Latest / The Index Podcast / Clean Coders: Is Uncle Bob Right That Buying Software Is Dead?
Transcript
- 0:08SPEAKER_03: One quick word from our sponsor, Triton One.
- 0:10SPEAKER_03: If you're building on Solana, your app is only as fast and as reliable as the infrastructure underneath it.
- 0:17SPEAKER_03: At Triton, we built the open source read layer for Solana ourselves.
- 0:22SPEAKER_03: The streaming engine that pushes data the moment it hits the chain, the transaction pipeline that lands your trades when everyone else is timing out.
- 0:29SPEAKER_03: All of it running on bare metal in over 20 data centers worldwide, served by the team that wrote it.
- 0:34SPEAKER_03: We power some of the biggest names on Solana, but with Pay AsYo Go, anyone can start.
- 0:39SPEAKER_03: Sign up, get an endpoint, pay only for what you use, head to Triton.
- 0:43SPEAKER_03: Serious Infrastructure for Series Builders.
- 0:45SPEAKER_03: I've worked with Triton for years at this point.
- 0:48SPEAKER_03: They've been around since the very beginning of Solana.
- 0:50SPEAKER_03: They've contributed open source software that's the core of the infrastructure of Solana that makes everything work since day one.
- 0:56SPEAKER_03: I couldn't recommend working with a better team.
- 0:58SPEAKER_03: Triton.
- 1:05SPEAKER_03: Hey everybody, welcome to the index.
- 1:07SPEAKER_03: I'm your host, Alex Cahaya, and today I'm joined by a special guest, Robert Martin, aka Uncle Bob.
- 1:15SPEAKER_03: He's on Twitter.
- 1:16SPEAKER_03: You can find him there.
- 1:16SPEAKER_03: That's where we connected.
- 1:18SPEAKER_03: He's known to most of the industry as Uncle Bob, author of Clean Code and Clean Architecture, and creator of Swarmforge, which is a tool for coordinating disciplined swarms of AI coding agents.
- 1:30SPEAKER_03: Bob, thanks for being on the show.
- 1:31SPEAKER_00: Thanks for having me.
- 1:32SPEAKER_00: This should be fun.
- 1:34SPEAKER_03: And you wait, do you go by Bob?
- 1:35SPEAKER_03: I do.
- 1:36SPEAKER_00: Yeah.
- 1:36SPEAKER_03: Okay, Bob, got it.
- 1:37SPEAKER_03: Yeah.
- 1:38SPEAKER_03: My I also I have an Uncle Bob.
- 1:39SPEAKER_03: Hi, Uncle Bob, if you're listening.
- 1:41SPEAKER_03: I'm excited to have you here.
- 1:43SPEAKER_03: First thing I always ask people is like, what gets you out of bed in the morning?
- 1:45SPEAKER_03: You know, why are you doing why are you build building uh Swarm Forge?
- 1:49SPEAKER_00: Oh my goodness, what gets me out of bed in the morning?
- 1:51SPEAKER_00: Well, I just love writing programs.
- 1:53SPEAKER_00: I've been a software guy for like, I don't know, 50, some odd, maybe 60 years now.
- 1:58SPEAKER_00: And that's what gets me out of bed in the morning.
- 2:01SPEAKER_00: And in in particular, right now, this AA stuff is just as exciting as can be, right?
- 2:07SPEAKER_00: This this is the future, and I want to be riding that wave.
- 2:12SPEAKER_00: It's it's a ton of fun.
- 2:14SPEAKER_03: I couldn't agree more, right?
- 2:16SPEAKER_03: Like every day is just something new and it just is moving so fast.
- 2:20SPEAKER_03: And I imagine you see a lot of it from your audience on Twitter.
- 2:23SPEAKER_03: What are you most excited about?
- 2:25SPEAKER_03: What's what's like keeping you up at night?
- 2:27SPEAKER_03: You just can't sleep.
- 2:28SPEAKER_03: It's so exciting.
- 2:29SPEAKER_00: Just how fast things are changing and changing in a remarkably positive direction.
- 2:35SPEAKER_00: Right.
- 2:35SPEAKER_00: This this change it is greater than any change I've experienced in my career.
- 2:42SPEAKER_00: It's it is it is really remarkable.
- 2:44SPEAKER_00: And the the potential is I don't even know how to scale it.
- 2:48SPEAKER_00: It's enormous.
- 2:50SPEAKER_00: So so I've I think the next few years are going to be uh very interesting and a lot of fun.
- 2:58SPEAKER_03: Well, tell me about your career.
- 2:59SPEAKER_03: I mean, you mentioned this is the fastest change you've seen, and I I feel it, but I think you've had a longer career than me.
- 3:06SPEAKER_03: Like, what is your background?
- 3:07SPEAKER_00: Oh, I started programming when I was a kid.
- 3:10SPEAKER_00: My mom got me a little computer when I was 12.
- 3:12SPEAKER_00: It was a plastic thing.
- 3:14SPEAKER_00: Uh, and that that was it for me.
- 3:16SPEAKER_00: I just wanted to play with computers from then on.
- 3:19SPEAKER_00: See the little blinky box behind me, that's a mock-up of a PDP-8, one of the first computers I ever programmed in, all in assembly language, of course.
- 3:27SPEAKER_00: And ever since then, ever since like 1964, I have been a programmer, writing programs, making these computers, do backflips, learning everything I could about operating systems and machines and everything possible.
- 3:41SPEAKER_00: Eventually became a consultant about 30 years ago and traveled the world and yelled at people, which is a great gig, by the way.
- 3:49SPEAKER_00: Started writing books, and you know, people started reading those books, apparently.
- 3:54SPEAKER_00: Uh, and I thought, you know, a year ago, I thought, you know, I've come to the end.
- 4:00SPEAKER_00: Moore's law has kind of stopped the growth.
- 4:03SPEAKER_00: The computers aren't changing much, the languages aren't really doing that much.
- 4:07SPEAKER_00: It's I think we've reached the end.
- 4:09SPEAKER_00: And I was about to hang my hat up on a hook, and all this AI stuff started, and I thought, whoa, I just dove right into that again.
- 4:18SPEAKER_00: Yeah.
- 4:18SPEAKER_03: Grab the hat.
- 4:22SPEAKER_03: I mean, isn't it amazing though?
- 4:23SPEAKER_03: And it's happened in such a short time.
- 4:25SPEAKER_03: You know, like it was like less than I feel like it's not even been a year, right?
- 4:30SPEAKER_03: Like since everything really popped, started popping off.
- 4:33SPEAKER_03: You know, there were signs before leading up to this for the past two or three years, but this past 12 months has just been absolutely nuts.
- 4:40SPEAKER_03: And I feel like even for people like us who are like in the weeds with this stuff, it's still just like lightning fast, and 90% of the population is still kind of unaware of what's happening.
- 4:52SPEAKER_03: I I am really interested in in this swarmforge thing.
- 4:55SPEAKER_03: What is that?
- 4:56SPEAKER_00: Oh, uh Swarmforge was an experiment.
- 4:58SPEAKER_00: His was, I don't know how to classify that.
- 5:01SPEAKER_00: It's an experiment.
- 5:02SPEAKER_00: What how do you deal with these agents?
- 5:04SPEAKER_00: So if you start, if you're a programmer, you start with one little agent nowadays.
- 5:09SPEAKER_00: So you type some prompts in and build some software for you.
- 5:12SPEAKER_00: And and you type in another prompt and build some more software for you.
- 5:16SPEAKER_00: And after a while, you realize that you're spending more time waiting for the agent than actually writing the prompts.
- 5:23SPEAKER_00: So then you fire up another agent.
- 5:25SPEAKER_00: And now you got two agents, and you're juggling two agents, and that works out pretty well until you realize you're still waiting, and then you fire up a third agent, and at some point you lose control of that.
- 5:37SPEAKER_00: And the idea of Swarmforge was to put the agents into a harness that gives them enough discipline so that you can feed them requirements at the beginning, and code comes out the back end, and you don't have to wait.
- 5:51SPEAKER_00: That was that was kind of the idea.
- 5:53SPEAKER_00: That's why uh why I put together Swarmforge.
- 5:56SPEAKER_00: My son actually put it together initially, and then I grabbed it and I thought, well, this is a good idea, but I've got my own ideas, and so nowadays we do not use each other's software.
- 6:07SPEAKER_00: If you're into agents, we do not use each other's software.
- 6:10SPEAKER_00: We point our agents at someone else's software, and then we tell our agents to build something like that.
- 6:16SPEAKER_03: Yeah.
- 6:17SPEAKER_03: I'm curious what you think about that.
- 6:18SPEAKER_03: This is something we think we've we've been talking about a lot because my partner is this guy named Brian Fox.
- 6:23SPEAKER_03: He wrote Bash back in the day.
- 6:25SPEAKER_03: He co-authored the GPL licenses, um, huge open source guy.
- 6:28SPEAKER_03: And now all open source software is being consumed by agents and in a lot, a lot of times like rewritten.
- 6:34SPEAKER_03: Like so it's like there's the I see two problems.
- 6:37SPEAKER_03: One is that, and then the other one is um contributions to open source, right?
- 6:42SPEAKER_03: Like Swarmforge is open source.
- 6:44SPEAKER_03: I was just checking the repo out where you're talking, and it's like, how do you manage, you know, a thousand people wanting to put up PRs that are all AI slop?
- 6:51SPEAKER_03: You know, and I'm curious, you know, what you think about both of those issues.
- 6:55SPEAKER_00: Uh Swarmforge is not a product.
- 6:57SPEAKER_00: I'm not offering it as a product, it's just out there for people to look at.
- 7:00SPEAKER_00: I don't pay much attention to the PRs.
- 7:02SPEAKER_00: Um, people can contribute all they want, but I'm not going to integrate other people's stuff into it because there's no point in doing that anymore.
- 7:10SPEAKER_00: If you want something like Swarmforge, don't take Swarmforge, point your agent at Swarmforge and have your agents build something better for you.
- 7:20SPEAKER_00: Right?
- 7:20SPEAKER_00: The the uh the days of buying software products uh is waning, right?
- 7:27SPEAKER_00: I've been writing books for years.
- 7:28SPEAKER_00: I've been using a tool called Scrivener.
- 7:31SPEAKER_00: And just yesterday I thought, Scrivener's okay, but there's a lot of things I would do differently.
- 7:36SPEAKER_00: And then I just talked to an agent and I said, you know what?
- 7:39SPEAKER_00: Make me a word processor that's like Scrivener, but doesn't do all that stuff I don't like.
- 7:45SPEAKER_00: You know, and 20 minutes later, I've got a word processor on my screen.
- 7:50SPEAKER_00: Yeah, it needs a few tweaks, but I think this is I think this is the way that things are gonna go.
- 7:56SPEAKER_00: Why buy somebody else's software when you have the power at your fingertips to make something that's perfect for you?
- 8:05SPEAKER_03: I yeah, I totally agree.
- 8:06SPEAKER_03: I've been talking about this a lot with people, it's like the idea of adaptive software.
- 8:10SPEAKER_03: And you know, in every industry, there are solutions to problems that are specific to that industry, but none of these solutions are specific to that customer, right?
- 8:20SPEAKER_03: Like everybody has the same problem, but their view of to solving it has tweaks.
- 8:25SPEAKER_03: And there's just no reason anymore to give proprietary software to a customer that can't be like molded to their organization.
- 8:33SPEAKER_03: It's just not a thing anymore.
- 8:35SPEAKER_03: And I think this is where software factories are starting to take off from a from a product standpoint.
- 8:41SPEAKER_03: Um, but really what you're selling there is inference and some your view of the world of like how software gets built with these tools, like your orchestration of the agents.
- 8:51SPEAKER_03: That's that's my view.
- 8:52SPEAKER_03: I'm curious what you how you view that.
- 8:54SPEAKER_03: Like, if software is basically the old model is dead, what is the new model?
- 8:59SPEAKER_03: What's the way that people build businesses in the software space delivering services to people?
- 9:05SPEAKER_00: Boy, I wish I knew.
- 9:09SPEAKER_03: I mean, we're all asking ourselves that question.
- 9:11SPEAKER_03: If you've been in this business for any time, right, you're like, what do we do now?
- 9:14SPEAKER_00: Yeah.
- 9:15SPEAKER_00: I mean, uh, you know, a couple of years ago, you could build the killer app, right, and put it on uh someplace and make a gazillion dollars.
- 9:24SPEAKER_00: I don't think that works anymore because everybody can make the killer app, and everybody can tune the killer app to their own particular needs.
- 9:30SPEAKER_00: So why create the killer app at all?
- 9:35SPEAKER_00: I think it's changing quite rapidly.
- 9:37SPEAKER_00: It may be that software products don't exist anymore, and that we all just customize the software, customize the behavior of our machines to match our needs without really thinking about software anymore.
- 9:52SPEAKER_00: We're just going to customize the behavior of our machines to our particular needs, and yes, we'll need some specialists who can do that really well, and we might call them programmers and we might not.
- 10:02SPEAKER_00: Who knows what we're going to call them?
- 10:04SPEAKER_00: But okay, all of that I think is still going to happen.
- 10:08SPEAKER_00: But the idea of going out to uh uh an app store and downloading an app, I think is gone.
- 10:15SPEAKER_00: Instead, you might you might have templates.
- 10:18SPEAKER_00: You know, there might be people who create templates that you can point an agent at and say, okay, if you're gonna use our template, you have to pay us a little fee, but otherwise your agent can build something like that.
- 10:31SPEAKER_00: Could go that way.
- 10:33SPEAKER_03: I mean, that's kind of interesting.
- 10:34SPEAKER_03: I I do think, look, I don't know.
- 10:37SPEAKER_03: The the problem I've seen and I've experienced is that actually deploying these things is like a production ready product that my team can use.
- 10:44SPEAKER_03: Let's say I'm at like a 10,000-person company.
- 10:47SPEAKER_03: Like that part's still kind of hard getting it right, even with AI.
- 10:51SPEAKER_03: Sure.
- 10:51SPEAKER_03: But then I, you know, I'm saying that and I'm like, that problem's probably gonna go away soon.
- 10:55SPEAKER_03: Like the the models are gonna get good enough that that that also just works.
- 10:59SPEAKER_03: Like the the you know, I remember the first time I vibe coded something, and it was something like I used to run uh a software development shop.
- 11:06SPEAKER_03: I was CRO there.
- 11:07SPEAKER_03: And you know, we I vibe coded something in a couple hours that would have cost you know six figures two years ago.
- 11:13SPEAKER_03: Yeah, right.
- 11:13SPEAKER_03: Like, you know, it's like just like making basic features like would have cost you a decent amount of money to get like a couple engineers to ship in two months, you know, and now you can make that in like five minutes, but still deploying it to production where it's like all the different components, like wherever you're storing things are over here, wherever you're hosting the front end is over there, like how it all connects, like that part is still takes some effort and some human you know action.
- 11:39SPEAKER_03: Um, you got to set up the accounts, you gotta give the credentials to the agent, right?
- 11:42SPEAKER_03: Like, but at some point, like when does that bled away, you know, and and this all this infrastructure becomes agent ready, you know?
- 11:49SPEAKER_03: And I I don't know if we're there yet.
- 11:50SPEAKER_00: I think that goes away pretty quickly.
- 11:52SPEAKER_00: Most of that is security stuff.
- 11:53SPEAKER_00: You know, how many passwords do I give these agents?
- 11:56SPEAKER_00: And what happens if the agents use the passwords in the wrong way?
- 11:59SPEAKER_00: Right.
- 11:59SPEAKER_00: All of that stuff I think is going to we're going to find a solution to that.
- 12:03SPEAKER_00: We're going to make that easier because it's absolutely necessary to make it easier.
- 12:08SPEAKER_00: And the financial motivation for making that easier is enormous.
- 12:12SPEAKER_00: Right.
- 12:13SPEAKER_00: So eventually you will simply tell your agent, okay, deploy this.
- 12:17SPEAKER_00: And it'll work out everything, and it'll have all the necessary securities, and there'll be enough guardrails around it so that the passwords don't leak and stuff like that.
- 12:26SPEAKER_00: But yeah, I that will get solved, and I I I gotta believe that's within the next year.
- 12:32SPEAKER_03: Yeah.
- 12:33SPEAKER_03: I mean, it's funny, your your commentary here makes the prospectus that just came out for Anthropics IPO seem less insane, right?
- 12:42SPEAKER_03: Like, because I think they lost like$48 billion or something, but their growth rate is through the roof.
- 12:47SPEAKER_03: And when you think about it, like the inference that they're selling and their model and how their tools like make that model work in a useful way to just like build things.
- 12:57SPEAKER_03: Yeah, I mean, that's just taking over the world.
- 12:59SPEAKER_03: It it that it's taking over everything.
- 13:00SPEAKER_03: Pretty much like that's the only software business in the future, not distant future.
- 13:06SPEAKER_00: Yeah, not distant, not distant future.
- 13:08SPEAKER_00: You know, everybody wondered how Facebook was ever gonna be profitable.
- 13:12SPEAKER_00: Yeah.
- 13:15SPEAKER_00: But it didn't matter.
- 13:18SPEAKER_03: At a macro level, what do you think of like in in your experience, you know, is gonna happen?
- 13:23SPEAKER_03: I hear like this bull and bear case because you know, a bunch of people are like, ah, they're overcommitted.
- 13:28SPEAKER_03: Anthropic, less so, but like open AI, especially, you know, with the with their Ada Center commits.
- 13:33SPEAKER_03: And if there's any hiccup in the market, we're gonna have like a house of cards moment just falling down and a big correction.
- 13:39SPEAKER_03: And I'm wondering if you believe that or not.
- 13:41SPEAKER_03: Part of me does.
- 13:42SPEAKER_03: The other part of me, like I've been talking to like every inference provider that is selling compute, like marketplace, all sorts of different inference providers, and they're all at max capacity.
- 13:53SPEAKER_03: They literally like are like, yeah, we're sold out for like months.
- 13:56SPEAKER_03: Yeah.
- 13:56SPEAKER_03: And so that tells a different story.
- 13:58SPEAKER_03: Like, is that gonna dry up or is this just so powerful that it's just like it's up only for indefinite time period?
- 14:06SPEAKER_00: Saying things are up only is always a little bit risky.
- 14:09SPEAKER_00: It wouldn't surprise me if there was a correction coming at some point.
- 14:13SPEAKER_00: However, the demand is so enormous and the potential is so enormous that any correction is going to be overwhelmed shortly.
- 14:20SPEAKER_00: I mean, you know, Elon Musk is talking about launching data centers into orbit.
- 14:25SPEAKER_00: Yeah.
- 14:25SPEAKER_00: And and that'll be profitable.
- 14:30SPEAKER_00: Who knew?
- 14:31SPEAKER_00: Who knew that it would be computation that would drive us into space?
- 14:35SPEAKER_03: Isn't it wild?
- 14:36SPEAKER_03: Isn't it wild?
- 14:38SPEAKER_03: It's absolutely crazy.
- 14:40SPEAKER_02: Yes.
- 14:41SPEAKER_03: So, okay, so uh why did you so if you're you know you're not it's not a business, you're just like have it there hosted, it's Warmforge, right?
- 14:48SPEAKER_03: It's on GitHub for people to use.
- 14:50SPEAKER_03: What was the inspiration for it?
- 14:51SPEAKER_03: Why'd you build it?
- 14:52SPEAKER_00: Try to figure out how to use these doggone things.
- 14:55SPEAKER_00: Yeah, that's my that's my whole thing.
- 14:57SPEAKER_00: I was an experiment for you.
- 14:58SPEAKER_00: How do we use this stuff?
- 14:59SPEAKER_00: And I like to put things up because I'm an author, you know, I like and people follow me.
- 15:03SPEAKER_00: So I put my ideas out where people can see them.
- 15:06SPEAKER_00: And some of the ideas are really bad and some of the ideas are really good, and I kind of let the chips fall where they may.
- 15:12SPEAKER_00: But yeah, my I am motivated to get this information out and to and to research it as much as I can.
- 15:18SPEAKER_00: And I'm having an absolute blast at it.
- 15:20SPEAKER_00: It's just I didn't think I was gonna be able to do it.
- 15:23SPEAKER_00: It is fun.
- 15:24SPEAKER_03: Last year it is fun.
- 15:25SPEAKER_03: I mean, for for people who like who I I'm an idea guy, right?
- 15:29SPEAKER_03: And now uh it's dangerous because I'm just technical enough to use these things, and now all my ideas are like right here.
- 15:35SPEAKER_03: Yeah, right in my fingertips, you know, it's like type prompts, and then they're the thing I wanted is there.
- 15:41SPEAKER_03: Uh and look, this is cool.
- 15:42SPEAKER_03: This experiment, I would say it's pretty like objectively pretty successful.
- 15:45SPEAKER_03: Like 3.9k stars, 3.9, 3,900 stars, 393 forks, 62 people watching it.
- 15:52SPEAKER_03: I don't know what that means.
- 15:53SPEAKER_03: They're just like watching for new, new, new things that you put into it, but that's like a pretty pretty decent amount of traction.
- 15:58SPEAKER_00: Well, I mean that I think that I think that's a reflection of interest.
- 16:03SPEAKER_00: There's a lot of people who are wondering how the heck do we get our hands around these doggone agents?
- 16:08SPEAKER_00: You know, and and I don't know if that's success or not.
- 16:10SPEAKER_00: You know, my view on the Swarmforge right now is that uh it's kind of a failed experiment.
- 16:15SPEAKER_00: But I don't know.
- 16:16SPEAKER_00: Experiments don't actually fail, right?
- 16:19SPEAKER_00: Experiments produce data, but the data that this experiment has produced for me has not been encouraging.
- 16:26SPEAKER_00: And and the reason tell me about that.
- 16:28SPEAKER_03: Okay.
- 16:28SPEAKER_03: Yeah, what does that mean?
- 16:29SPEAKER_03: Why are you not encouraged?
- 16:30SPEAKER_00: Well, so when I started this project with my son, right, the the models were the models that were around in February, right?
- 16:40SPEAKER_00: And they were okay.
- 16:41SPEAKER_00: They did good work.
- 16:42SPEAKER_00: I was very impressed with them at the time, but but the work they did was substandard.
- 16:48SPEAKER_00: Now, you know, you know, yeah, from my viewpoint back then, I was like, whoa, these guys are really cool, but but they need a little cleanup.
- 16:55SPEAKER_00: These guys need a little discipline.
- 16:56SPEAKER_00: These agents, they go off into crazy lala land, and we have to put them in corrals, and we've got to organize them so that they can produce things predictably.
- 17:07SPEAKER_00: And I worked for a good while on that.
- 17:09SPEAKER_00: Swarmforge was one of the outcomes of that.
- 17:12SPEAKER_00: And you look at the models today, and they don't need that kind of constraint, at least for the stuff I'm doing.
- 17:20SPEAKER_00: They can handle most of that stuff by themselves.
- 17:23SPEAKER_00: And I have kind of backed away from Swarmforge and thought, well, okay, maybe the harness, maybe this harnessing idea is the wrong idea.
- 17:34SPEAKER_00: Maybe there's a different way to organize these things, a way that is less bureaucratic, less uh legalistic.
- 17:42SPEAKER_00: You know, I had them crank down pretty hard.
- 17:44SPEAKER_00: I wrote all kinds of scripts and tools so that they could not escape from their own little job.
- 17:51SPEAKER_00: Every one of these agents had its job, and it only did that job.
- 17:55SPEAKER_00: And when they talked to each other, they couldn't talk to each other about the job because that would leak too much information.
- 18:01SPEAKER_00: So I, you know, I had them, I had a really locked down type.
- 18:05SPEAKER_00: And then in the end, I'm, I'm, it worked.
- 18:08SPEAKER_00: It was great.
- 18:09SPEAKER_00: I mean, I could type in prompts at one end and code would come out the other end, but it would take an hour or two to get anything done.
- 18:17SPEAKER_00: And then I would go to a single agent with a modern, a modern model and type in a little prompt and it would pop something out in five minutes, and the result was better.
- 18:27SPEAKER_00: You know, better code, better code, better organized, better tested.
- 18:31SPEAKER_00: Uh, and looking at that and going, well, I don't think that agent thing, I don't think that harness thing worked out real well.
- 18:40SPEAKER_03: Yeah, well, and it's it's hard to compete because if you're just using, if you're using clawed code with Fable, right?
- 18:46SPEAKER_03: Like, it just works.
- 18:47SPEAKER_03: I mean, and I so I was hunting for 10 days in the beginning of September.
- 18:51SPEAKER_03: I went elk hunting with my brother.
- 18:53SPEAKER_03: And before I left for fun, I asked, I used ChatGPT to help me plan the prompts, like just the the you know, through through Chrome.
- 19:02SPEAKER_03: And I was like, look, I'm going hunting for elk.
- 19:04SPEAKER_03: Here's the area we're hunting in.
- 19:06SPEAKER_03: Help me plan.
- 19:07SPEAKER_03: Like I'm a fur, it's my first time going.
- 19:08SPEAKER_03: I want to figure out like where are the elk and give me a topo map that's integrated interactive that has like all the different zones I should hunt.
- 19:15SPEAKER_03: And oh, by the way, there's a fire there.
- 19:17SPEAKER_03: So I want you to research the fire and make sure that I'm like in a safe space and like does that impact my strategy?
- 19:23SPEAKER_03: So ChatGPT, just the you know, the desk, not the desk, what is it called?
- 19:26SPEAKER_03: Codex, but just ChatGPT, like gave me a plan for how to build this application.
- 19:30SPEAKER_03: Then I did an A-B test and I gave it to Claude Code using Fable and Codex using whatever their top model is, and just like let them run in a loop and just burn tokens, right?
- 19:41SPEAKER_03: Just to see what would happen.
- 19:42SPEAKER_03: And it was really interesting.
- 19:43SPEAKER_03: Uh, and I have a video up on the on uh my my company's website about this where I did the A-B test.
- 19:48SPEAKER_03: Fabled smoked codecs, like absolutely smoked codex.
- 19:53SPEAKER_03: And I and I didn't really use many, like we I do have some like agents that I've loaded into both environments that we built that were like a little bit like Swarmforge, like there's like a critique agent, there's like a planning agent, there's you know, there's like a gray beard agent that's like the architect, but I didn't have it wasn't my own harness, right?
- 20:09SPEAKER_03: It was just these like specialized jobs.
- 20:11SPEAKER_03: Yeah and they just ran in a loop and then produced the result.
- 20:15SPEAKER_03: And Codex's result like looked okay, but it was not nearly as slick as Fables, and it didn't work.
- 20:21SPEAKER_03: Like it was full of bugs.
- 20:23SPEAKER_03: And it also burned all but like 1% of my tokens.
- 20:26SPEAKER_03: Like I was almost completely out, you know, for the month.
- 20:28SPEAKER_03: Whereas Codex, like it took like an hour, it just spun for an hour.
- 20:32SPEAKER_03: I mean, not Codex, um, Claude with Fable spun for like an hour, hardly used any of my limits, and was like actually usable.
- 20:39SPEAKER_03: Like I used it on the hunt and it worked.
- 20:41SPEAKER_03: Like we got four opportunities this hunt at Elk, you know, like massive monsters in some of the areas that it predicted, which is pretty cool.
- 20:49SPEAKER_03: So, anyway, I just bring that up because it was a fun experiment, like you.
- 20:52SPEAKER_03: Like I was just kind of seeing what it would do, and uh just out of the box, these things have gotten so much better so fast.
- 20:59SPEAKER_03: And so again, I bring it back to you.
- 21:00SPEAKER_03: The question is like, okay, what's next?
- 21:02SPEAKER_03: If if you think we need to be less strict about, you know, writing our own harness, for example, like what is worth building?
- 21:08SPEAKER_03: What are you what are you working on?
- 21:09SPEAKER_00: What I've been doing lately is saying, okay, maybe maybe the models are fine.
- 21:14SPEAKER_00: Maybe they don't need to be over-disciplined.
- 21:17SPEAKER_00: Maybe it's just fine to type into a model and have it work for 40 minutes or an hour and produce a big watt of code.
- 21:25SPEAKER_00: Now I want to see what the structure of that code is.
- 21:28SPEAKER_00: Um, you know, I'm a software engineer, so I want to look in there and I want to make sure that it's followed all the right rules and the dependencies are going the right direction.
- 21:37SPEAKER_00: And if I make a change to this part, it's not going to break that part.
- 21:40SPEAKER_00: You know, I still want to follow all those rules.
- 21:43SPEAKER_00: So the the kinds of things that I've been working on lately are visualization tools.
- 21:48SPEAKER_00: The worst thing I think you can do with the code that comes out of an agent is read it.
- 21:56SPEAKER_00: Because if you read it, you've slowed the entire thing down to human scope.
- 22:00SPEAKER_00: And that's the thing we have to avoid.
- 22:02SPEAKER_00: We don't want to slow down to human speed.
- 22:04SPEAKER_00: Humans are slow with code.
- 22:07SPEAKER_00: So I want to have some picture that I can see of the structure of the code so I can look at it and go, oh, I see what you did there.
- 22:15SPEAKER_00: Okay, but yeah, you got to move this part over there and you got to cut that one and move it over there.
- 22:20SPEAKER_00: And you didn't test this part enough.
- 22:22SPEAKER_00: I want to be able to have that kind of interaction with the agent at a very high level, a very kind of architecture software design level, without seeing a stitch of code, without caring what the language is, I want to see what the overall structure and architecture is so that I can tune it.
- 22:41SPEAKER_00: So far, the models are not good at that.
- 22:45SPEAKER_00: Or if I should say they're getting better, but they're not as good as I am at that.
- 22:51SPEAKER_00: So I want to impose my will over the structure of the code.
- 22:55SPEAKER_03: That's interesting.
- 22:56SPEAKER_03: I mean, well, that that's judgment, right?
- 22:57SPEAKER_03: That's human, that's human judgment that still is kind of missing, is what you're talking about.
- 23:02SPEAKER_03: And it's decades of architectural knowledge, right?
- 23:05SPEAKER_03: And that's something I wouldn't have.
- 23:07SPEAKER_03: Like I don't, I don't know how to do that personally.
- 23:09SPEAKER_03: I mean, I think there's a lot of industries where that's true still.
- 23:14SPEAKER_03: And there's some industries where the AI really cannot get that information, right?
- 23:19SPEAKER_03: It's not like they can just scrape that off of Google, right?
- 23:22SPEAKER_03: 30 years of experience in the automotive space, you might not be able to figure out some of those kinds of judgments.
- 23:28SPEAKER_03: What does that mean for us?
- 23:29SPEAKER_03: Like, is that does that eventually get sort of absorbed into the models as well?
- 23:34SPEAKER_03: Or probably.
- 23:35SPEAKER_03: Does that give us permanent value as humans?
- 23:37SPEAKER_03: Or are we just like what are we doing?
- 23:41SPEAKER_03: Yeah.
- 23:42SPEAKER_00: So let's start with a fundamental premise premise.
- 23:45SPEAKER_00: Fundamental premise.
- 23:46SPEAKER_00: All technology exists to serve humans.
- 23:49SPEAKER_00: Period.
- 23:51SPEAKER_00: So do humans have value?
- 23:53SPEAKER_00: Yes, because we are the target of the technology.
- 23:55SPEAKER_00: We are the ones who get served by the technology.
- 23:58SPEAKER_00: It is our needs and our desires that the technology exists to answer and fulfill.
- 24:04SPEAKER_00: So, yes, humans always have value.
- 24:06SPEAKER_00: Do human programmers always have value?
- 24:10SPEAKER_00: Will we always be programming computers?
- 24:13SPEAKER_00: And the answer I have to that is yes, of course we will, because the agents, the models, are not human.
- 24:20SPEAKER_00: They don't know what it means to be human.
- 24:22SPEAKER_00: They don't have human instincts or human desires.
- 24:26SPEAKER_00: They don't wake up in the morning and wonder what the hell am I going to do today?
- 24:30SPEAKER_00: Right?
- 24:30SPEAKER_00: They don't have that.
- 24:32SPEAKER_00: So someone, a human, has to act as the liaison between the models and the users.
- 24:41SPEAKER_00: Now that may not be writing code anymore.
- 24:44SPEAKER_00: That might be something else, some kind of architecture of the feature set and our architecture of the user interface and things of that nature.
- 24:55SPEAKER_00: But there will always be a human sitting between the machine and the users.
- 25:01SPEAKER_00: Because somebody's got to make that translation.
- 25:04SPEAKER_00: And that's what programmers have always done.
- 25:07SPEAKER_00: It's just that we did it in tiny little details before.
- 25:10SPEAKER_00: And now we're going to be doing it in uh much less detail.
- 25:14SPEAKER_00: Or dealing with far fewer fiddly little details.
- 25:18SPEAKER_03: So it brings me back to one of my earlier questions.
- 25:20SPEAKER_03: I want to ask it again.
- 25:21SPEAKER_03: You know, you were like, I was like, okay, so then what are we doing as entrepreneurs building software?
- 25:27SPEAKER_03: You know, like what does that mean for businesses, for like a business model now going forward?
- 25:33SPEAKER_03: Do you have any ideas?
- 25:36SPEAKER_00: I haven't got a clue.
- 25:37SPEAKER_03: Any wisdom for me?
- 25:40SPEAKER_00: I don't know why anyone would build software for someone else.
- 25:45SPEAKER_00: Because it's too easy to do.
- 25:48SPEAKER_00: For me to build software for myself.
- 25:50SPEAKER_00: It's too easy for you to build software for yourself.
- 25:53SPEAKER_00: There's there's very likely going to be a market for the liaison, the person who is skilled with agents who can translate it to the to the humans.
- 26:03SPEAKER_00: That I think there will always be a market for.
- 26:06SPEAKER_00: But is there a market for a software product?
- 26:11SPEAKER_00: Uh I don't know.
- 26:13SPEAKER_03: Yeah.
- 26:13SPEAKER_00: I don't know.
- 26:14SPEAKER_00: I don't know.
- 26:15SPEAKER_03: I mean, so basically, you end up with a bunch of services companies that are way more scalable because they can just service, you know, one guy can service 100 customers, you know, and each customer is paying them six figures to seven figures at enterprise, right?
- 26:27SPEAKER_03: Like they're big.
- 26:28SPEAKER_03: And and it and and basically what happens there is you're not making like one software product for one problem.
- 26:35SPEAKER_03: You're taking their entire software budget, the whole thing.
- 26:39SPEAKER_03: Like if they use 12 different things to run their company, all 12 are now built by this one guy, right?
- 26:48SPEAKER_03: Yeah.
- 26:49SPEAKER_03: Pretty much.
- 26:50SPEAKER_03: Yeah.
- 26:51unknown: Yeah.
- 26:51SPEAKER_03: So I always ask this, we're getting towards the top of the show.
- 26:55SPEAKER_03: What have I not asked you that I should have asked or that you wanted to talk about?
- 27:00SPEAKER_00: Well, let's see.
- 27:02SPEAKER_00: Programmers.
- 27:03SPEAKER_00: I I don't know who your audience is.
- 27:04SPEAKER_00: Is your audience programmers?
- 27:07SPEAKER_00: It's a mix.
- 27:07SPEAKER_03: It's entrepreneurs, it's builders.
- 27:09SPEAKER_03: There's some VCs, like some you know, venture capitalists that watch the show because I get a lot of like startups and stuff on the show.
- 27:15SPEAKER_03: It's kind of a mix.
- 27:16SPEAKER_03: So programmers.
- 27:17SPEAKER_00: Programmers.
- 27:18SPEAKER_00: If you're talking about programmers, you're gonna probably be hiring programmers.
- 27:20SPEAKER_00: If you're a business owner, you've got to be hiring programmers.
- 27:23SPEAKER_00: If you're a programmer, you're wondering where what the hell is gonna happen to me in the next few years.
- 27:27SPEAKER_03: Yeah, what do I do?
- 27:28SPEAKER_00: Okay, so my my view on this is fairly straightforward.
- 27:33SPEAKER_00: Programmers who get good at dealing with agents will be in demand.
- 27:38SPEAKER_00: And the demand will not decrease.
- 27:40SPEAKER_00: Everybody's worried, you know, if wow, if one guy can do what 10 programmers used to do, then 90% of the programmers are gonna be out of work.
- 27:47SPEAKER_00: No, the businesses are gonna be doing 10 times as much stuff.
- 27:51SPEAKER_00: That's the way it always works.
- 27:53SPEAKER_00: So the demand for programmers will remain extremely high going forward.
- 27:58SPEAKER_00: There's a little dip right now because nobody knows what the hell's going on.
- 28:02SPEAKER_00: But you know, a year from now, I think the demand for programmers is gonna be through the roof.
- 28:06SPEAKER_00: If they know the AI stuff well.
- 28:08SPEAKER_00: If they've if they've been playing with this skunk and they've got some feel for it in their fingers, then I think they're fine.
- 28:15SPEAKER_00: There's a whole bunch of new jobs that are starting.
- 28:17SPEAKER_00: My daughter just got hired by a big company, uh, and and her title is Agentic Architect.
- 28:26SPEAKER_00: Agentic architect.
- 28:28SPEAKER_00: What a great title.
- 28:29SPEAKER_00: Have no idea what that means.
- 28:30SPEAKER_00: I hope she does.
- 28:31SPEAKER_03: She'll figure it out.
- 28:32SPEAKER_03: That's the point.
- 28:32SPEAKER_03: I think that's I think that's the point.
- 28:34SPEAKER_03: She's gonna make it mean something.
- 28:36SPEAKER_00: But but clearly the business is looking out there, going big, we need people who know how to harness this stuff.
- 28:42SPEAKER_00: We've got to be able to take advantage of this stuff because everybody's going to be going ten times faster.
- 28:47SPEAKER_00: And if we're going eight times faster, we're gonna be behind.
- 28:51SPEAKER_03: Well, if if your daughter hears this episode, I hope she knows how lucky she is to be in the defining position, right?
- 28:57SPEAKER_03: Like she gets to make it up as she goes, whatever this thing means.
- 29:01SPEAKER_03: She'll do great.
- 29:02SPEAKER_00: No, she's been at this for as long as just about anybody else has right now.
- 29:05SPEAKER_00: And right, and that's you know, not a lot of time, but okay.
- 29:09SPEAKER_00: None of us have been at this point.
- 29:11SPEAKER_03: Did she study computer science or anything like that?
- 29:12SPEAKER_03: I mean, she must have been all around it with you uh growing up, but does she have a CS degree or anything like that?
- 29:17SPEAKER_03: I'm just curious.
- 29:18SPEAKER_00: She has a chemical engineering degree, right?
- 29:20SPEAKER_00: And she she, like so many other people, she transitioned into software years back uh and climbed the rungs and became a very accomplished software engineer and started playing with the uh AI agents and stuff as early as possible because this family is really focused on that kind of stuff.
- 29:40SPEAKER_00: And now she's got this job.
- 29:42SPEAKER_00: And you know, my other son is the CTO of a company, and they're all they're doing nothing but agent work, you know, trying to build some system with agents.
- 29:51SPEAKER_00: And my other son is doing doing lots of agent work.
- 29:54SPEAKER_00: They're all at this, they're all at this, they're all trying is running as fast as they can just to stay up.
- 30:03SPEAKER_03: Yeah, that's what you have to do right now.
- 30:05SPEAKER_03: Yeah, absolutely.
- 30:07SPEAKER_03: That's what everyone has to do.
- 30:08SPEAKER_03: It's got to drink from a fire hose every single day.
- 30:10SPEAKER_00: So if you're a programmer, I don't think there's a problem.
- 30:12SPEAKER_00: Right now, what about junior programmers?
- 30:14SPEAKER_00: How do you get them into the mix?
- 30:17SPEAKER_00: Because what one of the fears is that you don't have to teach them to code anymore.
- 30:21SPEAKER_00: You're just gonna hand them an agent.
- 30:23SPEAKER_00: And my solution to that is no.
- 30:27SPEAKER_00: You're gonna keep the CS programs just like they are.
- 30:30SPEAKER_00: They're gonna be learning how to code in school.
- 30:32SPEAKER_00: When you hire a new programmer, you don't let them touch an agent for a couple of months.
- 30:38SPEAKER_00: You treat them like an agent.
- 30:40SPEAKER_00: You have the senior people prompt them and give them exactly the same constraints that they would they would give to an agent.
- 30:48SPEAKER_00: Let them live in the corral for a little while and then let them out with supervision so they can control these agents.
- 30:55SPEAKER_00: Because one of the problems here is that agents are a power tool, right?
- 31:01SPEAKER_00: Uh an extreme power tool.
- 31:03SPEAKER_00: And if if you're a novice with a power tool, you're at risk of losing fingers.
- 31:07SPEAKER_00: So it'd be a really good idea to gradually introduce people from plain programming and writing a little bit of code to finally harnessing an agent and getting the agent to do decent work.
- 31:19SPEAKER_03: I totally agree with you.
- 31:20SPEAKER_03: And I think there's another really important point that a part of what you're talking about, which is, and I so I used to be an educator.
- 31:26SPEAKER_03: I've taught everything, kindergarten all the way up to grad school.
- 31:29SPEAKER_03: And yeah, everything.
- 31:30SPEAKER_03: And I already saw this when I used to teach high school, middle school with my, especially like my freshmen and sophomores.
- 31:36SPEAKER_03: They would go from being these like bright, ready-to-learn, eager kids to like ninth and tenth grade, they cannot do anything, like couldn't create a Gmail account, you know.
- 31:46SPEAKER_03: And uh, this was 15, 20 years ago, okay, for me.
- 31:52SPEAKER_03: And with AI, they are losing their it's gonna get worse, right?
- 31:57SPEAKER_03: They're gonna lose the ability to problem solve on their own and to learn anything.
- 32:00SPEAKER_03: And I think that's why you still need to learn how to code if you're going to be doing anything in computer science, is because you need to learn how to problem solve and think for yourself.
- 32:08SPEAKER_03: You know, these power tools need to be used with caution, not just because they can take your fingers off, but they can make your brain just rot because you don't think for yourself.
- 32:16SPEAKER_03: And I think that that's it, that's the thing that that I lose sleep over with my kids, you know, is that I want them to grow up ready to figure things out on their own.
- 32:25SPEAKER_03: Um, and then use these tools to help them figure things out.
- 32:29SPEAKER_03: Um, but not to, because I think what I see happening, and I'm guilty of it too, using AI for myself.
- 32:34SPEAKER_03: One day you wake up and you're like, who's driving here?
- 32:38SPEAKER_03: Right?
- 32:38SPEAKER_03: Like, am I even in charge?
- 32:40SPEAKER_03: Or is this ChatGPT thing like doing all my writing all my emails and communicating for me?
- 32:44SPEAKER_03: Is this really how I think?
- 32:46SPEAKER_03: You know, I mean, I feel like we, we, we, we're the matrix is becoming more and more real.
- 32:50SPEAKER_00: The matrix was the inversion, right?
- 32:52SPEAKER_00: The the people were the batteries that served the technology, right?
- 32:56SPEAKER_00: Yes.
- 32:57SPEAKER_00: We that's not us.
- 32:58SPEAKER_00: That's not where we're this technology will serve us.
- 33:03SPEAKER_03: I agree with that.
- 33:04SPEAKER_03: Well, look, I I really appreciate taking the time to come on the show.
- 33:06SPEAKER_03: Thanks so much for being here.
- 33:07SPEAKER_03: It was a super fun conversation.
- 33:09SPEAKER_03: My pleasure.