Paul Ford (2:26)

What do you want to know?

Peter Kafka (2:27)

Earlier this year you wrote a piece of the New York Times. Great headline. The AI Disruption We've been Waiting for has Arrived. That is a classic curiosity gap title.

Paul Ford (2:35)

Well, I was one of like five AB tested headlines.

Peter Kafka (2:38)

You got the right one. Okay, so what is the AI disruption? How has it arrived? What are we talking about?

Paul Ford (2:44)

There's a lot of AI disruptions. Right. But the one that I'm focused on, because I am a technologist and I run technology consulting companies with partners. I'm very worried and concerned and curious about what it's doing to coding and delivering software for people who don't know a lot of software. Most of the software in the world falls under this big bucket of custom and enterprise, which is if you ever wonder what SAP and Salesforce are, they're kind of the base of that. And you go and you build tools for companies that help them manage sales

Peter Kafka (3:16)

or do that as opposed to buying Ms. Word off the shelf.

Paul Ford (3:18)

That's right. So the opposite of off the shelf. Very custom, very bespoke. Usually you're starting with something and building up from it. That's been my whole career when I'm not a writer. And so watching a couple years ago, I watched everybody was like, ah, what's going on with AI? And it just kind of talks to you. But I watched it do really boring code things, things I won't even explain, but just like stuff that would translate 50 year old code into something much more modern. And that's the stuff that is like the absolute trillion dollar core of everything. That's like why you call Accenture, why you get giant consulting firms to advise you on how to transform your organization

Peter Kafka (3:54)

incredibly Bo in many cases very necessary.

Paul Ford (3:57)

Well, it's, it's, you know, every order for Macy's or every shipment that comes in on a container ship is tracked somehow, right? And there's taxes and there's, it's just all the governance and compliance. Every word that makes people fall asleep is software. And I saw a couple of years ago that, that suddenly the robot could do the thing. It could just kind of do a pretty credible job, get you started. And most of these projects, a lot of them fail, they just kind of, they're really hard. They take a long time. And I was like, wow, this is going to speed it up. But which means that things are probably going to change all over. And I think they are. They truly are.

Peter Kafka (4:33)

Tie that to vibe coding or maybe or should we tie that to vibe coding and sort of what kicked in late last year?

Paul Ford (4:39)

Okay, so late last year. Well, late last year was this had been kind of building up for a while. So OpenAI let you do a lot of coding. Anthropic let you do coding. Google's Gemini lets you do coding. What happens is you sit down and just as you would prompt it to answer a question, you know, make me a Bullet point and tell me where to go when I go to Tuscany, you know, that kind of thing. You could be like write me a script that converts from a legacy database format and it would do an okay job. But usually it was pretty buggy and you know, it was cool that it would kind of get it moving and it often could save you time. But it was more like that was sort of. When you hear about Copilot and stuff like that, that's what they did. They were assistants, they were like a little buddy. And it was a little faster than searching stack.

Peter Kafka (5:19)

You can do some work but you can't read. I always think of it like in non technical terms, like an internal.

Paul Ford (5:24)

That's right.

Peter Kafka (5:24)

Can run out and do a bunch of stuff. You wouldn't go out and take it to market. You'd want to fact check it or double check it or whatever.

Paul Ford (5:30)

You truly shouldn't. It's pretty dangerous. It would kind of hallucinate code just like it would hallucinate ideas. Right. But it was getting steadily better. And so you saw this kind of curve. And then what happened is around November of last year, Anthropic, the makers of Claude, nothing radically changed around the way that LLMs work underneath the large language models, underneath AI. But they built kind of a real nerdy product called Claude code. They'd been building it and they just did a few things to it that were sort of software things like it's going to think a little longer, it's going to go into more loops, it's going to do this, it's going to do that. But as a result of all those things combined with a slightly smarter model, it started to just write relatively good code in large quantities. And so it could build you a whole website. It could.

Peter Kafka (6:18)

So that's not just going from an intern to a first year associate. That's several levels up.

Paul Ford (6:23)

That's right. It still is hard to manage, you still have to know a lot, et cetera. But for those of us who been at it for a while, it was a very kind of shocking moment because it was partially because you could see it. We got there incrementally and then to see it actually kind of just come into the light was overwhelming. And I found myself. That's what the Times piece is about. I found myself just kind of compulsively trying to figure this out. I hadn't been coding a lot, I'd been mostly managing. And I was like, I gotta go understand this because I've never had a moment like this in my career. Thought I'd seen everything. And so I just was like. I just constantly was prompting, building, prompting and building. What happened is things I'd been putting off for a decade, little personal projects. That's where I started. They just yielded in, like, a weekend. And that was after 10 years. And then it started doing really, really boring, difficult things for me. Things that I don't understand very well, like chip programming, fPGA programming, which is. Don't even worry what it means, but just, like, know that it's one of the nerdiest things you can do. And. And I don't really understand how it does it, because I don't really understand it, but it just is like, yeah, you want to do that? We got it.

Peter Kafka (7:29)

So you're describing noodling on sort of side projects, right? The equivalent of like, someone building a table at home because it sounds fun. Yeah, they're into that. But obviously this has implications for the actual accentures of.

Paul Ford (7:40)

It's more like, oh, man, we got to rent that kitchen. And then you're like, you know what? Hold on a minute. And then you do the Sorcerer's Apprentice, and it's suddenly clean and there's a new stove.

Peter Kafka (7:49)

I guess what I was getting at is when this light bulb clicks on, is it, oh, this is going to allow me, Paul Ford, to do cool shit that I wanted to do but never got around to.

Paul Ford (7:59)

Versus.

Peter Kafka (8:00)

Or is it? And. Or this is going to really reshape the way professional software is made.

Paul Ford (8:06)

Well, you don't want to bring the fleet of Goblins in to destroy everything at work, Right? And everybody was kind of building, and we're trying to deal with this technology at work. So this was just. It felt alien, and it also felt like you had to prove it out. And everybody was kind of going through that experience. That's why it was such a sea change. Like, everybody was like, hold on a minute. If this is what it looks like, this will be really different in the future. And that's a thing to say, but it's another thing to kind of fully internalize. And a lot of people are still. I mean, I talk to people all day about this stuff. It is very, very hard for organizations to actually adapt to this level of change. I'll give you an example. Like, I'll do a proposal. I'll throw something over the wall, and I'll be like. Because what will happen is, instead of writing a set of bullet points, I'll just go ahead and build the thing they're talking about now. It'll look like real software. But I would never tr. It's just kind of whipped together. But it's still something you can click and work. It has a database, all the nice stuff. And then they'll go silent for like three weeks. And it was very confusing at first because I'm like, did I do something wrong? But actually it just turns out that humans can't metabolize this rate of change. And so they would just treat it like this alien object. And then later they'd be like, hey, that's really weird that you did that. We should talk.

Peter Kafka (9:13)

So is the. Again, just want to dumb it down for me, not for my audience. Is this. This is going to take time out of coding. This is going to take cost. I guess this is the same thing. Is this going to allow serious coders to do things they couldn't do before?

Paul Ford (9:31)

Well, here's the thing. It's so disruptive that everybody is absolutely sure they have the answer to that question and there's no clarity. Right.

Peter Kafka (9:37)

I mean, we've seen Wall street, right, go and slash the prices of all these enterprise software companies.

Paul Ford (9:42)

Well, you know, Claude says it's anthropic, says it's going to do legal and Harvey. Okay, so that's AI versus AI. Harvey goes down in its st. I don't know if it's public or anyway, regardless private.

Peter Kafka (9:53)

But yeah, yeah.

Paul Ford (9:54)

So suddenly the legal category loses value. Claude comes out with Claude Design and suddenly Figma and Adobe stock drops, but the market's dumb. I also have a feeling that there's a lot of blockchain money where they just read the paper automatically with AI and just trade.

Peter Kafka (10:12)

Yeah. And this is the same market that told us Peloton was a gazillion dollar stock five years ago.

Paul Ford (10:17)

Right. So let's go back to the original question, which is it's very confusing because up until now no one has ever been able to get enough software. Engineering was very expensive. It took a lot of time. You had to buy stuff off the shelf. And everybody has the experience of using tools that aren't custom to them but they have to use because the customizing is really, really expensive. So what we don't know is will a new world in which it's fast to customize, in which it's easy to make something that's really bespoke just for you and where engineering might be more of your service or kind of here to help you along with product and solution people, as opposed to this sort of alien entity bolted onto the org that does its own thing. How does that all fit into the future? Where this plays out is, do we need junior engineers anymore because a senior can do a whole lot more work. Do we need senior engineers because now a management consultant or a product manager can code all day and produce code all day. Are we all just QA for this thing? It's very tricky as to whether this is like a. Was it Jeevan's paradox thing where we're going to have more code and more resources and we can just have as

Peter Kafka (11:28)

the more, the more we can create, the more will be used.

Paul Ford (11:30)

Is that real or is this going to eat the market? Figma drops and Adobe drops, but nobody gave up Photoshop. Everybody's like, all right, well, I guess Claude has one too now, right? And it's actually a pretty big lift for these new orgs to, to get away from that lock in. So I hate to be the person who comes on the show and is like, boy, I don't know, man, but nobody. Thank you.

Peter Kafka (11:53)

That's a reasonable position.

Paul Ford (11:55)

You really shouldn't. I wouldn't. People are like, they will ask, can I lay off my whole engineering team? The answer is no or not yet. Well, I don't know about not yet because we just don't know.

Peter Kafka (12:05)

But I mean, one of the reasons that OpenAI is worth 852 billion and anthropic is now suddenly worth about that as well is not because part of the premise is, and people always say that's going to create new stuff and new drug discovery, but it's also, this is going to take away a bunch of cost is the explicit, implicit promise of these companies.

Paul Ford (12:25)

I think that's right. I do. You know, obviously the market likes a good zero sum narrative and it helps things make sense. I just think we're in for some weirdness and I think there's going to be winners and losers. There already are, right? There's already people who get laid off because AI is coming. Like block is a good example. So you're working at, you're working for Jack Dorsey and one day it's over. And one of the reasons given is AI. And so is it really, I don't know, because they haven't done it yet. But certainly if you got laid off, you feel that you lost your job because everybody got too excited about AI.

Peter Kafka (12:56)

Or you might be really angry at Jack Dorsey for other reasons.

Paul Ford (12:59)

But both things can be true. You really can have both.

Peter Kafka (13:02)

I've heard a bunch of folks like you who are actual technical people raving about this, saying the equivalent of what you were talking about. I did this thing at home over the weekend because it was. I always wanted to do it.

Paul Ford (13:15)

It's not just that it was hard, it's that it used to cost hundreds of thousands of dollars.

Peter Kafka (13:19)

And they would say, or I would need a team of 10 people and I could do, you know, I was a project manager and I would need 10 people to do this and I could do it on my own. So I get that appeal for a certain kind of person. Often those people will also say, you should vibe code Peter Kafka. And I'll say, I touched basic in the mid-80s and the last time I've touched code they said, no, no, no, you should do it. One is this really something a non technical person can do? And should we be playing with this stuff or the novelty.

Paul Ford (13:51)

Well, pause for a sec. Right? Sure, if you want to. So you as reporter might want to go do a few sessions just to figure it out. But don't look at it that way. I do think the way that we look at technology will change a little bit as a result, because something that was very expensive is now available to everybody. But it does still require a set of skills and understanding. And what I will tell you is that I've been building a bunch of stuff and I've been working on it sometimes for months. And the code part is fast, but the actual product thinking and understanding is still really hard. And a lot of times I end up going back on knowledge that I've gained over the last 20 years. Like, you know, I want to use embedded vector database technologies and like, just like all that stuff that piles up over a career, it tends to come back and you don't have that. You have other skills. So we have to translate your question into do you have something software shaped in your life that you'd really like to see? A lot of people are like, I really want the to do list. I really want that to do list that is in my head and it does. One duck quacks whenever I click the button. They can have that. Absolutely. You could sit down if you want the duck quacking to do list. You want a better way to organize notes. You want an archive of your all the podcasts you've ever done, but searchable. These are things you really can build. Now you might need a buddy. We could do it right now.

Peter Kafka (15:14)

Because even that step though, and I've had this experience and I've compared notes with a few other people who are like me, who interact with technology and write about it are not technical. And we've gotten the same, you should make a thing. And we all are stuck on what would I make.

Paul Ford (15:27)

I think that's it.

Peter Kafka (15:28)

And when I've asked them like chatgpt that question, they come back with the most boring. Like, you could literally rename your file folders. And I'm like, well, that's wrong. So does that tell us more about sort of our failings as people who should know more about technology? What's the gap there?

Paul Ford (15:44)

No, I don't think that's that at all. I mean, I think it's you not seeing the world as a set of software shaped problems is good for your readers. Right. You see things in terms of the problems your readers might have. Right. So I don't really need you to go off and become a technologist if you're going to keep providing utility the world like you do. And in fact, if I was your boss, I wouldn't want you to. I'd be like, whoa, Peter's over there dabbling, don't do that. I want you to.

Peter Kafka (16:10)

Well, actually the bosses are like, go use AI. Yeah, but we can come back to that narrative, right?

Paul Ford (16:16)

I mean, what I can tell you, like, you would expect, look, these things are good at like making a lot of bullet points and PowerPoints. And you would expect that consulting would be nervous right now. It's not like numbers are going up. Consultants are really busy. There are certain places where people are getting laid off. But mostly it's like Accenture just made a deal with anthropom to have 30,000 Claude certified, this, that or the other. I think there's an enormous number of people whose job is to answer these questions. And they seem to be doing pretty good when we all thought they might be doing pretty bad. So I think what's happening is, yes, the boss is saying, hey, we got to accelerate everything with AI because they heard about it and they want to get those margins. And yes, some people are getting laid off because maybe they overhired back in

Peter Kafka (17:00)

the day or maybe they're not hiring the new college grads.

Paul Ford (17:03)

Yeah, that's right. But ultimately, you know what's funny is like, somebody asked me if they're like, what's this going to mean for local journalism? And I'm like, who is going to write the restaurant review? Who's going to be like. Because I think what happens is people are like, all right, well, this thing could cover little league baseball games just from the stats. But that's not what it's.

Peter Kafka (17:22)

We already have that we do.

Paul Ford (17:24)

Right. A lot of these patterns have actually been around for 20 or 30 years and people are just seeing them for the first time and they got a real nice glaze on them and they're like, whoa. Huh. It actually seems like a person. But ultimately like, I don't know man, you seem to be doing okay. I mean, so why would you go care about software?

Peter Kafka (17:41)

So the fact that I can't think of what I would vibe code other than like literally a joke app. Like I want an app that just tells me how great I am. Like ChatGPT always responds to my question. Wouldn't that be funny? Haha. Okay.

Paul Ford (17:53)

You don't see the world as product shape problems? No, I do. Right. And so like that is just like. But I don't know the value of you seeing the world in product shaped prompts. You should just call a guy like me.

Peter Kafka (18:04)

Because one of the other scenarios you people talk about excitedly is, well, this is going to open up the world and let lots and lots of people code. And in the same way that pick your disruptive technology and media blogger or Spotify or whatever, allow people to make music, GarageBand, make music or make media and distribute it widely. And yes, most of it sucks, but it will eventually allow a Mr. Beast, and whether that's good or bad, to create really, really popular stuff in North Carolina at age 19 with really no background. Does that scenario make sense to you that there are people who have coding and product skills that are going to be able to take advantage of this in a way that they couldn't have before?

Paul Ford (18:47)

I mean, take it with a grain of salt because I can't remember the exact stat, but I think year to year App Store submissions are up like 30%. So more people are making software more quickly.

Peter Kafka (18:57)

But I don't know if that's good.

Paul Ford (18:59)

I am a person who tends to believe that more people making things is

Peter Kafka (19:02)

pretty good just because you'll still have a hit rate.

Paul Ford (19:04)

Yeah, it also is really good with the performative and exhausting and rote aspects. Just getting stuff into the App Store, getting it built and tested is a real chore. Getting a website deployed onto a server is a real chore. So it helps with that. So that last mile gets a little bit shorter and I think that's good because it's boring stuff.

Peter Kafka (19:24)

I was going to ask you about. You mentioned this in a different conversation. First mile and last mile. That AI is good at first mile and bad at last. I guess you kind of answered the

Paul Ford (19:31)

second part well, no, it's good at clerical last mile of like I need to make something really good that people like that is just as hard as it was before. But the last mile of now I need to put it on a web server at a URL. And that kind of like that got a lot easier that Claire because it's just like everything always is changing and everything is like I got to use Amazon Web Services or whatever. It's freaking exhausting. And so a lot of passion projects and a lot of little apps didn't get shipped just because the. The form filling bureaucratic aspect of software sucks so bad. And I think that's gotten a lot easier as a result. So I think you're seeing just like friction go away there. But back to the earlier point you made about. I remember when I was an early. I was like an OG web writer even before the Word Weblog and I was convinced that what everybody would do if they wanted to participate in this world was learn how to program and build their own systems, right? Because I'm in my 20s and I just had a thought like that the reality is no, it took LiveJournal and Blogger and WordPress and all that stuff. When that stuff showed up and people were like, I just put the words in the box. That looks cool, I'll do it. But building infrastructure is actually a really. Only very few people want to build infrastructure.

Peter Kafka (20:38)

Even on the writing and music part, people will say they want to do it and then it turns out it's easy but not rewarding or I'm not good at it or no one's consuming this stuff and they move off.

Paul Ford (20:48)

I have Ableton Live, which is a digital audio workstation. I dabble with it. I don't have any talent in the world. It's really fun for me. I like to learn about music and synthesis. End of story. Right? Does that make me a musician? Am I suddenly this, that or the other? No, but I'm having a good time.

Peter Kafka (21:02)

Another scenario, maybe it's I guess related to what we've been talking about is no, you're not going to be a coder, but you are going to be in an org and you're going to be using word processing software or whatever and you want it to do a specific thing that's important to you and normally you would not be able to ask for that because that would be a whole new set of software. But now there's a world where you can tell someone, hey, add this thing or subtract this thing. Is that a world?

Paul Ford (21:27)

Absolutely. I think about This a lot. One of the things we built something at work that literally is like a lot of dashboards for health orgs. But you can talk to the dashboards. You can be like, how's this doctor doing? And then it answers by going to the database that wasn't there before. You couldn't do that before. Now you can do that. I also have a friend, different use case, but she's kind of always in my head. She works in immigrant rights and this use case is in my head because she has to cut and paste from. I think it's like Salesforce. Once a month or once a quarter, she has to do all this clerical work in order to go and get funded. It's just not a good use of her time. It's the opposite of what she should be doing. But she has to do it or she can't get her funding renewed. No one's going to do that for her. No one down the hall has got that on the queue. She can have that. Maybe not at that org. Maybe it's moving slow, but the future is you're going to ask questions of your systems and you're going to say, I need this report every month and there's no reason you can't have it.

Peter Kafka (22:22)

And you will direct that request, you think, to an internal IT person, to some equivalent of Accenture that's still providing software services.

Paul Ford (22:32)

You might address it directly to the box inside of.

Peter Kafka (22:35)

So that is the kind of work you might do. And you wouldn't think of that as coding and just be, I want this product feature.

Paul Ford (22:40)

I think that is 100% coming. You will say, hey, I really want this report once a week. And the little eager beavers inside of Salesforce, or whatever its mascot is, it's got that little bear boy. They go off and they start typing for you in little animation and it comes back and they're like, is this what you wanted? And I think you're going to see a lot of that with every SaaS and software and subscription tool, is that you'll be able to direct it a little bit more.

Peter Kafka (23:05)

Does that feel transformative or just like normal progress?

Paul Ford (23:08)

That's more transformative than the coding. You got to remember, I love technology. I think of myself as a programmer, but somebody once asked me, how do we get more teens into the world of digital power? How do we get them that? You can teach them to code, but if you really like, we're in Manhattan right now. The power in this town is not people coding. It's Excel and PowerPoint right. And so like teaching people really good Excel skills is probably better to give them more like actual economic power than even coding. And it's infinite what you can do with Excel and what they're doing with Excel right now. So anything that makes that easier lets you do more with it, gives it better, gives the data more integrity checks. The data is really, really good, kind of for society.

Peter Kafka (23:51)

We'll be right back with Paul Ford. But first, a word from a sponsor.

Narrator/Announcer (23:57)

I'm Mitch, first, two time IndigoCell champion, championship MVP and forward for the US Women's National Team. Before I went pro, I graduated from Harvard with a degree in psychology, which comes in handy more than you think. Any athlete pursuing greatness knows there's a certain mentality you have to have. What people don't know is what that costs. In my podcast, Confessions of an Elite Athlete, I sit down with the best athletes in the world and explore the psychology, mindset and unseen battles on the path to greatness. So take a seat and learn from the Confessions of an elite athlete on YouTube or wherever you get your podcasts.

Paul Ford (24:41)

Foreign.

Narrator/Announcer (24:46)

I'm Maria Sharapova and I'm hosting a new podcast called Pretty Tough. Every week I'm sitting down with trailblazing women at the top of their game to discuss ambition, work ethic and the ups and downs that come on the path to achieving greatness. We'll dive into their stories and get valuable insights from top executives, actors, entrepreneurs and other individuals who have inspired me so much in my own journey. Journey Follow Pretty Tough wherever you get your podcasts.

Kara Swisher (25:16)

Hi everyone, Kara Swisher here. We just won the Webby Award for the best interview show in news, business and society. And I've had some great guests on my podcast on with Kara Swisher. Here are some you don't want to miss. Tristan Harris, the co founder of the center for Humane Technology, I talked to him about his biggest worry when it comes to development and deployment of AI. Hint, hint. It has has something to do with the CEOs and how they stand to profit. I interviewed documentarian Louis Theroux. His latest documentary, into the Manosphere, focuses on the incredible and horrifying influence this group of individuals has, especially on young men and boys. And recently I caught up with Katie Couric, Amy La Rocca and my brother Jeff Swisher to debunk some of the fads and misinformation behind the billion dollar wellness industry. And we talked about the important medical tests that are actually worth your while. All of these conversations are available now you can find them on YouTube or wherever you get your podcasts. And we've got plenty more lined up for the summer, so be sure to subscribe to on with Kara Swisher to catch them all.

Peter Kafka (26:20)

And we're back. I've gotten pretty comfortable with the using ChatGPT the last year or so again, pushed by my employer to use a

Paul Ford (26:29)

lot more of it.

Peter Kafka (26:30)

And so I'm like, pretty comfortable that some of what it makes is good, some of it's okay and I can make it better. Some of it is bad and or wrong. I just asked ChatGPT to plot a trip for me and it gave me the wrong train time. I wouldn't have known that if I hadn't double checked.

Paul Ford (26:44)

Yes.

Peter Kafka (26:45)

How does that play out in a world of software where stuff either works or doesn't?

Paul Ford (26:50)

It's funny because there's so many ways for it to break. I'll tell you what's wild. The issue just described is totally real, which is you can end up with something very, very sloppy. But there are all these coding techniques like test driven development or type safety and sort of all these things that are really nerdy and really boring to implement, but they're not boring because you can just tell the system to do it. You'd be like, hey, I want. Every single variable needs to work this way and you need to double check them and you got to write a zillion tests and it's little simulated brain is like, okay, got to do it. And it adds an extra 20 minutes to the process, but it's checking its own work the whole way as it goes. And I think you've seen that pattern even in chat, right? Like if you tell it to go back and look again, it'll get a little smarter. It's just that it's just a robot in a loop, right?

Peter Kafka (27:37)

I mean, I feel like I'm just writing chat on an angry note. Like, I feel like a year into like me using it all the time. I shouldn't have to say check the actual train schedule because I'm going to take a train. It's important that I take the train at the right time. If you give me the wrong information, that's bad. And we just had to go through

Paul Ford (27:55)

that whole process that is inherent to the technology. It's incredible, incredibly hard to fix that. And so what you need to do is find loops where you're actually feeding it the train schedule and then getting it to interpret, letting it. I had this. I'm building something right now where I needed it to do a web search. And kind of, it's like a very light kind of fact checking, not real fact checking, but I just wanted to make sure that links were connected to ideas, right? And you can't get Claude to do that. And it's hard to get ChatGPT to do it, but Google Gemini will do it for you. So I was like, all right, let's switch over. We'll do Google Gemini just for that part, blah, blah, blah. It comes back. And I was like, oh, good. I click on the links, they're not there, they just don't exist. Because I forgot to say you're doing web search. It just was like, oh, they want URLs. I got them for you, man. And it made up like 20 Bloomberg links, which felt bad. I just felt bad for everybody. And so then once I figured out just the right little magic code, then it went and it always made sure to look at the web and that the URLs were real. And that was just computer doing computer. Like one of my big things with this thing is we everybody wants to give it everything all at once. Right now the reality is, to me, the first job of AI should be to clean up the old computer's mess. Like all those old invoices or all the old OCR documents or the podcast transcripts or whatever, Go clean all that up, index it, make it nice and easy to use, make it easy to search, and only then do we get to move on to like phase two.

Peter Kafka (29:19)

But even that indexing part, you feel like it is going to do at a high enough accuracy that you're not any more worried about it than if you'd asked a human to do it.

Paul Ford (29:28)

There's a really good rule. There's a technologist person named Laurie Voss, and he's great, and he wrote a blog post in which he said he was working in AI, works in AI. And it was just like, look, if it's going from more text to less text, if it's summarizing, you're going to get good results. If you let it make up text, you won't get as predictable a set of results. It's along those lines, right? And that is a really. So if you think about what is deep research, it goes out to the web, finds a bunch of text, and then it summarizes it. And it looks like a miracle because it summarized. It's like this big research report, but that's all it's doing. It's kind of like it's going out and it's boiling stuff down.

Peter Kafka (30:08)

One of the things we've taught ourselves in the last couple years is to identify AI, or we think we're identifying AI. That thing Donald Trump posted, that's an AI post and you're probably right. Or that person has too many fingers. Okay, that's AI. Or that writing has too many EM dashes.

Paul Ford (30:22)

God help the polydactyl community. I bet they have had.

Peter Kafka (30:26)

And by the way, some of the ChatGPT writing is writing that I like. But anyway, there's the sense that you can sort of see it, feel it, and again, you may be wrong. Does that exist in code? Can you see? Oh, this is code made by an AI and it's not good. It's off because a computer made it.

Paul Ford (30:44)

Unfortunately, it's the opposite. It's kind of really good and consistent because it's really hard to code and exhausting and everybody kind of gives up. So when you see one of the funniest things I've noticed as a writer, the writer part of my brain is you can always tell ChatGPT because the paragraphs are balanced and they're sort of nice. Everything is actually kind of organized in a way that humans just don't do. So what you get is all this output that doesn't have jagged edges. And with code that can be really good, all the docs are there and everything, it does tend to blob stuff up. There's all sorts of weird one offs. But in general, and this is the really tricky part, maybe the scary part is a lot of times you're not looking in, it's just telling you, I did it. And then you go to the website and the website loads and you're like, all right, I guess that's okay. Especially for the personal projects, if it's at work, everybody's got to read it. That's the rule. But if it's just Paul alone at night, nobody needs to look at that code you wrote.

Peter Kafka (31:37)

What in nerd circles is a very famous Bloomberg cover story, 2015. What is code? It's great. It's also technical.

Paul Ford (31:44)

Thank you.

Peter Kafka (31:46)

If I understood it correctly, you were essentially saying, hey, you, middle manager, upper middle manager. You actually need to know. You don't need to know how to code, but you do need to understand how this technology works.

Paul Ford (31:57)

It was a funny thing because that piece was about me explaining what code and what the technology industry. But yeah, it's business week, right? So they're like, could you make it for the R reader who is a manager. So I think if I'M talking to people now. What I want them to know is that it just got a lot cheaper in some ways. Not in every way, but like you can kind of have your thing now or you should at least be talking about having your thing and these sort of like, like if you've got like an 18 month delivery roadmap, you might want to really sit down and talk about that.

Peter Kafka (32:30)

So is the update from, if you update that story from 10 years ago, 11 years ago, is it everything I said is still true and it's just that much more important for you to understand this or because we have made the tools more accessible. You, manager should be more hands on about this process. Don't just hand it over to the nerds and let them come back in three months.

Paul Ford (32:52)

I mean, when I wrote that, I'd never really managed anything significant before and now I'm thinking about myself. Managing is funny. Like managers, you're trying to communicate in as few words as possible and you're existentially exhausted all the time. So I don't really want to ask them to do too much looking into the code. I think what I really want managers to know is that there isn't a magic trick here. It is a set of processes that are emerging. Your developers are probably as overwhelmed and confused as they've ever been. They're feeling sometimes defensive and sometimes excited. And you can either ride that wave or you can wait for it to turn out. What you can't do and what no one can do is kind of get control of this because it's so big so fast, we forget ChatGPT. What was it like 35 was the one that kind of broke over. That's 2022. We've had four years for the entire scope of the Internet to just happen in that amount of time. And it's really only in the last two that it's just gone bananas. There's a lot of hucksterism, there's a lot of false promises and there's a lot of stuff. There's something absolutely real here. There's a lot of transformational stuff. It's time to internalize it. But everyone is just as overwhelmed and no one really has solid answers, including the LLM companies.

Peter Kafka (34:11)

Again, appreciate that you don't know the answers, but I'm going to keep asking you anyway. There is this fear discussion about the fear that using AI makes you dumber. At least in the humanities. The more you ask it to write your book report, the stupider you're going to get because you did actually need to read the book. And also you needed to think about how to express your idea idea and express it. And you can now just short circuit the whole thing and ace your test

Paul Ford (34:35)

or ace your homework.

Peter Kafka (34:39)

Does that exist in code? Do people who were hand coding worry about what happens to their skills when they are outsourcing this stuff?

Paul Ford (34:46)

Coders are funny because a lot of them are relatively progressive and somewhat adjacent to the universe you just described. And they all just. Everybody just got real quiet around November. No, I think you'd be. It's a pretty intense position to say I will not be using any AI tools. I think just about the whole industry has transitioned into this at least a little. A lot of people don't have access. They have a little bit of copilot access. They don't have full on quad code Mac. So they're able to touch it and learn. But a lot of big orgs won't let this stuff in just yet. So a lot of people are working the way they used to. I don't think that they're absolutely. People feel they're getting dumber. Absolutely. But they also felt that way about copying and pasting from stack overflow or they felt that way about. Any intellectual industry always has this real intense hierarchy as to who's really smart

Peter Kafka (35:38)

or not, but also who's paid dues. And part of it is literally I paid my dues, full stop. Also I paid my dues and I learned stuff while I was doing that.

Paul Ford (35:46)

Yeah. But that never. In technology there's always is going to be a shortcut coming down the pike. Unlike in the humanities. Unlike in culture where you kind of really. Unless you do the work, you can't fake it. Then you're just plagiarizing if you're faking it. But here the shortcut is just built into the culture. So I think that it's a little more complicated culturally than that. But the reality is you need to be really honest about what you know and don't know. There are certain things that are just magical to do because you never could do them before and you won't know what's happening inside. And that would be really bad. Like I'm learning FPGA programming. Okay. Don't worry about exactly what it is. But let me tell you what it's absolutely critical to high frequency trading. Okay. So because it lets you just do a certain kind of thing really, really quickly. I'm learning it to make little synth noises. It goes beep boop. Very little chance that I can destroy the American economy that way. But if I was doing it, if I was feeling lazy and I decided to just go ahead and load up my chips with all the algorithms, I could take down my fund. And so that's where somebody needs to know what's really going on in that situation. I don't need to know when I'm playing. And the funny thing is, now that I've done a little of it, I know five times more than I used to. So I'm backing it out that way. So there's other ways to learn here too. You make it, it works. And you go, okay, it's not done until you can tell me what it's actually doing. That's okay, maybe.

Peter Kafka (37:11)

So you're still using your brain, using it different ways, different things.

Paul Ford (37:14)

I have learned more about coding in the last three or four months than ever knew in my, like, it's more actionable and so on and so forth, including lower level stuff. But that's not every brain. A lot of people are just going to be like, I got it done. And there is a lot of that going on. There's a lot of people going. We allow people to vibe code. They turned in good working code, but it's very clear that no one has read it and it's been automatically tested. Now it's my job to review it. What am I reviewing? There's a lot of really existential questions about what the job is. Anyway, we'll be right back, but first,

Peter Kafka (37:46)

a word from a sponsor.

Paul Ford (37:49)

I'm Asted Herndon and this is America. Actually, we're all talking to each other to see what did we do wrong? What did we not see?

Peter Kafka (37:56)

I'm in Washington D.C. this week to interview Ruben Gallego.

Paul Ford (37:59)

He's a Democratic senator from Arizona and he's been thinking openly about running for higher office. But he's recently run into some hot water because of his connection to Congressman Eric Swalwell. I have to learn from this and I will learn from this. But, you know, for me, it's not a 2028 question. It's about what it means to be a better first boss in my office and also a better senator to my constituents.

Peter Kafka (38:24)

This week on America actually, we asked

Paul Ford (38:26)

Gallego about predatory behavior in Washington, his plans for immigration reform and more. Is the U. S. China rivalry ultimately a race to build the future? The United States and China are the two countries that are really inventing the future. The future is being financed by Wall street, invented in Silicon Valley as well as Shenzhen. I'm Jake Sullivan and I'M John Finer and we're the hosts of the Long Game, a weekly national security podcast. This week, author Dan Wong joins us to discuss America's lawyerly society, China's engineering state, and why derangement might be a prerequisite for Superpower. Our status. The episode's out now. Search for and follow the Long Game Wherever you get your podcasts.

Narrator/Announcer (39:15)

Wedding season is here and your wallet is already sweating. Between the Bachelorette in Vegas, the destination ceremony, the registry gifts, and the outfits for every single event, being a good friend has never felt more expensive. I'm Vivian to your rich bff. And on this episode of Net Worth and Chill, we're breaking down exactly how to survive wedding season without going broke. We're talking hidden costs. You forgot to budget for how much you actually need to spend on a gift flight and hotel hacks that could save you hundreds. And my most unhinged but totally legal money, tips for stretching every dollar. Because celebrating love shouldn't mean sacrificing your financial future. Listen wherever you get your podcasts or watch on YouTube.com YourRichBFF.

Peter Kafka (40:00)

And we're back. The normals are being told we need to learn to use AI. We need AI skills. Reese Witherspoon said, it's time for us women to learn AI. I think I properly criticized for that. But beyond the gender part of it, is learning AI something middle aged people should do? Should my kids be learning AI or is the whole premise of AI that it just does the things for you without you having to learn it?

Paul Ford (40:26)

I mean, I'll tell you what I would really like people to know is I'd love it if more people understood this as a technology. When people understand it as a it's a weird magic box that does some stuff and they don't know if they like the dudes who are in charge of it. And there's a lot of folk narratives that are emerging around that. A good example of a folk narrative that happens to often be true is, boy, there's a lot of data centers. And I don't know if I like them in my community because they're emitting a lot of pollutants. And they're big and they're big, they're

Peter Kafka (40:53)

ugly, and I didn't vote for them.

Paul Ford (40:54)

Yeah, exactly. So that's simultaneously something people have rallied around, but also a point that makes sense. Right? But I think that we don't have that going in the other direction. We have like, no, no, it'll do your taxes. It's gonna should I marry It I don't know. And the reality is you've got got a really complex and weird set of technologies with its own history that actually have very hard limits. There's just certain things they can do and they only seem intelligent. They're simulating language, they're not even simulating intelligence. Or maybe they are and you can have that fight. But I feel that when I think about. So my son, I have 14 year old twins. My son wants to, he wanted to learn to program and it is a little tricky because I'm like. And he was into it, he wrote a book about Python and so on. And I'm like, I still want him to learn that. I think it's really important he learns it. I think that he will have tools that will teach him if he is engaged, if he wants to learn, he will have a tool that will let him learn a zillion times more quickly in a highly interactive way. In this, if he just wants to slop together some bad game, throw it together and it kind of works, he can also do that.

Peter Kafka (42:07)

You don't have to learn Latin to know how to write English. Maybe it's helpful to have known Latin.

Paul Ford (42:14)

There is that. But at the same time, if you want to make the game better, you got to actually go to the root and figure out what's happening. And so I don't know, when you're coming up as a journalist, you had to rewrite and revise a lot, right? And I think if you hadn't done that, you wouldn't be is good, you just wouldn't, you know it. And like what is that process here? What is revision like? Revision can be talking to the LLM and saying make it better. It can be looking at the code, we just don't have it yet. It's too soon.

Peter Kafka (42:44)

I think I know what you're going to say to this one, but what do you make of this token maxing moment we're in where the tech companies and now non tech companies are bragging and incenting their people to use AI so they can burn tokens and then bragging.

Paul Ford (42:59)

So there's two schools of thought here and there's one. There's this idea that like I'm going to put in one prompt and the little eager beavers are going to go and they're going to make me so much software and they're going to come back and they're going to say here's your software sir, and I'm going to kite board the whole time. Okay, that's like the Silicon Valley narrative right now. And so what they're doing is they're like, guys, go as many agents as you want. Show me progress, solve every problem. But I don't think most problems, there is a class of problems that can break down to that. Like really big, big. Make me a custom database that does blah, blah, blah, like really, really granular, boring, big stuff. Maybe, maybe some of it breaks down that way, but I think it's just bros broing out. What I'm finding with this stuff is even I burn a lot of tokens personally. I have two Claude super max accounts. I'm using the prototype because now I'm doing stuff at work to kind of like prove out points and to pitch and. And I'm doing stuff at home. So I'm probably in a weird way a token maxer. But it's iterative and I occasionally send it off to do research and it

Peter Kafka (44:06)

comes right, but you're not bragging about your tokens used and you don't have a leaderboard where you can say I've

Paul Ford (44:12)

ashamed of my token maxing.

Peter Kafka (44:14)

This is the right time to talk about.

Paul Ford (44:15)

Yeah, no, no, no, I don't like the ecobox. I'm going to a climate event after this.

Peter Kafka (44:19)

I guess what I'm getting at is from the outside, it seems to me pretty obvious this is, hey, we must show progress. This is a way we can demonstrate it, even though it may not actually. Not only does it not really show progress, it might be counterproductive. But I can also imagine people saying, look, we just have to get the things moving and if we leave it up to the individual workers to decide how much AI they want in their lives, they're going to get lapped by the guys down the street or just by the technology is going to come swamp our company. I don't want to get too specific here, but I can think of several media companies where there's just a lot of impulse to use AI and it seems kind of not well thought through. But also waiting for AI to come take over your business is also bad. And I can see them just saying, let's just go run at the problem. And yes, we will run into the wall, but let's do it.

Paul Ford (45:10)

You're back to terms like change management and governance, right? And nobody wants to have that conversation because that's. That would get in the way of all the fun. But that's what we need. It's sort of like, look, let me take a deep breath back and just sort of be my consultant self for A minute. Not my writer self, not my technologist self. Enormous, significant changes have occurred and there are tremendous economic impacts. And they're happening on an industry, industry, industry by industry basis. And it's very similar. It's new infrastructure in the world. It's not one new thing. It's as if suddenly they just replaced all the plans. Plumbing. And you're like, whoa, this is weird. The toilet screams when I flush it. And you're like, well, okay. But yeah, that's how it works now. That's where we're at. And so you have to figure out how to live in a world of constantly shrieking toilets. And what is your next step and what do you do? But what this involves is that the CEO ideally would be the person or the CTO or the CIO would be the person who would really understand that there's a lot of ambiguity here and a lot of concern, and wouldn't just put foot on the gas in order to get maximum revenue or to, say, burn more tokens, but would be like, what are the outcomes that are going to actually help us drive forward the business that we already decided that we're in? And everybody's like, well, wait, what if we could be. And this has happened every 10 years. It's like, we are now a technology business. And they always put like a. There's always a prefix or a suffix. For a while it was E, then it was I, now it's AI. And I think we're kind of that simple as a species. So to me, I'm just like, boy, could you just. You gotta take a breath for your people.

Peter Kafka (46:51)

Let's wrap it up. By going back to where we started and this idea of AI disruption, you've made a convincing case of what this is doing to software. And I think we sort of have our heads around. All right, there's probably a good case that entry level work at places like consulting firms, law firms, where you're doing a lot of drudge work in your first couple years that is impacted anything.

Paul Ford (47:14)

Where the job was to search the web and cut and paste things you found that's not as safe as it used to be.

Peter Kafka (47:20)

So that's impactful enough. Do we think this is also a precursor to other changes in the way we work and interact with the world? That sort of. Because there's a lot of people who are listening to this go, I don't do any of that data entry equivalent stuff. I'll be okay. Should they feel that they're going to

Paul Ford (47:38)

be okay, People should be aware of this stuff. I'll tell you, it's not necessarily that AI is coming for your job, but your life touches software and systems in a lot of different ways that you may not be fully aware of. That might change. Your bank software might change. There's all that. But I do think that one of the things I learned is a lot of times I'd hire people, I'd work with people, and you'd realize people don't often know what business they're in. They don't know where the money comes in. They know that they're a graphic designer, they know they're an accountant and they do the work. But they may not even. Even the accountants may not really understand the revenue. And somehow the salespeople always make a lot of money because they're close to it, right? Go figure out where the money comes in, because that is actually like there's just going to be a lot of money.

Peter Kafka (48:27)

It's also the best beat reporting advice ever. Understand the business you're going to write about. You think you do, but you actually, where does the money start and where does it end up?

Paul Ford (48:35)

And I tell you, actually, AI is fantastic at this. It's incredibly good at explaining where the money flows because it's got all that generic information. Go understand that, because that's where change is going to happen. And it might not be your department, but it's like, just like get your head around it and figure out where you want to go and who you want to stand next to, as some change might come down. The bike. The one thing that I keep an eye on, there are three or four really major, what are called world model labs. As I was coming over, I was on a Citi bike and I was right behind a Waymo because Waymos are kind of piloting New York and they still have drivers, I think, but they're checking us. And I mean, I love them, I love Waymos, I can't help it. But anyway, regardless, let's put that aside. Waymos are run on a thing called a world model. The world model is different than a large language model. It has a whole model of how physical reality works. And then Waymo, with a good world model coming down in price in New York City, is a real job muncher. It is not great for jobs, just flat out. And so I like the product, but I also like, like people having jobs. And so that is real tricky. And that's going to be like governance has to figure that out. World models could also help robots make things in factories. World Models could trim your hedges. And so we're not there yet. It's just not there. But that is the thing that I will say. I don't know if the large language models really keep me up at night because I just see so many new things happening while maybe some old things are going away. And I just have that of kind. Kind of that Internet instinct that, like, it's probably going to be a wash because everybody just.

Peter Kafka (50:11)

But if Elon's dream of the robot comes true, and many people are dreaming

Paul Ford (50:15)

of the robot, I can't even remember China's robot. It turns China into like one giant 3D printer, right? And so that would be. And we're not there yet. But that's the only sci fi scenario where I'm actually like, boy, we are not ready. We are not ready for all physical manufacturing to be handled. Because you could prompt it and it would write some code to run the robot, and then the robot can prompt itself. And now off we go.

Peter Kafka (50:39)

I told you I was wrapping up, but I'm wrong because you reminded me. Sometimes in your writing you say, I'm going to get yelled at for this. You're very defensive about it. And I didn't really fully understand what you were talking about. And then it became clear when I was researching this. You're on Blue Sky.

Paul Ford (50:54)

I was, yeah.

Peter Kafka (50:55)

Oh, you were?

Paul Ford (50:56)

No, I am.

Peter Kafka (50:57)

And I don't spend much time there. But apparently you're getting a rash or a shit for saying anything remotely positive about A.I.

Paul Ford (51:03)

i knew it. And look, I guess what I wanted

Peter Kafka (51:06)

to get at is the Blue sky versus Twitter of it aside, how do you think about talking to. And again, we're carving out Blue sky and Twitter, which have their own sets of weirdos. But just talking to a normal person who is thinking about this, whether it's their job they're worried about or their kid's job, or they actually feel really uneasy about being in a Waymo.

Paul Ford (51:27)

That creeps them out.

Peter Kafka (51:28)

How do you talk to them about that without sounding like you are working on behalf of Elon Musk and Mark Twitter?

Paul Ford (51:34)

I mean, a certain number of people will say that if you say anything about this technology that isn't first and foremost about its cost to society, its ecological cost that you shouldn't, then everything else you say is suspect. But most people are just trying to figure out where they're at. And what I see. I think what's really tricky here is a lot of people have kind of made a little bit of their identity. They tend to be very they're kind of humanities adjacent and they're like, stay away. Don't want anything to do with this. And then I did an interview recently with somebody who's super low vision named Andrew Leland. And these are amazing tools for him because he's able to hack tools for himself as a very low vision blind person with this stuff that he didn't have access to before. But he's also a writer and he really feels mixed about it. And I think those are normal feelings. I just, honestly, I didn't even want to come back out and start writing and doing stuff about this. I like the technology. I find it fascinating, but I didn't want to participate in culture. But whenever I get asked, I feel that I kind of. I'm on the hook. I'm like a technology explaining guy, right? This is the biggest thing that's ever happened in my career. It's not the only big thing, but it is a wild thing really quickly. And I sort of sat with myself and I was like, I am going to get yelled at because people don't want to hear what I have to say. And I could be wrong. That journalist part of myself, I just really could be wrong. But what I see is a really significant change. I find it very overwhelming, sometimes very scary. But it's also incredibly interesting. And things that I could not do for 10 years, I can do in a few days. And I felt ethically, and I still feel ethically on the hook to just narrate that change. And if people think this guy sucks, he should go to hell. He is sitting in Anthropic's back pocket. Then that's fair. All more power to you. I don't kind of care. But it's too big of a change to turn away from, given what I know about this industry. And so I lean back in and it feels like, you know in the movie when the guy's got his kitchen table and the helicopter lands on the lawn and they're like, we need you back in for it.

Peter Kafka (53:44)

One last job.

Paul Ford (53:45)

No, no, Me and Cindy are doing a good job. I don't want to go. I don't want to go. And so that's where I'm at.

Peter Kafka (53:53)

We're gonna leave it there with the aliens coming to take you into the helicopter.

Paul Ford (53:56)

They're normal aliens. They're just normal.

Peter Kafka (53:57)

Well, they're men in black suits.

Paul Ford (53:59)

Yeah, yeah, just your normal colonel.

Peter Kafka (54:01)

And if you don't like that image, just imagine toilets screaming. That's the image I can't get out of my head. Now.

Paul Ford (54:05)

I think I'm actually thinking of the movie MacGruber.

Peter Kafka (54:08)

Paul Ford, you're awesome. Thanks for coming.

Paul Ford (54:10)

Hey, thank you. Anytime.

Peter Kafka (54:12)

Thanks again to Paul Ford. Really great to have him in the studio. Thanks to Charlotte Silver, who produces and edits the show. Thanks to our sponsors, who bring it to you for free. Thanks to you guys for listening. See you next week.