Apr 8, 2026Guest

Vertical slices, Solo, and empathy — product engineering with Aaron D. Francis

Show notes

Kent talks with Aaron D. Francis about product engineering: why ticket-taking implementation is losing ground to agents, what a vertical slice from UI to database really means, and how Aaron’s desktop app Solo came from a painful problem—not a feature spec.

They go deep on scratch-your-own-itch products, separating agents from dev stacks, Jobs to Be Done, why users bring you solutions instead of problems, and how empathy (and letting go of “technically correct”) changes what you ship.

Transcript

Everything we said

00:00

Kent C. Dodds

50 years ago, the, the software developer who really understood the domain and the problem space and the users and the problems, they were way more valuable than the one-

00:10

Aaron Francis

Mm-hmm

00:10

Kent C. Dodds

... who could just take a Jira ticket and turn it-

00:12

Aaron Francis

Yep

00:12

Kent C. Dodds

... into code. That, um, taking a Jira ticket and turning it into code is so much less valuable because-

00:18

Aaron Francis

Mm-hmm

00:18

Kent C. Dodds

... we can, we've got tools for that now. But, um, and, and of course, like, there's the architect and the like, you know, guiding and all of that stuff. But really, uh, m- m- I am of the p- uh, the opinion that we have not seen the end of models improving. I, I think that models will continue to improve.

00:34

Aaron Francis

Sounds like a safe bet.

00:36

Kent C. Dodds

Agent, agents will continue to get better. Maybe they won't. Being product-focused and understanding the user's problems, that was really valuable 50 years ago. It will continue to be valuable. Hey, everybody. It's Kent C. Dodds, and I'm joined by my good friend Aaron Francis. How are you doing, Aaron?

00:53

Aaron Francis

I'm doing great, and I feel like I should correct the record. Aaron D. Francis. If we're going middle initial- ... you know, might as well tell the people. Aaron D. Francis. It's good to be here. Thanks for having me.

01:02

Kent C. Dodds

That's great. Aaron D. Francis. Um, I, uh, sometimes when I say Kent C. Dodds, people think, uh, it's Kency, as in, like, that's just my first name is Kency.

01:11

Aaron Francis

Literally, I only ever call you Kent C. Dodds. So it's a good brand. Whatever you're doing is working.

01:17

Kent C. Dodds

Uh, fun quick story. I've been watching this, uh, true crime, uh, YouTube channel called Mr. Ballen.

01:23

Aaron Francis

Mm-hmm.

01:23

Kent C. Dodds

And, uh, his name is actually Michael B. Allen, uh, but, uh, his Instagram is Michael B. Allen, and so people would DM him and say, "Hey, Mr. Ballen." And-

01:34

Aaron Francis

Amazing

01:34

Kent C. Dodds

... that's where that came from.

01:36

Aaron Francis

Oh, man. I wish I had a cool nickname like that.

01:39

Kent C. Dodds

Yeah. Yeah. Um, well, cool. We're talking about product engineering. Uh, that, in my estimation, is where, um, uh, the, like, the... I- if software development ever goes away as a discipline, the last skill that the last software developer will have that's still valuable is product engineering. Uh, so what, what do you think about that, Aaron?

02:00

What is product engineering to you?

02:03

Aaron Francis

I think that is a directionally correct take. I think we both agree the death of software engineering has been greatly exaggerated. But I think we would also both agree that it's changing, um, and that, like, what we as software engineers, uh, should be doing, uh, is, is morphing a little bit.

02:25

And that's both... It can be terrifying. I think it was terrifying for a while, and now it's exciting.

02:32

Kent C. Dodds

Yeah.

02:32

Aaron Francis

Um, I think we've both gone through the valley of despair, and we're on the other side.

02:37

Kent C. Dodds

Yeah.

02:38

Aaron Francis

Um, and I agree that I think the future is more, uh, I don't... Some- maybe higher level is the right term. I think, um, what we've been doing, both you and I as, as educators, what we've been doing for a long time is teaching people, "Here's how you write this specific little piece of code in a maintainable way.

02:59

Here's how you implement any of those acronyms that Vercel has, ISR, SSR, PP," whatever, all those stuff. Like, "Here's how you do this, like, super low-level technical stuff." Um, and I think the agents have just gotten really good at that, like, rote, like, implement, uh, SSR. Like, they're just, of course, they still make mistakes, and we still need to verify and all that.

03:22

Um, but that level of ticket-taking development, which is what I call it, where you just, like, you sit in the bowels of, you know, AT&T or some bank somewhere, and you just grab tickets off the board and implement it. I think that is, um, becoming less valuable. And what you're calling product engineering, I think is the correct term, is becoming much more valuable.

03:47

And to me, that is, um, we can, like, hold constant the, the talk to people side for a second and just think about, like, the development side. Like, what does this thing look like at the surface, like, where the user interacts with it? How does that go down to, um, you know, maybe the front end, the back end, the database?

04:08

Let me take this whole vertical slice and think about from kind of soup to nuts, what is this thing supposed to do? How is it supposed to work? And then, of course, you layer on the whole, like, well, what do the users want? Like, what, what do the-

04:22

Kent C. Dodds

Mm-hmm

04:22

Aaron Francis

... actual humans want? And that is, I think, where you can go from a mid-level, like, valuable employee to a senior super valuable employee, is you actually are hopefully talking, but at least thinking about, like, what should we build? What do the users want? What do they expect to happen here? And I think that kind of starts to encompass, uh, the product engineering moniker.

04:48

Kent C. Dodds

Yeah. Yeah. And, uh, taking into account the fact that the users might not know exactly what they want.

04:53

Aaron Francis

Mm-hmm.

04:53

Kent C. Dodds

And, uh, I listened to a book recently, uh, titled "Jobs to Be Done."

04:58

Aaron Francis

Mm-hmm.

04:58

Kent C. Dodds

Um, which is the Clay Christensen idea to, um, uh, the, the basic i- uh, one of the premises in there is that the users don't really know what they, uh, what they want, but they do have a job to be done. Um-

05:11

Aaron Francis

Mm-hmm

05:11

Kent C. Dodds

... and so, like, figuring out, um, what that, uh, actual job is and, and really understanding the problem space is a, a tricky thing. Uh-

05:22

Aaron Francis

Mm-hmm

05:22

Kent C. Dodds

... it's, it's hard to learn. Uh, what, what are you currently product engineering, Aaron? What, what are the things that you're working on right now?

05:29

Aaron Francis

So the, the most product product thing I'm engineering right now is a, a desktop application. And the app is called Solo. And of course, it's, you know, ostensibly for AI, but it comes at it from a slightly different angle. And I think this is like a, maybe, uh, an interesting thing to surface was, um, I was, you know, doing the stuff that we all do, got Ghoste open and got 12 tabs

05:59

and, you know, anything you can dream, you can build. So I'm working on three different projects at once. And- One of the things I started to notice was like, well, I've got all these tabs open in different projects, and the agents are spinning up npm run dev themselves, and so they're backgrounded somewhere and I can't see it.

06:18

Kent C. Dodds

Mm-hmm.

06:18

Aaron Francis

And so I think, uh, something we should put a pin in was I had a huge problem that I was experiencing, and I thought, "This sucks and I hate it." And so I had, uh, in the past, built out a Laravel, uh, package for, like, basically like a terminal multiplexer written in pure PHP to solve the same kind of problem.

06:41

And I thought, "Well, I'm not living in PHP anymore. Um, what if I built a desktop app for it?" And so that's what Solo does is, um, you've got all those agent orchestrator tools, which are all great. This is more of a pure, like, a terminal multiplexer agent organization. There's an MCP server so the agents can talk to each other and all that kind of cool stuff.

07:05

Um, but it is very much a scratch my own itch. Um, it separates the agents from the dev stack, so you can, like, spin up your dev stack and then the agents can read it, that sort of stuff. Um, and it's been, uh, a surprise hit, honestly. Like, I put it out and thought, "This will be marketing for other stuff that I'm working on."

07:24

Kent C. Dodds

Mm-hmm.

07:24

Aaron Francis

And people were like, "I just want that, and I want it to do these 50 things." And I'm like, "Oh, shoot, I didn't think of all those 50 things." And so that's also been an interesting journey of receiving all the DMs and the emails and the feedback and trying to, like you said, figure out what does this person actually wanna do?

07:41

Kent C. Dodds

Mm.

07:41

Aaron Francis

'Cause they say they want a button here. I don't think they want a button there. What are you actually trying to do? And so deciphering all of that feedback has been, uh, it's been a lot of fun, truly.

07:51

Kent C. Dodds

Yeah. That, that is a, a really tricky thing, and, and it's come up, uh, time and time again in these conversations is, um, really sussing out what the, the actual need is. Uh, I have an example. When I was, um, really active in building open source libraries that, um, people would use in the React ecosystem, and I had, uh, somebody ask me for a specific prop on a React component-

08:16

Aaron Francis

Mm-hmm

08:16

Kent C. Dodds

... um, to control, uh, whether a certain thing was rendered, uh, or, or, uh, how to, like, it was basically add an if statement to updating state, right? Like, oh, it's just another if statement. You know, you get many, many such cases. Um, and I, in those situations, it's just really good for the person who understands the system to step back and think, like, look at the system holistically and what is something that we could do to change the way that it works foundationally

08:47

to eliminate that problem from the first place. Like, make it so they didn't even have to open up the, the, uh, issue. And, um, I ended up creating what I called the state reducer pattern in the React ecosystem. We don't need to get into it, but it, um, it worked out super well. So that's the key in, uh, uh, in that particular scenario, uh, is figuring out ways that you can create a simpler solution than what they are, uh, they asked for in the first place that completely eliminates the problem

09:18

that they brought to you.

09:20

Aaron Francis

Yeah. It's the same as, like, you know, back in ye olden days when we would write code with our fingers. Um, it's the same kind of deal where you're like, "Okay, well, I need to introduce a new affordance here. I could just bolt it on, or is there, like, a data structure that makes, that unifies all of these, uh, all of these channels or all of these ingress points or surfaces or whatever?" It's the same thing we used to do and, and still instruct the agents to do in the code,

09:50

which is like, hang on, back up. We are like, we're littering state all over the place. What if we introduce a state machine and we just have one place where those transitions happen? It's the same thing when the user comes to you, and they're like, "Hey, can you throw a button on the top right to do this?" And you're like, "Why?" And I got, like, I got most of my, most of my pre-Solo, like, user interaction when I was, I worked, uh, in person many years ago for five years at a local, uh,

10:21

local services firm. It's a property tax protesting firm. And I was building internal software, and I was the only software developer, and there were, like, 15 or 20 people there. And my users were, like, sitting right there, you know? I would build-

10:34

Kent C. Dodds

Nice

10:34

Aaron Francis

... the software, and they would be like, "Hey, we need to, you know, we need to have this." And I would say, like, "What, what are you trying to do?" Was always my question.

10:43

Kent C. Dodds

Yeah.

10:43

Aaron Francis

You know, I'm trying to be kind to these people and not say, like, "Your solutions suck. What are you actually trying to do?" But in my head, I'm like: What are you... Like, w- tell me why this even came into your mind. And they're like: Oh, well, I'm, I need to export. And I'm like, oh, okay. We need an exports concept. We don't just need to tack buttons onto every screen. We need, like, an idea or, like, a container in which we can put all of our export-shaped things.

11:10

Um, and that was just, like, super valuable experience for me because if something was not clear or bad or needed, uh, tuning up, I would, like they would say, "Hey, Aaron, why can't..." or, "How come..." or, "Why does this work?" And I'd be like, "Oh, that's a good point." So I learned a ton in, in that experience that I'm hopefully carrying on till today.

11:30

Kent C. Dodds

Yeah. That, being... This is another thing that just keeps on coming up is being really tight with the user is super valuable. And it, it does change. Uh, if you've got, uh, a million users, that's a lot harder, uh-

11:44

Aaron Francis

Mm-hmm

11:44

Kent C. Dodds

... to do, surprisingly. Like, you've got so many of them, but it's, it's still h- kind of harder to do because you need to find some sort of representative, uh, user, uh, to, to really understand their problems. And, um, I, I talked with Dylan, uh, earlier today, so folks, listen to that episode too. We talked about that a lot with his work on the Vercel domains, uh, API-

12:05

Aaron Francis

Mm-hmm

12:06

Kent C. Dodds

... and all the monitoring and everything he put in place. But I wanna, uh, ask you, Aaron, about the, um- When, when you're talking with the users, what are the sorts of questions that you can ask them to really get clarity on the problem that, um, they're presenting to you? Like in, in that-- I mean, you can-- You're literally sitting right next to them and they say, "I need button here."

12:27

Aaron Francis

Mm-hmm.

12:28

Kent C. Dodds

But like, how do you, um, extract from them the, the clarity around the problem so that you can, um, build the right solution?

12:36

Aaron Francis

Yeah, I think, um, you know, I think trying to get as much information about, like, what is in their head. And so, like, right now, people will DM me and be like, "Hey, in Solo XYZ." And one of my first questions is usually, like, either what, what are you trying to do? Because if you're trying to do agent-to-agent communication and you want me to build this whole thing, well, I like agent-to-agent communication, but in my head, I kind of have this other channel,

13:07

like, mapped out for that, and maybe that's better. And so I think part of it is you need to extract the underlying intent. You also, you have to have a vision for your product-

13:17

Kent C. Dodds

Yeah

13:17

Aaron Francis

... which I think you have to square that circle somehow. Um, and then the other thing is, like, I will often ask, "What did you expect to happen?" Because no matter what kind of vision I have, if all the users are expecting something else, I kind of just need to conform to that in one way or another.

13:36

Kent C. Dodds

Mm.

13:36

Aaron Francis

Or at m- at least, like, honor what they expected somehow. Because something I'm, like, constantly trying to relearn is, as much as I think I'm a very smart boy, if everyone thinks that it should work one way, it's possible that it might maybe should work that way.

13:54

Kent C. Dodds

Mm-hmm.

13:54

Aaron Francis

And that's something that I'm constantly having to tell myself because I thought, "Oh, what a clever solution." But we've done this with code for a thousand years. We wrote our clever solution and then came back six months later and to ourselves we're like, "That guy was an idiot." And so now I'm trying to like, I'm trying to close that loop where everybody is saying, "Hey, when I clicked this, I thought that was gonna happen." I don't want to go into the pattern of, "Well, you know, actually, the reason this is better..." It's like, okay, what is the fundamental mismatch?

14:24

And if I'm really, uh, if I'm really set on that path, well, then I need to change something about how it presents itself so that they expect that rather than have this like, "Well, actually, it should work that way," versus, "I don't care, man, I want it to work that way," and trying to close that gap. So it's a lot of like, what did you think was gonna happen?

14:45

What does it do in your other tools that you use? Like, what are you trying to do? Why are you trying to do that? Those kinds of things to hopefully get below, like... And we can't blame the user for having a bad solution. They know nothing about the architecture underneath. But if we can find out what they're trying to do, we might find a nice, neat home in the existing architecture to satisfy their need rather than bolting on another button.

15:10

Kent C. Dodds

Yeah. Yeah. Absolutely. And sometimes even, like, outside of the existing architecture too, that's when you take a step back and think, "Okay, maybe, maybe the, the decision that I took o- way back when-

15:23

Aaron Francis

Mm-hmm

15:23

Kent C. Dodds

... was the wrong one, and I, I actually need to go this way."

15:25

Aaron Francis

Mm-hmm.

15:25

Kent C. Dodds

And luckily with agents, um, it's pretty easy to kind-- Well, relatively easy to explore-

15:30

Aaron Francis

Easier, yeah

15:31

Kent C. Dodds

... those different, different things. Uh, what you said reminded me of The Design of Everyday Things.

15:36

Aaron Francis

Mm.

15:36

Kent C. Dodds

This is a book by Don Norman, I think, um, where, uh, he talks about, like, how door handles are designed and, uh, different things like does it suggest pull or push? You know-

15:48

Aaron Francis

Mm-hmm

15:48

Kent C. Dodds

... that, simple things like that. Um, and sometimes we have, uh, our own intuition for how things are to work, but, uh, we can't pay ourselves. And so you do, you really do have to, uh, listen to the users. Uh, you said earlier, uh, y- you were talking about how you have this product vision.

16:07

Aaron Francis

Mm-hmm.

16:07

Kent C. Dodds

Uh, and it, it sounds like there's a little bit of contention between product vision and what everybody is telling you, like-

16:13

Aaron Francis

Mm-hmm

16:13

Kent C. Dodds

... uh, they want this thing to be. Uh, I'm curious how you deal with that friction. Uh, and maybe before that, like, what is, what does a product vision mean? Like, what is the vision for a product?

16:25

Aaron Francis

Yeah. I think, um, like, having a point of view on how something should work, not necessarily like at the feature level, but like at the product level. Um, and so like going back to the property tax days, I had a vision for like how this... It was like this CRM, process automation, invoicing, uh, land valuation, this whole thing.

16:54

Just having a vision for like, okay, all of this stuff needs to work together such that, kind of like my North Star was, um, the computers can do literally everything that is computable and then surface information to the human. That was kind of like, this is what I want. I want-- And this was pre-AI, so now the computers can do a lot more.

17:16

Kent C. Dodds

Mm.

17:16

Aaron Francis

But I wanted the, I wanted the system to automate as much as possible, and then when it could not automate anything further, it surfaced relevant, what we would call context now, to the human. So it would be like, "Hey, this person has not signed their contract. It's due on this day. Click this button if you want to send an email. Otherwise, here's their phone number, and you can call them right now." Um, so that is kind of like, that's less of a, "No, it needs to have a left sidebar instead of like top nav," and

17:47

more of a, "Here's where I want to go with this thing." And so for Solo-

17:52

Kent C. Dodds

Mm

17:52

Aaron Francis

... it's less like I want this to, I want this to be, uh, it's, it's less I want this to be the place where people come, and they spin up 15 different worktrees and fire off a bunch of agents, and they never look at it. Um, and that informs the actual features that get built. And so when someone comes to me and says, "I would like to see a git diff in Solo," that fits with what I want it to be.

18:18

Now, how I implement that and how they want it implemented is still an open question. But what I want Solo to be, that fits like the vision that's in, in my head. And if somebody comes to me and they're like- Um, I want, I want you to recreate Cursor. I'm like, "Well, that's not what I wanna do."

18:35

Kent C. Dodds

Mm-hmm.

18:35

Aaron Francis

Like, I don't wanna do that. Cursor's great. Just go use Cursor. And so kinda having some north star of, like, what you're building towards, and then the actual, like, implementation paths then become somewhat flexible. Like, you're just trying to get to the top of the hill. Whether you go up the left side or the right side is more of an open question.

18:57

But if you start just taking everybody's feedback and implementing it as if it's gospel truth, then you end up with something that's bloated and doesn't do anything the way that anybody wants. It's much better to have, like, a point of view that excludes a lot of people, but really includes, like, the narrow slice that you're going after, instead of, like, something that's so broad that everybody's like, "Well, this is about 40% of what I need." 'Cause then nobody uses it.

19:24

And so trying to square my vision with what people want and separating that slightly from specific feature implementations.

19:35

Kent C. Dodds

I think that is really smart, um, because, like, uh, okay, so we've got X.com, the everything app.

19:43

Aaron Francis

The everything app.

19:44

Kent C. Dodds

Well, I mean, it's not actually the everything app.

19:46

Aaron Francis

No.

19:47

Kent C. Dodds

May- maybe, maybe they want it to be eventually.

19:49

Aaron Francis

No.

19:49

Kent C. Dodds

But, like, I'm not gonna use the everything app to, like, I don't know, write my software or run my terminal commands or, like-

19:55

Aaron Francis

Or pay your friends. Like, we've got-

19:57

Kent C. Dodds

Yeah

19:57

Aaron Francis

... apps for that.

19:58

Kent C. Dodds

Yeah. So, um, I think e- especially when you're getting started too, a- and with you, I imagine-- I, I mean, maybe in the future you wanna hire people to help you with Solo, and, like, it just continues to grow. But certainly right now, uh, you wanna maintain that, that vision.

20:14

Aaron Francis

Mm-hmm.

20:14

Kent C. Dodds

And you mentioned, like, you wanna get to the top of the hill, whether you go left or right to get there is fine. But, like, if you just go ahead and implement every feature, you're not gonna get to the top of the hill.

20:24

Aaron Francis

Nope.

20:24

Kent C. Dodds

You're just going, doing circles around it the whole time.

20:27

Aaron Francis

Yep. Exactly.

20:28

Kent C. Dodds

And, and on top of that, um, if you do just implement every possible feature, um, then, like, yeah, sure, you're gonna slow down. But it's the agents. Okay, whatever. Maybe you don't slow down. Maybe you're just as fast. But the, the onboarding experience and the, the usability of this thing, it just becomes so cognitively difficult for humans to use.

20:52

Um, and, and agents as well. Uh, you know, with that, that context does bloat. And so, um, yeah, so when, when the users come to you and say, "Hey, I want this, I want that, I want you to build Cursor," um, having that north star just really helps you to say, "Oh, that doesn't quite fit, but, like, let me try and get to the core of why you want this to be Cursor."

21:15

Aaron Francis

Right. Yep.

21:16

Kent C. Dodds

And is there something that I can do that is still in line with my vision that would address the, the challenges you're having? And maybe there's not, and that they're outside of that slice you're talking about.

21:27

Aaron Francis

Yeah. There's, like... People have come to me and said, "I need you to build in work trees." That's, like, a very clear, discrete request. And my answer's like, "One, I don't use work trees," right?

21:41

Kent C. Dodds

Same.

21:42

Aaron Francis

I don't, I, I don't, like... I am still a h- a single human, and it-- I find it hard to even, like, reason about the same app in six different slices. Like, I can, I can much more easily be like, "I'm gonna work on this app, and while it's cooking, I'm gonna work on that little side project, and then my personal website." That's fine for me.

22:02

And so, first of all, I don't use the thing that the people want me to build, which is already, like, a red flag to me of, "Hey, man, you don't, like... You're not gonna have the touch," right? You're not gonna have the, "I know which parts are prickly and which parts aren't," 'cause you don't use it. And so that already raises a red flag to me. But then you go, like, you go further into the process and you say, uh, "Hey, I, I don't use them.

22:25

What are, what are you trying to do with it?" And it's like, "Well, I just wanna be able to have, you know, changes not stomp on each other." And you're like, "Ah, okay. Well, that's different. Um, and so what about, like, discrete checkouts? Like, what if you just checked it out again? How does that work for you?" And they're like, "Well, then I have this problem with node modules bloating." And I'm like, "Okay. Well, that's interesting.

22:45

Like, what else have you tried?" And then finally we can get to somewhere where it's like the nugget of the problem is you wanna be running multiple agents on one project. Now, is that something that I need to figure out how to let you do? I don't know. Kind of an open question. But the request was, "I need work trees," and the reality is, "I just wanna run multiple agents safely without them stomping on each other." And those are different things.

23:10

One is a solution-

23:11

Kent C. Dodds

Yeah

23:11

Aaron Francis

... and one is a problem. And so I'm always trying to get to what's the real problem here.

23:16

Kent C. Dodds

Yes. I, uh... That's, that's a really big takeaway for everybody listening right now. Users will often just naturally come at you with solutions. Um, a- at least this is true of technical users-

23:28

Aaron Francis

Mm-hmm

23:28

Kent C. Dodds

... and maybe in some cases non-technical users. But I think I, uh, speaking for myself, um, I just have this natural tendency to be like, "If they would just do it this way," you know?

23:38

Aaron Francis

Mm-hmm. Yep.

23:39

Kent C. Dodds

Then it would be so much better. But, uh, without all the, the context of the, the bigger solution, um, then I'm as just hallucinogenic as, uh, the AI that I'm- ... using to help me with this stuff.

23:52

Aaron Francis

Yep.

23:52

Kent C. Dodds

Um, so yeah. That's a really, really good point. Trying to-

23:54

Aaron Francis

And it holds for non, non-technical users. I mean, at the property tax firm, that was the export button. It was like, "Can you just put an export button here?" And I'm like, "We don't even have exports. Like, I don't even know what you're trying to do." And in reality, it was just like, "Oh, I just need to download, you know, these five properties." And it's like, oh, okay. Well, maybe export fits there, but maybe there's just, like, a, a better way to do that.

24:17

So everyone comes with, "Can you just do this?" And that's when you're like, "Well-" Why?

24:24

Kent C. Dodds

Yeah, yeah. Well, and maybe, like, I need to export this so I can do that. And you're like-

24:28

Aaron Francis

Mm-hmm

24:29

Kent C. Dodds

... "Oh, actually, I could make that way easier than an export button." Uh, you know.

24:32

Aaron Francis

And we did. Oh, I gotta tell you this. We totally did. We, um, so I did build CSV exports, but I also built live Google Sheet exports, so they could do, uh, you know, the equals import data, and then give it this signed URL, and all the data would show up in Google Sheets, which is what they actually wanted. And so if you di- if you dig far enough, you'll find something that is, like, maybe even way better than what their proposed solution is.

24:58

It was actually really cool.

24:59

Kent C. Dodds

Uh, that i- that is a great story. Yeah, I appreciate that. Um, I, I really liked what you said about, um, when somebody ca- like, they came to you asking for work trees, and you're like, "I don't know, man. I'm not the right person to build that."

25:12

Aaron Francis

Mm-hmm.

25:12

Kent C. Dodds

"I, I don't use that." Like, that's not part of my workflow. And I had situations like this all the time, uh, as I was working on open source libraries too. Like, uh, especially when I built this library, I was using it actively, and then I move on. I'm not doing an AngularJS anymore or whatever. And, um, people are asking me for features and stuff, and I'm like, "Uh, I am the least...

25:35

Like, I'm the worst person to solve this problem."

25:37

Aaron Francis

Mm-hmm.

25:38

Kent C. Dodds

"Um, would you like to maintain this library now? Like, you are way better because you have the problems." And handed off a lot of libraries that way.

25:45

Aaron Francis

Yep. There's just something that, like, oh, you just don't get if you're not in the scene, right? Like, you-- Everybody has a, a brother or a cousin or an uncle or somebody that knows a guy who knows a thing or something. And, like, whenever you're, when you're like, "Hey, man, I wanna, like, I need to, I need to get a new car." And he's like, "Ah, I got, I got a car guy." That's kind of how it is.

26:08

Like, I'm not a work tree guy. I'm not in the scene. I don't know the, like, oh, definitely do this. Don't do that. Oh, this is gonna bite you hard, or this is gonna lock up your, your Git index or whatever. Like, I just don't know it. And so not being in the scene is really hard, which is not to say that, like, you can only build things that you use. But I do think if I were to go out and try to build something for dentists, I would have a really hard time because that's just not my world.

26:35

I don't know the language they speak. I don't know the problems they run into. I don't know why they're furious with their software. I'm sure they are, but I don't know why, and I don't know how to fix it. I've never worked in a dentist office. And so I think there's some portion of, like, you kind of have to be embedded. And this goes back to, like, having a point of view. I can't have a point of view on how dentist-

26:55

Kent C. Dodds

Yeah

26:55

Aaron Francis

... scheduling software should work, 'cause I've never been a front desk, uh, worker at a dentist office. And so I think that can be great news because if you're embedded somewhere weird, like I was in a property tax firm, and I'm, you know, we made that property tax firm incredibly profitable because there was n- there were no software engineers embedded in property tax firms.

27:18

And so, like, if you're listening and you're thinking, "Well, my, you know, my, uh, aunt and uncle run a, a window washing, you know, company that makes a million dollars a year, and they do it all, all on paper." Well, go talk to your aunt and uncle. Like, there's a huge opportunity, I think, for bespoke SaaS for people that are embedded in places that have not traditionally been worth touching from a SaaS perspective.

27:46

Kent C. Dodds

Yeah. A-and I, I'm glad that you called this out because, um, what I think some people are, are... um, might mistakenly take away from this is, okay, well, I just have to work at a developer company, like a company building for developers, because that's all I know is I'm, I'm-

28:02

Aaron Francis

Mm-hmm

28:02

Kent C. Dodds

... just a developer. But, uh, I think that if you, if there was an opportunity to go and handle scheduling for dentist office, like, uh, that, that's not just a, oh, well, definitely not, 'cause I don't know anything. That's a, here's an opportunity to go learn-

28:17

Aaron Francis

Mm-hmm

28:17

Kent C. Dodds

... uh, about that. And what's, what I find really interesting about all this, and I said that, I think on every, uh, episode so far, um, nothing has really changed. Like 50 years ago, the, the software developer who really understood the domain and the problem space and the users and the problems, they were way more valuable than the one-

28:39

Aaron Francis

Mm-hmm

28:39

Kent C. Dodds

... who could just take a Jira ticket and turn it into-

28:41

Aaron Francis

Yep

28:41

Kent C. Dodds

... code. And, um, that has just, um, accelerated, uh, or increased now that, um, taking a Jira ticket and turning it into code is so much less valuable because-

28:53

Aaron Francis

Mm-hmm

28:53

Kent C. Dodds

... we, we've got tools for that now. And so, um, and, and of course, like, there's the architect and the, like, you know, guiding and all of that stuff. But really, uh, m- I am of the p- the opinion that we have not seen the end of models improving. I, I think that models will continue to improve.

29:10

Aaron Francis

Sounds like a safe bet.

29:11

Kent C. Dodds

Agent- agents will continue to get better. Harnesses will get better. Maybe one day they'll be able to do the architecture stuff. Maybe they won't. Either way, being product-focused and understanding the user's problem, uh, was really valuable 50 years ago. It will continue to be valuable. It w- And as I said at the beginning, it will be the last valuable skill that a software developer has before the industry just goes away, if it does.

29:36

Aaron Francis

Mm-hmm. Yep. No, I agree. I think, um, and I, I think that that speaks to hopefully a proliferation of, you know, micro SaaS is if you can be the engineer that can speak to the people, well, then you just wield these tools that have made everything easier.

29:57

I mean-

29:57

Kent C. Dodds

Mm-hmm

29:57

Aaron Francis

... I think a, a big reason we've had monolithic SaaS for so long is, uh, you know, people say code is not the bottleneck. I don't know, man. It's, it's taken a long time to write a lot of this code historically.

30:10

Kent C. Dodds

Mm-hmm.

30:10

Aaron Francis

And now it doesn't. And so I think, you know, Salesforce is never gonna come down to the level that would make, you know, a two or five-person company a lot of money. They're just not gonna-

30:21

Kent C. Dodds

Mm

30:21

Aaron Francis

... do it. But now with all these tools, uh, you can do everything a lot faster, which is a double-edged sword. You can do the wrong thing incredibly fast and-

30:30

Kent C. Dodds

Yeah.

30:31

Aaron Francis

... feel like you're making a ton of progress. And so I think that is I think that is the skill that a lot of us are learning right now is, how do I do the right thing incredibly fast or at least directionally correct rather than, uh, do 10,000 lines of code a day on the wrong freaking thing? And so talking to like figuring out painful problems that human beings have, hopefully human beings with money, and then using these new like superpowers

31:01

that we have to hopefully solve people's problems. They get a lot of value, you get a lot of money, and then everybody-- like the whole world becomes more efficient. I, I know that that's an optimistic take on the impending doom, but I think it could be true, right? It could possibly come to fruition.

31:17

Kent C. Dodds

Well, it's, it's certainly not gonna be true if we don't believe that it's possible.

31:21

Aaron Francis

Correct.

31:22

Kent C. Dodds

So-

31:22

Aaron Francis

Exactly.

31:23

Kent C. Dodds

So we gotta manifest that. Um-

31:25

Aaron Francis

Yes.

31:25

Kent C. Dodds

A-and I, I completely agree with you that I, I think that, um, the, the big players will continue to big- be big players.

31:33

Aaron Francis

Mm-hmm.

31:33

Kent C. Dodds

Um, but, uh, but I do think that, uh, bespoke software, um, and smaller teams are a, a big part of that future. And so if you're listening to this and you're like, "You know what? I actually do have this really strange obsession with dentistry"-

31:51

Aaron Francis

Mm-hmm

31:51

Kent C. Dodds

... then like there's a really big opportunity for you, uh, to go and AI-ify the dentist's office or, or whatever. Don't, don't just go AI-ify it. Go and ask them questions, interview them-

32:04

Aaron Francis

Mm-hmm

32:04

Kent C. Dodds

... and, uh, say, "Hey, listen, I just wanna understand your problem."

32:07

Aaron Francis

Mm-hmm.

32:08

Kent C. Dodds

Uh, "What is, what is your problem?"

32:10

Aaron Francis

Yep. The whole, the whole world runs on, uh, executive assistants and secretaries that have been at their business for 25 years. That's how-

32:20

Kent C. Dodds

Hmm

32:20

Aaron Francis

... the majority of small businesses work. Like you go to a plumber, you go to a title company, you go to a real estate office, and nine times out of 10, there is one person who's probably underpaid keeping the whole thing in their head, right? I have my-

32:37

Kent C. Dodds

Hmm

32:37

Aaron Francis

... my favorite HVAC company, which is horrible to say 'cause my HVAC keeps breaking so much that I have a favorite HVAC company. The whole thing, the whole thing is literally run by this woman who is the owner's assistant, and she's truly been there for 30 years, and everything-

32:55

Kent C. Dodds

Hmm

32:55

Aaron Francis

... is in her head. And those companies are in every single town across every single country. And you go in there, and it's just like files stacked up on their desk, and they know where every single one is. But you can't, you can't argue that that's more efficient than some dinky little piece of crud software built specifically for them that could make them super efficient, and you could probably rip out in, you know, two or three months with the back and forth of like actually understanding what they're doing.

33:25

And that is just like a huge opportunity that has historically not existed. I think the question is how do developers get into those spaces? And that's still like, that's still hard, right? If you don't-

33:37

Kent C. Dodds

Hmm

33:37

Aaron Francis

... have an in, like if maybe you come from a long line of dentists and you're the black sheep that's a software developer, well, let's, let's like rely on that. Go talk to all your, you know, siblings and parents or whatever that are dentists. But I do think the software-ification of the world will continue. It's just finally gonna start bleeding into these little micro niches, um, which is an opportunity.

34:01

Like don't, don't go out and try to build, uh, for example, Cursor. When everybody's trying to build Cursor, go find some $1 to $5 million company locally and be like, "Hey, I think I worked here in high school. Can I just like come shadow y'all for a week and see what's going on and see what happens?"

34:20

Kent C. Dodds

All right, Aaron, but what, what about all the introverts who don't wanna talk to people? What, what do they do?

34:25

Aaron Francis

Oh, that is me, first of all, so I feel, I feel your pain. Um, I think you have a couple of options. One is, um, one is you, you partner up with a friend. Um, you know, there are, there are people that like to talk to other people. Um, those people do exist.

34:46

And so I still think the, you know, the sales, sales guy or girl, developer guy or girl as a team, that's awesome. And frankly-

34:55

Kent C. Dodds

Mm-hmm

34:55

Aaron Francis

... I think that can work super even better now, now that the developer half of that team is basically like 10 people if you get the right-

35:03

Kent C. Dodds

Yeah

35:03

Aaron Francis

... you know, tokens. So I think that's still totally viable. Um, and in fact, you could probably find your, uh, domain expert on the non-technical side of the house. Like-

35:15

Kent C. Dodds

Yeah

35:15

Aaron Francis

... who's your, who's your best friend that you went to high school with? Where does he or she work? What do they do? Call 'em up. Maybe, maybe now is the time for you to finally start that business together that you've talked about since high school or whatever. I think another, another option is, um, you just kinda treat it like a engineering problem. Like you treat it as a fact-finding mission and not like-

35:38

Kent C. Dodds

Hmm

35:38

Aaron Francis

... a networking mission. And I think reframing it slightly in your head to, there is a pot of gold out there somewhere, and my job is to systematically eliminate and find, you know, the, the right options, and I'm gonna treat this like, I'm gonna treat this like a spreadsheet. I'm gonna, you know, talk to all these people that I know and reach out on LinkedIn and get lunch with...

36:01

I mean, there's nothing that old people on LinkedIn want more than somebody that is younger than them to say like, "I just wanna learn about your industry." Ah. People on LinkedIn love that crap. And so especially if they're like above 60 and you're below 40, they're gonna think, "Ah, the next generation." And so- ... just treat it, treat it as like a, I don't know, like an optimization mission of there's something out there for me, and my job is to go uncover it.

36:27

Um, and then just face reality that you're gonna have to talk to a few people, but on the way to some mission that you actually want. I think that framing could be helpful, but, uh, finding somebody that's super outgoing and good at sales that you implicitly trust, that's the real move. If you can pull that one off, that one seems a lot easier. Yeah, I completely agree. I, I actually, I had an opportunity to do exactly that, um, not for dentists, but for, uh, cosmetology schools- Yeah, there you go ...

36:56

things. And, uh, I, I didn't pursue that because I just love being an educator- Mm-hmm ... and I decided to go this direction instead. Um, I, I recommended a bunch of my friends to, um, to, like, pursue this idea because it is, um, like, yes, okay, agents are gonna be able to build everything, but this guy has so much experience and knowledge- Mm-hmm ...

37:20

in this, like, really tight niche that, um, yeah, of course, like, w-we can build a company. We will be the AI that puts this, you know, this industry- Yes, exactly ... on its head. And, and because we have all of this knowledge and experience, we can encode that into the- Mm-hmm ... AI that we build, and that's gonna be better than whatever some Joe Schmo throws together- Mm-hmm ...

37:42

on a Replit, you know? Mm-hmm. Exactly. And the connections that your friend has. Yeah. He can walk in any door and be like, "Hey, I've been in this industry for 20 years. You know me. We did this, we did that. Hey, what's up?" Um, and that kind of stuff, like, when OpenClaus are sending a billion emails a second, having a human that knows another human is gonna be pretty powerful.

38:03

And so even that is like, if you're embedded in an industry or you've got a friend that is, that is becoming increasingly valuable. Yeah, absolutely. Um, so working with developers or building a product for developers, um, relatively straightforward if you're a developer. Like, you kind of- Mm-hmm ... have a good idea, uh, and you've got connections, you got people who are willing to use it.

38:25

Developers often are willing to put up with stuff as long as we are aware, like, this is alpha software. Okay, great. I expect bugs. Working with non-technical people is a little bit different. You were really fortunate to be in the room with the people- Mm-hmm ... as they were using the software. What would you say to somebody who is not? Um, maybe they, uh, are building...

38:47

Like, if they were building internal software, then they can go do that. But if they're building something for millions of users that are- Mm-hmm ... non-technical or technical- Mm-hmm ... um, do you have any advice or, or ideas of how somebody can develop a product-focused mind, um, in a setting like that? Um, I don't have any advice for anybody that's building something for millions of users, so I'll just say that upfront.

39:10

Never done it, wouldn't, uh, be fair to opine on it. I think generally speaking, um, having, uh... I, I find that it's nice when, like, the, the right thing to do and the strategic thing to do are the same thing, and I think this is a case where, um, being, like, empathetic towards the user is incredibly valuable because you have to imagine that, like, truly, like imagine somebody is trying to

39:40

do their job and go home, and they're as stressed out as you are about the world, the economy, the kids, baseball practice, whatever. And your tool is the surface that they spend most of their day in. And so when they- Hmm ... send you an email, it's like, "This thing freaking..." And you're like, "Okay. Well, that hurts my feelings, first of all." Like, you can say that. I'm, I'm frustrated.

40:02

And then after you calm down, think about like, okay, this person is just like me. They just want to do their job and go home at the end of the day. And like, how can I put myself in their shoes where I was trying to do this workflow, and then I hit the back button, and you didn't have, like, an exit intent, and I lost all my work.

40:22

Yeah. That happened to me, like, an hour ago. It was very irritating. Yeah. It's frustrating. Super frustrating. And then you just kind of like, you have to develop that, um, that empathy for the other humans that are actually using the thing. And that can be, like, proactive empathy. You can think, okay, well, if they're, like, at the end of this form and they hit back, they should know they shouldn't hit back.

40:46

But out of, out of love and care for the user, I'm gonna give them a little warning here instead of something that I think engineers can fall victim to is, well, you should've known. Like, that's how it works. Yeah. That's technically correct. And I think you kind of have to, like, start to let technically correct go out of your mind as you move into- Yeah ...

41:06

product and away from, uh, pure implementation. A unit test that is technically correct, oh, that's great. That sounds awesome. A product that is technically correct but makes everyone furious doesn't matter. You lose. Game over. And so I think moving more towards the product side, you have to think, all right, well, it's not technically pure, but it is way more usable if I surface this thing in two places.

41:31

Even though it's still just one model entity or whatever, if they're looking for it here, yeah, maybe I should put it there as well. And so having that sort of like, uh, human-to-human interaction, even if you're not sitting right next to them like I had the benefit of. Yeah. Yeah. Uh, it, it does require a, a significant amount of, uh, empathy for the user.

41:52

Uh, and um, actually, it, it makes me think of that, uh, "Design of Everyday Things," uh, book again, where he, he says that it's just really easy for us to, um, to say, "Oh, well, like, just read the manual," or, uh, human error, uh, sort of thing. And actually, even the human who made the error kind of assumes that they were in the wrong, and they, they admit like, "Yeah, I shouldn't have done that.

42:16

Like I know it's, you know, you're supposed to right click, not left click," whatever. Mm-hmm. Um, but a, a product engineer sees things like that and they, they assume fault. They assume responsibility, and they say, "I'm gonna make it so it's impossible for the dumbest user to use this wrong." Yep. And I don't, I don't know if it's true or if it's apocryphal, but there's a story, and I'm gonna say it as if it's true.

42:38

There's a story about, uh, Disneyland or Disney World, and people kept cutting across the grass. And so the, the maintenance person came to the big boss and was like, "Hey, can we put up a fence here? 'Cause people keep cutting across the grass here." And the big boss is like, "No, we need to put a sidewalk there. We need to pave it." Like, that is what the people are expecting to be a reality. They think at, to get from A to B, I walk this way.

43:02

And I think those are maybe called cow paths, where it's like- Hmm This is where the cows walk. This is where the people go. And so instead of, like, erecting these barriers because it's technically correct that the sidewalk goes that way, we should give them an affordance to do the thing that everyone is naturally trying to do. Let's make it easier instead of saying, "No, there's a fence here.

43:23

Technically, you gotta go that way."

43:25

Kent C. Dodds

Yeah. That's, that's very good. Aaron, as we, we wrap up here, uh, I, I'm gonna ask you here in a second for one practical thing that people can do to improve their product sense. Um, but before we get to that, is there anything else that we didn't really touch on that you were kind of hoping we would?

43:42

Aaron Francis

Um, no, not really. I mean, I think between the two of us, we're as optimistic as they come, and we've both had our, our doom days. But I really do believe that, like, the industry is changing, but it's not dying. Um, we just have to, we just have to adapt, but that's okay. That's what we've been doing since you mentioned Angular, which is a thousand years ago.

44:03

Like- ... we've just been doing this for forever. Um, and it is a little, it's a little hard to be like, "I guess I'm not a, um, software craftsperson anymore, but maybe I could be a product craftsperson."

44:17

Kent C. Dodds

Mm-hmm.

44:17

Aaron Francis

And that feels pretty good to me. Um, and so I'm sympathetic to the people that are dooming and are like, "I love, I love to code." Um, and I think, I think there are other things that we will grow to love as, as time progresses on. But, um, it's definitely topsy-turvy, but I feel a little bit optimistic.

44:36

Kent C. Dodds

Yeah. I think that's good. And, and I think that the thing we should fall in love with is the problem. Like-

44:42

Aaron Francis

Mm-hmm

44:42

Kent C. Dodds

... fall in love with that problem. That was very clear from your, like, how you created Solo, is, um, you, you loved the problem. Well, you actually-

44:51

Aaron Francis

Mm-hmm

44:51

Kent C. Dodds

... hated the problem, but-

44:52

Aaron Francis

Mm-hmm

44:52

Kent C. Dodds

... but, uh, but in, in such a way that you decided to, to produce a solution, um, that is actually making you money, uh, right now.

45:00

Aaron Francis

Mm-hmm.

45:01

Kent C. Dodds

You, you've got customers who actually care, and-

45:03

Aaron Francis

Mm-hmm

45:03

Kent C. Dodds

... that is a really special thing. Uh, and you-- I think you only get customers who care, um, when you care for the customers. Um, and so yeah, thank you for sharing some of your insights-

45:15

Aaron Francis

Mm-hmm

45:15

Kent C. Dodds

... and wisdom with us, Aaron, today. Uh, what's the, the best place for people to keep up with what you're doing and, and, uh, yeah, uh, if they have-

45:22

Aaron Francis

Uh

45:22

Kent C. Dodds

... any follow-up questions or anything?

45:24

Aaron Francis

X.com, the Everything app. And as we mentioned at the top of the show, Aaron D. Francis is my handle there. So you can find me on Twitter at Aaron D. Francis.

45:33

Kent C. Dodds

Fabulous. Thanks so much, Aaron D. Francis, and thanks everybody for watching. We'll see you in the next one.

45:38

Aaron Francis

See ya.

Get the next one by email

I’ll send you some more. Not too many.

Join 7,500+ other readers.

Read past issues