Paul Ford: Hi, I’m Paul Ford.

Rich Ziade: And I’m Rich Ziade.

Paul: And this is The Aboard Podcast, the podcast about how AI is changing software.

Rich: AI. AI. AI.

Paul: My God, it’s a lot, right? Okay, so we have so many things going on. Let’s actually talk about the—

Rich: Let’s dive in.

Paul: I made a deck.

[intro music]

Paul: I’m wearing a light-colored shirt but a concealing sweater. You’re wearing a dark-colored shirt. We’re getting a lot of feedback about how to orient ourselves. So we’re trying here, folks. We’re trying. I’m trying. I’m looking at you right now.

Rich: Yes.

Paul: Okay? And we’re also gonna try to introduce a little AV, but real early start today. Okay? I did this this morning. We’re gonna try to, like, visually up our game a little bit.

Rich: We sound like executives who can’t figure out the WiFi.

Paul: It’s real. This is, that is what we are. Okay, so what I want to talk about today, today the day—

Rich: Yes.

Paul: —is right now, an enormous amount of news is coming out about new stuff in AI.

Rich: It’s funny, I have a sort of parallel in my brain that, like, the velocity of AI is sort of, like, the way it sort of spits out stuff is similar in the news about AI. It’s like AI behaves like—AI news behaves like AI.

Paul: Well, and actually there was a—

Rich: Very weird.

Paul: —syndicated book reviews that were AI generated in a lot of major papers, [laughter] which was another sign of like, everything kind of falling apart.

Rich: Yeah.

Paul: But what I thought we could do is, like, a brief recap of, like, some of the stuff that’s going on. And then I want to talk about some themes, because I actually think that the narrative—you and I have been talking about this, the narrative of the AI companies is very different from what’s actually happening.

Rich: Yes.

Paul: So—

Rich: But there are keynotes every two weeks. It really feels like it.

Paul: So Google I/O was a few days ago, as we were recording this.

Rich: Google I/O predates AI.

Paul: Yeah.

Rich: It’s Google’s big event.

Paul: So they had an article. It was, like, they launched a hundred things. They’re launching so much AI stuff. Just—and not just AI, like, web and Android and I mean, it’s Google. They’re huge.

Rich: Yeah, sure.

Paul: But what’s interesting to me is they’re really going after developers around AI and there’s a, you know, I was watching one of the videos and there’s this real focus on like a stack emerging inside of Google.

Rich: Okay.

Paul: And you’ve got your, your models, you know, your LLMs.

Rich: Mmm hmm.

Paul: And then you get your AI framework top. And this, if you’re a software person, this is just classic. Stacks refer to layers of abstraction, right? And the very bottom layer of abstraction in the AI stack is going to be that, that model.

Rich: Mmm hmm.

Paul: And then you go up to the developer tools, where I’m writing my code and I’m organizing things and I’m working with a team.

Rich: Classic stuff.

Paul: All the way out through deployment.

Rich: So just real quick, I didn’t watch Google I/O, because I have a life. Unlike you.

Paul: Yeah, no, I do not. Yeah.

Rich: Was this, like, the big reveal or no, this was one of 20?

Paul: There is no one big reveal.

Rich: Sure.

Paul: There’s a lot going on that’s, like, optically exciting. Like, their video generation is getting surreal and really good.

Rich: It’s scary. I watched a few of them and it’s scary.

Paul: Like, VO3 is good and it can talk and so on.

Rich: Yeah. There’s a few announcements. Okay.

Paul: So there’s a lot of big announcements in optics. But what you’re seeing is them essentially trying to get the Google back into this world.

Rich: Yeah.

Paul: Like, Gemini is going to have these new features, and then you’re going to write apps with it and deploy them on Android. Like, they’re trying to bring it and lock it into their ecosystem.

Rich: Yes, yeah.

Paul: So there were a lot of, like, conversations where people just had code on the screen and they’re talking about how the code is going to talk to the LLM and it’s going to call functions and the word agent shows up. But it’s, like, a hundred things. Like, dozens and dozens of them are AI. But it’s all starting to live inside of that big Google ecosystem, right?

Rich: Sure.

Paul: Let’s go over to Anthropic for a minute.

Rich: I’m a fan.

Paul: I like Anthropic, actually. Of all them, I kind of find Anthropic, with Claude, kind of the most…feels like software to me.

Rich: Also, lowest drama, maybe? Generally? Just, I don’t hear a lot of dramatic Anthropic news.

Paul: No, no.

Rich: So thank you for that.

Paul: They actually seem to be managed like a company, which is exciting.

Rich: Yeah, yeah. All right, what is this?

Paul: So they’re throwing out a whole lot of stuff. They’re doing, like, they’ve got Claude playing Pokémon.

Rich: That’s cute.

Paul: It’s cute. And what it means is it plays Pokémon and it learns what works and doesn’t. And then it goes and it creates another prompt for itself to be like, “Okay, here’s how I’m going to play a little better.”

Rich: Right.

Paul: And what they’re really emphasizing is they’re going to just be, it’s going to think a lot longer. Like, it’s not drastically smarter.

Rich: That seems to be a common theme.

Paul: But it can go away—

Rich: Thinking a lot longer.

Paul: It can go away for a half hour or an hour and work on something. And then something might be code.

Rich: Mmm hmm. Mmm hmm.

Paul: And then at the end it might deliver a pull request to GitHub, like, “Hey, I fixed this for you.”

Rich: Yeah.

Paul: “Do you want to bring it into your world?”

Rich: On its own?

Paul: Yes.

Rich: So, meaning you didn’t say, “Go fix a thing.”

Paul: Well, you might have said, “Go fix the thing.” Or you might actually have it in GitHub, and you might go, like, that’s one of their examples, like, you know, “Hey, there’s a problem with this merge.” And the person responds, “Claude, could you address this feedback?” And then Claude goes in as the programmer and kind of fixes what the GitHub issue is.

Rich: Okay.

Paul: So then you get into this world where like, “Hey, we wanted to deploy software, we ran the test, the test didn’t pass.”

Rich: Yeah.

Paul: And instead of me going, “I should find out why the tests are passing or not,” you say, “Claude, can you do that?”

Rich: Yeah.

Paul: And then it reports back. Good? Not good? I don’t know. Because actually, the more I think about this stuff, the more—I had a talk with one of our engineering leaders yesterday, literally about this stuff, and he’s like, “The thing I’m really worried about is how are we going to promise really good quality software to our buyers if we don’t know the code?” Everybody needs to be involved all the way through the process, because on Thursday night when the server goes down, you can’t be like, “Claude, fix it.” You have to—somebody needs to know what is in there.

Rich: I mean, that’s a theme across a lot of what AI is helping you shortcut. We love convenience.

Paul: Yeah.

Rich: Even at work, we love convenience. But if things go south and the boss or the client or the customer wants to know what’s up and you don’t have a clue because you’ve never looked inside? You’ve never had to?

Paul: Did you ever read about, like, food and medicine before the FDA?

Rich: Must have been bad.

Paul: It’s just, like, sawdust and morphine. And you’re like, “Wow, I just wanted white bread.”

Rich: You know what used to throw me off is that like the tonic—first off, you got sodas at the pharmacy.

Paul: Yes.

Rich: And the tonic kind of had, was delicious and refreshing. But it had cocaine in it.

Paul: Yes.

Rich: [laughing] And it was also medicine.

Paul: Yes.

Rich: But it also was, was, it was thirst-quenching.

Paul: Because this was an awesome country. It was so good. [laughter]

Rich: We can’t do this anymore, man.

Paul: You can go to the weed store, but they’re closing all those down.

Rich: No, well, they shouldn’t be selling out of the back.

Paul: That’s a really—

Rich: That’s a different thing. That’s New York City.

Paul: Or the front. Or the side.

Rich: Yeah.

Paul: Okay, so—

Rich: Keep going.

Paul: So, Claude—

Rich: Okay.

Paul: And the other thing, too, like—agents. Agents. Agents. Agents. Agents.

Rich: Agents is—I have a, I get it. I think it oversimplifies, but I think the reason agents have taken is because it’s a great marketing tool. It’s a great way to communicate what’s going on.

Paul: We’ll come back to that a minute. Hold on to that for one minute. Let’s—

Rich: Yeah. So OpenAI.

Paul: OpenAI. OpenAI, interesting. Now, MCP is a way for people to talk to—it’s an API protocol for, like, talking to systems and getting results from, from an LLM. You send it a prompt and you get data back.

Rich: Yeah.

Paul: And so it’s just that plus a lot of LLM gravy and niceties. Anthropic proposed this standard. It’s taken a while, but now OpenAI is like, “Hey, we’re on board. We’re gonna do this with you.”

Rich: Okay, so this is important, because it’s the first cross-platform, platform-agnostic standard that if people adhere to it, I may have Anthropic tools doing some specific task and then talking to OpenAI tools over here to finish the job, for example.

Paul: Correct. So like, so, so that’s really good. Like, that means something. We’ll talk about that in one sec. It does stand for Model Context Protocol.

Rich: Yeah.

Paul: So that’s helpful.

Rich: Yes.

Paul: That’ll get everybody aligned. But here’s, like, what can you do with it? You can say, like, let’s say you’ve got a store on Shopify and you’ve got a large language model that understands stores on Shopify. You can say, you know, “Show me all the pink shirts with funny slogans.”

Rich: Yeah.

Paul: Right? And like that, and if, you know, Shopify’s MCP server is supporting that kind of input and that kind of query, you can start to do AI-y things with websites and with databases and platforms without having full access to all their data, or so on and so forth.

Rich: You can make a call.

Paul: That’s right. So if I have a big Shopify store with 900 different components, I could be like, “Which one of these are compatible with Sony Stereo?”

Rich: And it’ll go out and check some…

Paul: I mean, it does its thing tool.

Rich: Yeah.

Paul: It does its thing. So it’s a way for, like, AI to do its thing in the context of the web. And so, you know, HubSpot is supporting MCP. So like, so you know, you can be like, “Tell me all the leads that are in New Mexico.”

Rich: It’s a way to open that knowledge capability, that AI capability, to the rest of the world.

Paul: And kind of, yeah, glue all those worlds together. And so our friend Anil Dash wrote a big piece. We’ll put a link in. But you know, “This Is Web 2.0. 2.0,” which the idea is like, hey, we’re kind of getting back to the place where instead of everybody running off in their own direction, they’re starting to talk to each other again, and there’s things to do—knitting things together is always good for an open commons and ecosystem. It means that you get open stuff and closed stuff and everybody’s kind of talking and working together.

Rich: I’m going to share a less optimistic—not optimistic, suspicious view of MCP.

Paul: Sure.

Rich: I think the way I look at it is AI showed up and started talking to people first.

Paul: Yes.

Rich: It became a human interface tool, the chat box and a robot responding back very thoughtfully blew everyone’s minds. We’re still kind of processing it. And it’s giving me pictures and it’s writing me essays and it’s breaking education. We’ve talked about all this. And what MCP is proposing is that now machines can talk to machines.

Paul: Mmm hmm.

Rich: Right? Ultimately, it’s usually sparked by a human, for some goal or purpose.

Paul: For now.

Rich: For now. But if it has to go out and fan out and have HubSpot talk to Claude AI via this protocol, that’s something it can do. I think when you look back on what Web 2.0 promised and where we ended up, we still ended up with, like, five companies. Yeah, so that’s a bigger thing. And I think what’s happened with AI is that the borders around the different countries were drawn, like, in the first 10 minutes. Like I’ve never seen—I’ve been in tech for a long time, and I’ve never seen four mega-companies sprout out of the soil in like a minute.

Paul: It was like a James Cameron movie. Like, sort of—

Rich: Yeah, it’s just unreal. And I’ve seen, in the past, companies, you know, hat in hand, agree to standards.

Paul: Oh yeah.

Rich: You know, it’s like, “We’re on board. Of course we’re on board.” But then what you see is the classic mechanisms around creating moats and such and such. So that’s—

Paul: I’m drinking a lot less these days, but if you ever want to get a couple drinks into me, we can talk about the OpenOffice XML versus Microsoft Proprietary XML.

Rich: I think it’s the only way to get you to talk about it.

Paul: It’s a very deep and buried set of emotions.

Rich: Yeah. All right.

Paul: So the other big—

Rich: [sighs and laughs simultaneously] This is—

Paul: The other big open AI is—remember, a lot of people are listening to this. They can’t see the slides. You got Jony Ive wearing his Jony Ive glasses. Kind of got his arm around Sam Altman’s shoulder, and he’s just whispering the word [deep British voice] “aluminium” into his ear. Just, “Aaaaluminium.”

Rich: I can’t process if that’s a real photo. That is uncanny at this point.

Paul: So OpenAI bought…Johnny Ive? And his company I/O. It’s confusing.

Rich: Johnny Ive had been working on—

Paul: Big ticket, though. $6.5 billion, billion with a B.

Rich: Of AI money.

Paul: Who knows? Who knows at this point what’s real?

Rich: We can’t even tell if the photo is real.

Paul: I mean—

Rich: Let alone how they paid for the thing.

Paul: Don’t forget, also, Altman’s got that World Coin thing. It’s the orb that scans your eyes so that you can get the Bitcoin. All right, so that’s what we’re up against.

Rich: He’s moved on from that. I’m gonna call it.

Paul: No, they’re still running around with that orb.

Rich: All right, fine. Run around with the orb. Yeah, so Ive had been working on, sort of, new interfaces. A big premise for him, by the way—and I just happen to have been reading up on what his goal is—is to get people to benefit from all this technology without staring at screens. That’s like a big goal of his. I don’t know if you knew that.

Paul: Altman or Ive?

Rich: Ive.

Paul: Yeah, I mean, because you got to show the old boss what’s up.

Rich: Yeah. Show the old boss, even though old boss is not really around anymore. But yeah.

Paul: Yeah. Well, the old new boss.

Rich: Old new boss. Yes. So, and—

Paul: They’re going to regret letting him go with their stupid phones.

Rich: There’s a nine-minute video on the internet of these two just sharing an ice cream sundae, and it is just unwatchable. The whole thing is unwatchable. God bless.

Paul: I want to bring business back. You know, we just—so we were opening a new office, right?

Rich: Yes, we are.

Paul: And we had a big conversation about what to name the conference rooms.

Rich: Yeah.

Paul: And where did we land?

Rich: 1, 2, 3, 4.

Paul: Yep, that’s it. I just want—I want to get back to that, man.

Rich: Yeah.

Paul: Just like—

Rich: Trees were in the running. Birch…

Paul: I made a spreadsheet. [laughter] I had HTML elements, because I was supposed to be the creative one here, right?

Rich: Yeah, yeah, yeah.

Paul: It just kind of came down—

Rich: “Ended up to 1. I’ll see you in 3.”

Paul: Because we’re about business and it’s just easier to like—

Rich: Yes.

Paul: So anyway, but they’re sharing—

Rich: Anyway, I’m curious to see what they come up with. I’m sure it’ll be fascinating.

Paul: I can’t wait. [laughing] They’re buying the keynote, you know, they’re buying the like, [even deeper British voice] “Aluminium.”

Rich: He’s a thoughtful, out-of-the-box thinker who views technology as submissive to him, which is actually incredibly effective. And credit to Steve Jobs for pretty much making him his peer for years. And it created immense value. I don’t blame them. $6 and a half billion of, like, Eyeball Bitcoin Money? God bless.

Paul: Who knows? Who knows what money even is anymore? All right, so that was our news roundup, and I see three themes emerging.

Rich: Okay.

Paul: Let’s run through all three and then go back. So one is interoperability is starting.

Rich: Great.

Paul: Big companies are making it so that these pieces of software can talk to each other. The youth might not remember, like, you couldn’t share a word-processing document for 30 years.

Rich: You had to print it.

Paul: Yeah, it’s just, like, Word and WordPerfect couldn’t open each other’s stuff.

Rich: Yeah, yeah.

Paul: So interoperability is actually a big deal. And the fact that these things can talk is how you build a tech industry.

Rich: Mmm hmm.

Paul: The other thing is that agents are showing up. And I think agents represent kind of a first pass at what are the methods we’re going to be using to talk to these systems.

Rich: Mmm hmm.

Paul: And there was a good—this is something I wanted to come back to, which is, there was, Simon Willison, this is from his webblog. I’m going to read it to you. He went to the Anthropic Developer Days. He was getting frustrated at the fact that every talk at the Anthropic developer conference has used the word agents dozens of times, but nobody ever stopped to provide a useful definition. “So I’m now in the prompting for Agents Workshop and Anthropic’s Hannah Moran finally broke the trend by saying that at Anthropic: ‘Agents are models using tools in a loop.'”

Rich: What a ridiculous definition.

Paul: I think that’s good, though. It’s a model—

Rich: Okay.

Paul: —and you use some code on top of it, and you’re doing it over and over again. You’re not just prompting it.

Rich: That’s a better definition.

Paul: No, it’s the same. I just said the same thing.

Rich: I’m going to give you credit for this.

Paul: No, give her credit.

Rich: Fine.

Paul: The agents are model using tools in the loop. So it’s, like, “I’m going to run this code and I’m going to kind of keep running it and I’m going to wait until I get the result I want and then maybe I’ll run some other code.”

Rich: Yeah.

Paul: That’s all it is. Because we’re going to keep saying the agents are like little people, out in this industry.

Rich: Yeah.

Paul: But that’s it.

Rich: Yeah.

Paul: And then I think the third thing, so we got interoperability. We have agents as a model. And then I think developer experience is starting to become really important. We didn’t talk much about it, but Claude is getting way more focused on code. They updated their code client in the terminal. They’re starting to work with third-party code tools like VS Code in a more mature way. And Google was all about developer experience.

Rich: Yes.

Paul: Here’s what I see. I see huge, massive progress, big announcements—Jony Ive and so on. But it is starting to get folded into the classic structures of our giant tech industry. I need developers to build with this. I can’t keep all the value myself, as the AI company. I have to distribute it out to developers, who will go build the next generation of products.

Rich: And it’s interesting. I mean that’s a, a lot of people view that as a corner of all this, not the thing.

Paul: I think they’re starting to say, “Whoa, I guess we’re going to need that.”

Rich: Interesting.

Paul: I don’t think Gemini will be enough. I think they’re really focusing on, like, you’re going to build apps up—

Rich: Is that 10% of this revolution?

Paul: No, I doubt it. I doubt it. I think it’s—

Rich: Is AI going to revolutionize, like, how restaurants are run, and then how factories are run?

Paul: Here’s what—well, it depends. There’s a lot of different ways. I’ll argue pro…let’s go factories. I’ll argue pro and con.

Rich: Yeah.

Paul: And we’ll start with con. No, not really. There’s going to be some AI toolkits, code that operates robots is going to get a little faster, and you’re going to be able to automate a lot of things around the factory, and you could do kind of like an AI walkthrough.

Rich: You still need developers is what you’re saying.

Paul: You still need developers to build that stuff and you’re going to need to train people and you can’t really get that last mile, the humans, out of the loop. You still need the humans, but you might find a lot of efficiencies.

Rich: Yeah.

Paul: Phase two here, the pro would be, we’ve got all these automated systems for manufacturing and doing things in the factory. Those are controlled by code. And we’re going to continually iteratively improve how the code gets created to operate the robots to make them more efficient, including having an AI system monitor how everything is going, automatically suggest improvements, and we’re going to have this incredibly accelerated continuous improvement model—

Rich: Okay.

Paul: That sometimes will be automated. You won’t need consultants. They’ll just make everything better in the factory. And we’re going to start working towards that because we think in a couple of years, it’s just going to be like—

Rich: That also requires expertise.

Paul: You can’t get humans out. You can’t. Like, this fantasy that you could somehow throw away all the developers, everything. Like, I don’t think it’s real. I think—

Rich: Five years?

Paul: Who cares? Like, the last mile is going to remain long. [laughter] No, because in five years, you’re going to have a whole class of new problems that we never thought of before that you still need humans for.

Rich: Yeah.

Paul: Because humans just fill the void. If you show us a vacuum, you’re like, “Oh my God, I can write Photoshop in five minutes with this thing?” Everybody’s going to go, “Great, but I need Photoshop that also sings and dances.”

Rich: 3D. Yeah.

Paul: Yeah, exactly. And be like, “Well, I got it to do that, too.” But then who are you—

Rich: Endless appetite.

Paul: We’re never done.

Rich: Yeah.

Paul: What do you think, they’re going to be, like, “Oh my God, that’s enough polar bears in my—” You know, they’re going to want 50,000 polar bears in the Coca Cola commercial.

Rich: I think this is helpful guidance and advice. And I think you’re right. I don’t know if developer is the right word. I don’t know, it’s, I think a new set of skills are going to materialize off of this stuff.

Paul: There is a—

Rich: Developers are going to, they’re going to appropriate it early because they’re closer to it, they understand it and whatnot.

Paul: We are in a class here, I think we’re like, we’re kind of experiencing this, too. And the term that, the term that always gets bandied about in consulting and so on becomes “solutions,” right? Like, “I deliver solutions.” A lot of times, that’s consultants, sometimes, that’s coders sometimes. But I think there is this sense of like, “I have a problem. Can you decompose it into something that a prompt could make us a little bit further along to solving the problem?”

Rich: Yeah.

Paul: “And then let’s see where we hit walls. And then we might have to do some—” Like, you’re delivering a solution, or AI-assisted solution. So I think—

Rich: I like that, because I think that’s something that demystifies it a little bit.

Paul: I mean your job is—

Rich: “I’ll solve problems for you.”

Paul: Because your job title can’t be like, “I figure it out.”

Rich: Yeah.

Paul: But that is what it is.

Rich: Yeah, yeah.

Paul: “My job is to know how these tools all work together so I can go figure it out.”

Rich: Do you think people who become experts in AI in five years will have to code?

Paul: I don’t know if that really matters. They’re going to have to think algorithmically. But they might have a—here’s what’s really nice about this technology, if we take all the drama away? It’s really helpful about syntax, and syntax is hard.

Rich: Yeah.

Paul: Like I, when I go, I’ll tell you, there are scripts that I write after 20-some years in Unix and there’s certain little corners of syntax I just can’t remember. I always gotta go look them up.

Rich: Totally.

Paul: And that’s never gonna be a problem for me again in my life.

Rich: Yeah.

Paul: And don’t take that away from me. That is the best thing ever. Because that is meaningless knowledge. It is, like, where the semicolon goes.

Rich: Totally.

Paul: There’s no value to society.

Rich: Yeah.

Paul: It’s what a computer needs. So I do think, so yeah, sure. People should still know how the things work.

Rich: Yeah.

Paul: Like, you should—if you choose not to, then you can’t complain when it doesn’t work.

Rich: Yeah.

Paul: Right?

Rich: Yeah.

Paul: But let me ask, I mean, here’s the same question. You think in five years from now, there’s a problem with your toilet? Often there is. You calling a plumber, or you calling a robot?

Rich: Calling a plumber.

Paul: I think that’s real, right? Maybe 20 years from now, robo-plumber comes?

Rich: Yeah.

Paul: We’re not there yet.

Rich: Yeah, but self-repairing toilets, Paul.

Paul: [laughing] Yeah, that’s…

Rich: Self-healing toilets. Let’s end it on that.

Paul: I just want you to imagine the liability insurance issue that is created by that idea.

Rich: Or good times.

Paul: Yeah, there are like 25 class-action toilet-related law firms in the future.

Rich: [laughing] Yeah.

Paul: Is what we’ve got. I skipped ahead. So everybody should have a great week. If you wanna get in touch. Hello@aboard.com. Go check out our website. There are some new and exciting things on it. We’re gonna talk about them more. But we’re [whispering] soft launching.

Rich: So soft. Have a wonderful week.

Paul: Gentle velvet launch. Goodbye, everybody.

[outro music]