if you really feel like you need to let a lot of people go, maybe just wait a little bit and see what happens if you build some CLIs and see what the team creates first, before

Coté (56:18.158) Yeah.

Brandon (56:35.602) we have to make any huge move either way.

Coté (56:38.444) Yeah, we had vibe coding, token maxing — now we need "just chill maxing." Just let it ride, make a command line interface, and chill out, see what happens. You know, one of the best ways — you might be feeling stressed, might have seen some absurd presentation — there's something in every large organization that's actually quite relaxing, if you

think about it — like, over in Texas, other places, it's really hot out, and you've got these — they're creeks, pretty much, something between a creek and a river that I think needs a better name — flowing bodies of water. You sit in a tube and just say, "I'm gonna go where the river takes me." In a large organization, that thing is bureaucracy. Do we have any of that this episode?

Brandon (57:24.978) Right.

That's right.

Brandon (57:32.387) This week I wanted to highlight Mark, who works over at Databricks. We were talking last week about how it'd be great if there was some way for everything to just route to the cheapest model, so you're not burning through your tokens. He fully admits he works at Databricks, and there's a solution they've announced called — introducing "I'm the Agent," hopefully I said that right.

He says it's open source and free, and it basically does what we were describing. I haven't used it, not necessarily endorsing it, but hey, sounds good to me. If you're worried about using too many tokens, maybe check one of these out — there's also OpenRouter, I think. Someone should do the homework on which one of these is best, then report back so we can tell everyone, "if you're worried about your tokens, use this one." And then, of course —

it's summertime, so I always think everyone needs more stickers. You're probably not going to conferences in July and August — there aren't that many — but you're probably going to the beach, going to the pool, wherever you are, you've got water bottles, you need to put stickers on them, and you need a Software Defined Talk sticker on there. That way, when your family asks what it is, you get to explain what a tech podcast is, and they completely zone out, but it's still kind of fun. So if you don't have stickers, send your postal address

to stickers@softwaredefinedtalk.com and I'll be happy to send you a sticker anywhere in the world. Also, as I mentioned earlier in the show — our new transcription service, Riverside, makes transcripts, but unfortunately it can't handle the accent in Coté's last name — it's misspelled throughout the whole transcript. Maybe we'll find a way to fix that. But if you'd like to look at the transcripts,

let me know if you think they're good — if so, we'll keep doing them, if not, we'll figure something else out. So check all that out.

Coté (59:29.806) Well, there are a lot of conferences. I'm finally going to WeAreDevelopers next week in Berlin, July 8th–10th — my coworker Orrin and I will be speaking there. There's also DevOpsDays Graz, September 4th–5th. Cloud Foundry Summit, where I'll be speaking, is in Heidelberg, September 21st–22nd. We've got DevOpsDays Rockies — there's a discount code for that if you go to softwaredefinedtalk.com slash this episode number, which you can find by scrolling up and seeing that it's 579,

or really any of the past few episodes, since we've mentioned it several times. There's another WeAreDevelopers event September 23rd–25th. DevOpsDays Dallas at the end of September, DevOpsDays Vilnius, and I'll be speaking at DevOpsDays Istanbul on October 24th. I think there must be a tradition where conferences in Turkey are always on a Saturday, from what I can tell — well, maybe just the ones I speak at. It's like,

Brandon (01:00:25.385) Really, is that their thing?

Coté (01:00:29.528) it's got to be on Saturday. In Europe and America you don't really have conferences on Saturday, it just doesn't happen. But that's the one country where I've been to a few that are like that. Anyways —

Brandon (01:00:36.561) No, no, no.

Brandon (01:00:41.459) We know they renamed the country — did you learn this, watching the World Cup? It's no longer spelled like the bird anymore, they've got their own way of spelling it now.

Coté (01:00:45.092) Yeah? My slides? Yeah, yeah, okay.

Coté (01:00:57.994) That's good, okay. And then there's VMUG Connect in Orlando, October 20th–22nd. And I'll also be speaking at Cloud Native Denmark in November — the 19th — which seems like a long way away, but I'm sure it'll happen quickly. Now, speaking of things that are a long way away that you want closer — what do you have to recommend this episode, Brandon?

Brandon (01:01:17.031) Well, this recommendation I'm going to attribute to my wife — we've long been H-E-B shoppers here in town, and we went on a little getaway, not too far from the house, and needed to order some groceries. My wife was lured in by the Walmart+ trial, which offers free grocery delivery on every order over $35. We did it at the vacation place, and it's great, because the last thing you want to do on vacation is shop

Coté (01:01:41.526) Mm.

Brandon (01:01:47.07) for groceries. They brought everything, we had a nice time, had everything we needed, so we're continuing our free month — actually, I think she signed up for the whole year. So now we're trying it at the house too, which surprised me, because she's a true H-E-B loyalist. We're actually going to start ordering some of our groceries from Walmart. We've identified the few things we can't get — the good tortillas, that'll still have to be an H-E-B run.

Coté (01:02:12.492) Yeah, yeah.

Brandon (01:02:12.851) There are a few other things too, but if we can get most of it delivered, I think that's going to beat shopping at H-E-B. We'll see. This is an early recommendation, but I think it holds promise. And of course, what I really want is for H-E-B to adopt a similar format — I know H-E-B bought Favor, and they have

Coté (01:02:23.895) Oof.

Brandon (01:02:40.861) curbside, but I need them to commit to full delivery, an annual price, without marking everything up. That's what Walmart's doing — they're not marking up everything, that's the key. The tortillas can't go from $2 to $3, I don't want that. I want H-E-B to commit to having a good delivery service. In some ways I feel like I'm doing this for H-E-B — I'm betraying you, H-E-B, because I need you to see the light and offer me

a really nice delivery service like Walmart's. We'll see what happens.

Coté (01:03:13.176) So is grocery delivery not a common thing over there?

Brandon (01:03:18.675) It is common, but here in Austin at least it usually comes with a pretty significant price increase. You can do Instacart, which adds a fee on top — Instacart will go almost everywhere, but it's significantly more expensive. I don't want to pay that much more. It's like Amazon, same thing —

Coté (01:03:24.406) Yeah.

Coté (01:03:27.671) Yeah, yeah.

Brandon (01:03:47.219) with free shipping, I feel relaxed ordering — I don't want to sit there constantly seeing everything get tacked on. That's what I want to see from H-E-B. Of course, I'm sure at H-E-B they'd say, "we can't make any money that way." I'm like, I don't know — Walmart's doing it, so —

Coté (01:03:52.502) Yeah, yeah, yeah.

Coté (01:04:01.166) I love that — that's part of the dark art of pricing, where you think you're being all good, like, "I'm bringing you new value, right, customer?" And the customer's like, "you sure are, but I don't want to pay for it."

Brandon (01:04:15.283) Yeah, no — actually, I think it's so interesting, and maybe it's a much deeper discussion around value. It's become so easy to order stuff — I almost feel like — you can contradict this, but — take Amazon, for example. I'm doing Amazon a favor:

I expect delivery, and in exchange, I don't expect Amazon to have any stores at all — I don't expect any retail experience. It's kind of like H-E-B — instead of having all these H-E-B stores, opening more of them — there are probably three stores within a two-mile radius of my house — it's like, listen, you don't need this many stores if you just deliver.

Right, you can just deliver. And a lot of times when you go into an H-E-B now, a lot of what's happening is H-E-B employees doing curbside shopping. It's like, listen, you guys can just make this a warehouse and deliver it to me — because they've got these huge industrial carts with multiple orders on them. It's basically just a warehouse anyway — why don't you just load the truck or van when it's done and drive it to me?

Coté (01:05:13.892) Yeah, yeah.

Coté (01:05:26.161) Yeah.

Coté (01:05:34.434) Yeah, yeah.

Brandon (01:05:35.463) And you don't have to have the cashiers and everything up front — just take all that out. I realize people are going to laugh at this — like, "yeah, I'm doing you a favor by not having a retail location" — but that's kind of how I think of it.

Coté (01:05:46.978) Well, it sounds like what you want is a headless H-E-B — just work on the command line interface, you don't need a UI. Yeah — I ask because here in the Netherlands, grocery delivery is very common. There's no Instacart — the chains, Albert Heijn and Jumbo, do it all themselves. Kim does the buying, but they'll always have something you can buy that eliminates —

Brandon (01:05:49.811) You

Brandon (01:06:04.06) They just do it, it just comes. Is there a markup, or is it all part of —

Coté (01:06:16.765) the fee. Yeah, yeah. And that's why we end up with, like, five bottles of shampoo sometimes — the whole arbitrage trick is identifying items that are everyday items you know will eventually get consumed, or fun items the kids will consume. You don't want to end up with a bunch of Brillo pads or —

Brandon (01:06:17.284) Okay, free delivery if this happens. Okay, perfect, right?

Brandon (01:06:41.874) Well, I think you're hitting on it — I'm totally open to that kind of emotional manipulation, right? Sure, give me the "free delivery" with the shampoo, even though you're making all your margin there — it just makes me feel better. What I can't stand is a markup on every single item — it's like death by a thousand needles. I paid too much for the eggs, too much for the milk, too much for —

find some way to wrap that up so it just makes me feel good — a subscription is another way of doing that — wrap it so I feel good, but definitely don't show me an itemized receipt with everything marked up.

Coté (01:07:20.503) Yeah, yeah. Well, oddly enough, my recommendation this week is hyperlocal — remember when that was a phrase, "hyperlocal news"? We found a new grocery store a couple weeks ago — I've been working on my Dutch, so I'll probably get this wrong — it's called Supermarkt Merwijk Vader Grasmeer. I can never say Grasmeer correctly, but it's a little —

Brandon (01:07:28.422) Absolutely.

Coté (01:07:50.008) they have — what would we call it — an international grocery store, they call them "toko's" — this isn't quite a toko, but an international grocery store means it's got Asian stuff, Middle Eastern stuff, that kind of thing. They don't have tortillas, sadly — when it comes to international grocery stores here, anything Texas-and-below doesn't seem to exist.

Brandon (01:08:07.258) Interesting. Bummer.

Brandon (01:08:18.437) Okay.

Coté (01:08:19.349) Except for Suriname, I guess. But anyway, it's a new grocery store, right next to a giant Albert Heijn, and it's fantastic — very orderly, clean, and nice, which is one of the things you want in a grocery store. You don't want to buy your food somewhere that looks dirty — this place is brightly lit, easy to browse. And they've got three things —

well, they've got three things. One, you can buy a giant bag of frozen shrimp that doesn't cost fifty euros — trying to buy shrimp in the Netherlands is usually very difficult, they want to sell you a little ramekin of shrimp for eight euros. But here you get a giant, Costco-class bag. And they also have an olive bar, and because

Brandon (01:08:55.473) Okay.

Brandon (01:09:09.508) Okay.

Coté (01:09:17.645) it's Middle Eastern, it's actually good olives — not the bad grocery-store olives, actual good olives. But then, Brandon, they sell prime rib. You go to the back, they've got the meat section, and the first time I went — I've been three times now — the first time I walked by the meat section and looked at it, then walked by again,

Brandon (01:09:21.042) Hmm.

Brandon (01:09:24.562) what are you talking about?

Brandon (01:09:34.08) Okay, there you go.

Coté (01:09:46.402) and then finally walked by and the butcher was standing there, and you could tell he knew this guy was coming back for some of this meat. So I was talking to him, saying, "this is really good meat," and he was like, "yeah, I know."

Brandon (01:09:52.658) Ha.

Coté (01:10:02.687) So I got myself some delicious prime rib, and it wasn't even that expensive. I'll put a link to it in the show notes — if you're in the Amsterdam area, you should check it out. They don't have pine nuts as far as I can tell, but they've got a great selection of everything. Their produce section is also top notch — four types of tomatoes,

Brandon (01:10:22.151) Wow.

Coté (01:10:25.785) all sorts of things. It's a great place. And it was great because — there was a heat wave here in Europe, I don't know if you heard — and it's one of those produce sections with the plastic strips hanging down, because it's super cold in there. So I went in for a bit just to hang out and cool down. The Supermarkt —

Brandon (01:10:45.17) Have you thought about telling the butcher, "listen, I've got something for the heat, it's called ice"? I think we can give it out for free, at least small portions. You've gotta float that idea — especially with all the Europeans going, "it's a heat wave!" I'm like, no, that's called Monday, that's what we call Monday in summer here in Texas.

Coté (01:10:56.301) That's right.

Brandon (01:11:13.592) Every time people are like, "there's a heat wave in Europe," I'm like, guys, this is nothing, I don't really have any sympathy for you. But yeah, for all of Europe — guys, ice, it's right there for you, all of you can have it too, it's not that complicated.

Coté (01:11:20.055) Yeah, yeah.

Coté (01:11:26.593) Yeah, maybe I could become the Ice Ambassador. Anyway — Supermarkt Merwijk Vader Grasmeer, go check it out. And, to add — during that heat wave, I saw a distressing number of people driving around with their windows down, and I just wanted to say, "you're doing it wrong." I don't know what you think you're doing — saving gas money or something — but just treat yourself, on the hottest

Brandon (01:11:35.186) All right, I'm checking it out next time I'm there.

Coté (01:11:56.044) days of the year — roll up the windows, pump the AC, and figure out somewhere to drive to. You don't even need a destination. But if it's hot, turn on the AC, that's what you've gotta do.

Brandon (01:12:01.051) Amen, amen. Yeah.

Coté (01:12:24.107) Well, speaking of things you've gotta do — you just did one, which is listen to another episode of Software Defined Talk. This is episode 579, so if you want to see links to the things we mentioned, the conference discounts, get some free tickets

if there are any left for WeAreDevelopers, the discount on DevOpsDays Rockies, and links to the things we recommended and stuff we didn't get to, go to softwaredefinedtalk.com/579. You can also join the Slack channel, do things there. And if you've listened this far and don't subscribe to my newsletter, go to cote.io/newsletter and subscribe. If you don't like newsletters, I do a lot more on my weblog nowadays —

you can check out the RSS feed, bookmark it — go to cote.io/weblog and check things out there, it's good stuff. And with that, we'll see everyone next time. Bye bye.

Brandon (01:13:09.969) All right, Coté, fantastic recording as always. I thought I'd ask you one question in the after-show — since it's summertime, the question is this: do you think the people who run beach resorts, resort pools and things like that, could learn something from technology — specifically, the idea of burstable capacity? I want to talk about it around umbrellas. I think umbrellas are very

Coté (01:13:32.301) Mm.

Brandon (01:13:38.49) inexpensive relative to the resorts they're at, and every time you're out enjoying the water in the hot summer, there's never enough shade, especially around midday — you can see exactly where the shade is, and there's a crowd of people lurking, waiting for someone to leave so they can grab an umbrella. I think — if the umbrella need is, say, ten,

why not buy twenty, just double it? I haven't bought a lot of umbrellas, but it seems like such an easy win — most of the time the umbrellas won't be used, you won't need them, but when everyone wants one, they can just grab them, and you can turn the whole pool deck or beach into a nice shady area. It doesn't cost that much. So

my thought is: that's what the world needs to learn — just buy a lot more umbrellas and have them there. I don't think it increases your costs much, and I think it improves customer satisfaction by an order of magnitude. People should do that. What are your thoughts — is this a good idea, and why don't people do it?

Coté (01:14:48.757) Yeah, I —

I don't think there's much to add except to say yes. In my experience, going places poolside or beachside, yeah, you can always use more umbrellas. And I think if you charged five bucks, or even ten euros, for one, people would buy them. But more to your point — you shouldn't really need to charge, just have a lot of umbrellas.

Even if they cost a couple hundred units of currency, it's probably going to be great. Maybe have shittier couches in the lobby if you're having budget problems, because, man, if you can't find a good umbrella, that's a bummer. No one wants to sit in direct sunlight forever.

Brandon (01:15:38.162) Well, I think it's one of those things — just find the budget somewhere else, don't nickel-and-dime people to rent an umbrella. It comes back to what you're going for: if people leave thinking, "that was a great place, so easy to find shade, we were by the pool the whole time, it wasn't crowded" — or even if it was crowded, it was easy to find shade — that's really all anybody wants. Not that you really notice it consciously, but take that budget from, you know, the little stand —

make the burgers a dollar more, or add fifty cents to the soda — no one will notice — and put that in the umbrella budget. Every time I'm at one of these places I think about it — do the people running it ever think, "if I were managing this resort, the first thing I'd do if I hired someone to run it would be to send them there with their family for a week and just say, stay there, tell me what you think."

I think if you spent a few days at a resort just enjoying it, you'd come away with five easy things you could suggest to make it better. Towels are another example — this might be more of a European thing — one place we went had this complicated system: you took a towel, gave them a card, got the towel; when you returned the towel, you got your card back. And I was thinking, what's the net loss of towels per summer? There's this whole system, and you have to staff people to manage the towels — versus, what if we just had the towels out, with a big bin, and when you're done, you throw them in the bin? How many towels a year are we really losing,

compared to the cost of the whole administration we're putting in place to manage them? Is it really that much money?

Coté (01:17:42.706) I bet there are hospitality conferences with talks and panels about linen management — how should we manage the towels? Yeah, I've noticed that too — the places that are stingy with towels, you have a little less fun. And the places that just have towels available seem awesome at first, and then, in a good way, it's just normal, it's fine, not a big deal. And yeah — I like the emerging "Brandon pricing theory": the customer will actually pay more, just not for that thing. You've got to find something else they'll pay more for — pay more for the burger, maybe the fries cost more, we've already maxed out the drinks so maybe we can't raise those, but —

Brandon (01:18:11.258) Brain.

Coté (01:18:39.393) just figure out something else to charge for.

Brandon (01:18:40.208) There's something else, right, something simple. I've been to so many conferences — if I said, "hey Coté, here's 300 towels you can give away as swag," one, you'd be like, this is such a pain — you'd be standing there trying to give them away, and people saying "I don't want it, I don't have room for a towel." But turn it around: print the resort's name on the towels,

put the URL on there, a discount code — and if someone "steals" the towel, great, now every day they get up and use it and see "resort.com, use code ABC for 20% off." It's literally a piece of swag — if you tried to hand someone a towel on the way out, they wouldn't want it. But if someone feels lucky to have "stolen" a resort beach towel, it becomes marketing.

Right — you're just, this is a marketing expense, fine, that's what I'd say to the marketing team — take a little budget out of swag and put it in the towel budget, and every towel we "lose," we call it swag, done. In fact — I'll give the marketing team credit — every towel we lose, count it as an impression on your web metrics, bring it to your QBR, show what a good job you're doing.

Coté (01:20:07.181) That's right, towels are marketing.

Brandon (01:20:09.306) Done. Well, we've solved so many problems this week. I appreciate everyone standing by and watching us. We'll be off recording for a couple of weeks, but we've got some great episodes coming, so there'll be plenty of YouTube content and podcasts to listen to. We'll be back recording in the middle of July. With that, we'll talk to you all later.

Coté (01:20:30.91) That's right. Bye.