The two types of engineering post Claude 4.5

[00:00:00]

Introduction: The New Reality of Coding

Fraser Kelton: Welcome back everybody. Welcome to the, the, what do we have? We have a weekly episode of Hallway Chat. Is that we've been hitting.

Nabeel Hyatt: not so much. We had a short, ~um,~ let's call it sabbatical break, but, ~but, uh, but~ we're back. Hey, everybody. I'm Nabeel.

Fraser Kelton: I'm Fraser and we can, we can get into it. It is like you, you've been down the rabbit hole of Claude Code and building hundreds of little

Building Camp: Multiplayer Knowledge Work

Fraser Kelton: apps.

Nabeel Hyatt: ~a, I have a any,~ you know, there's no room for any, any of the new Apple TV shows or anything like that. It's, it's~ it's all just, it's, it's, it's~ talking with founders or, uh,~ uh, or, or~ coding.

Fraser Kelton: You know what that actually Yeah, that's exactly right. Is the, a topic from now, like a year, year and a half ago, is your, your evenings were filled with, ~um,~ mid journey image creation. Like your hobby then was mid journey and now your hobby is building like robust multiplayer products that, I don't know, maybe you don't want to talk about the one right now, but the one that I most recently tried, what do we, we tried to scope it.

We thought pre-Claude it was a team of 10 for like seven or eight months.

Nabeel Hyatt: Yeah. ~Yeah.~ We can [00:01:00] talk about it.

Fraser Kelton: And you had cranked it out and, yeah, go ahead.

Nabeel Hyatt: like a couple of hard weeks. Let's be clear this, we should talk about long running agents and one shoting apps and this kind of sense that maybe with one prompt I can make a whole app.

I think a lot of that is still speculative and still a year away. It's a little bit like how we talked about, it's the summer of agents two years ago, and

Fraser Kelton: Yeah.

Nabeel Hyatt: is like the, the winter of agents was just a couple of months ago. I feel like long running agents are kind of in that camp.

Fraser Kelton: ~Yep.~

Nabeel Hyatt: so this was work, not a one shot app , I don't know how many times I prompted it, but it was an awful lot of back and forth. It's called Camp. ~Um,~ I built another call it like Claude Cowork, you know, how do you build a Claude code for information work. ~. Um,~ in this case, the kind of experiment was what does it feel like to do group work?

And you know this, 'cause I've complained about this since the beginning of ChatGPT, that there's no real product that has thought about what native multiplayer feels like in a, ~in a, in a, um, in a~ model LLM chat GPT world, the best we have is this world where you have [00:02:00] outputs that you share, which is a little

Fraser Kelton: Yep.

Nabeel Hyatt: you, you make a webpage and then I can share the webpage with somebody else.

That is, no one would say that's the same as you guys working together on a webpage.

Fraser Kelton: No. No.

Nabeel Hyatt: I work through Claude, I hit the publish button, and then I can share the published output. That's different than sharing your work. And so I know it's a hard problem. I've talked to researchers about it, but, if I'm gonna get smart about this, 'cause I think it's the future of work.

Why not? I don't know if anyone's ever gonna use this product, but by iterating on it. You get more native on the problem itself. And so

Fraser Kelton: Yep.

Nabeel Hyatt: it's similar to this idea that you see inside of our startups we work with right now, which is like, oh, instead of writing PRDs, why don't the designers just make the prototype and show you?

Why not try and experiment and kind of get some firsthand knowledge of what it feels like to be kind of native multiplayer working in small teams on something with an AI copilot .

Fraser Kelton: yep, yep. And like the, the amazing thing is it works, it's like way beyond the, ~the,~ like the level of a prototype, right? ~Like it is,~

Nabeel Hyatt: ~Oh~

Fraser Kelton: ~I, I, I,~

Nabeel Hyatt: ~robust than, than Claude Cowork to be, to be~

Fraser Kelton: ~yeah, yeah. Like,~

Nabeel Hyatt: ~Uh, with different scope. But, but also I've been working a little bit longer than they have.~

Fraser Kelton: I think you'd, it would be like a beta product. Here is a question for [00:03:00] you, ~um,~ because I'm trying to figure out how, how engineers and technical teams, ~uh,~ in this moment are navigating this. ~Um,~ I'm assuming that there's a, a reasonable amount of open source that you're building on top of, or like pulling in when, when and where it makes sense.

Open Source in the Age of Models

Fraser Kelton: ~Um,~ I was talking to a founder the other day and they said. ~Uh,~ they're, they're rejiggering their entire engineering org around this change. ~Um, and we, we can get to that as a whole other topic in a moment, but~ the thing that I'm curious about is he said that they're, they're very intentionally using a lot less open source, ~uh,~ packages.

They're instead saying, ~um,~ go find inspiration from this package for what we are trying to do here, but just take the p write the piece of code that I'm interested in. Yeah. Because I don't, I don't want to inherit all of the other gunk that's like, associated with it. Just like, just get the piece, get inspiration for the piece that's relevant for this.

Nabeel Hyatt: ~It~ Well in, in a sense that's an incredible, ~uh,~ comment because I think it's actually quite insightful. ~Um,~ it's in a way a, ~a a re~ a programmer [00:04:00] recasting of, of the conversation we're having in SaaS, right? Which is

Fraser Kelton: ~Right.~

Nabeel Hyatt: Hey, instead

Fraser Kelton: Hmm,

Nabeel Hyatt: Workday, I can take a handful of screenshots of Workday

Fraser Kelton: Right.

Nabeel Hyatt: remake it for myself internally.

Now, again, this is all slightly more hyperbolic than it really is every day, but it's the same version. It's like if software is free, why wouldn't I just get the version I want versus inheriting all the things I don't want? And, and that's the code

Fraser Kelton: Yeah.

Nabeel Hyatt: of that as well in my case.

~Um,~ look, the thing that you don't get with that, if you're just looking at outputs,

Fraser Kelton: ~Yep,~

Nabeel Hyatt: is all of the nuanced. Technical infrastructural decisions that frankly the models are still not particularly incredible at.

Fraser Kelton: yep.

Nabeel Hyatt: the

Fraser Kelton: Yep.

Nabeel Hyatt: I started with, I did start a, a little bit ways down. ~Um,~ it's a fork of an open source product, ~um,~ called Chorus, which is from the founders of Conductor, um, conductor, which is a, um, AI coding orchestration product. It's first iteration early days was an AI chat app. And so, and a very well-built ~way~ AI chat app. So that gave me a base and importantly that team has particularly good [00:05:00] taste and so I knew they had made good underwriting decisions on those things.

And so instead of getting the like median outcome AI slop version

Fraser Kelton: Yep.

Nabeel Hyatt: thinks I should build,

Fraser Kelton: ~Yep,~

Nabeel Hyatt: a tasteful. Open source implementation that got me started. So then I could just immediately, it can already talk to multiple models. It does good model

Fraser Kelton: ~yep. Yep.~

Nabeel Hyatt: it does good, ~um,~ subagent work. And then there I can be like, okay, now let's just think about how do we share context amongst users?

How do we make it multiplayer? And then I ripped out their whole backend, the first thing I did, and then inserted a new backend, ~um,~ from a company called Convex. ~Um,~ because they, I get a whole bunch of things for free. It's like a supabase competitor as a new database. So I can do multiplayer, ~uh,~ in real time easier.

And it's also type safe. So it's like, anyway, a bunch of technical things I get for free. So I, ~I, I~ don't know that rebuilds everything is my, my like personal experience. I think like all things, like the stuff that is average, that is commodity, you'll have a harder and harder time extracting any value from the market out of, because I can just rebuild it. [00:06:00] Or send a, send a few dollars Claudes way to rebuild it for me, ~um,~ is really what's happening. So yeah, net benefit to anthropic again. Um, and then there always will be best in class . It's just everybody's, the floor moves up to

Fraser Kelton: ~Yep, yep.~

Nabeel Hyatt: like, I also know Convex has made a bunch of database really, really interesting choices that are top of the line. I would love those outta the box.

Fraser Kelton: ~Yep.~

Nabeel Hyatt: so I, of course, I'm gonna use, use Convex. I think the deeper problem Frazier is how, ~uh,~ who am I asking and how am I figuring out which of these resources repose open source toolkits, which ones are best of breed, which ones should I use?

Fraser Kelton: ~Yep.~

Nabeel Hyatt: there's no Gartner for AI toolkits,

Fraser Kelton: ~Yeah.~

Nabeel Hyatt: ~oh,~ not that that was ever

Fraser Kelton: Well, in many, in many cases, the person making the decision is, is Claude code. Right? And, ~uh,~ I met a company, ~I don't know if we talked about this two or three weeks ago, that has been, you know, building a very popular e-commerce, uh, well, I met a company that's been~ building a very popular, ~um,~ open source package [00:07:00] for, for a domain.

And I'm not sure that, that they wouldn't want me to talk about it, but, ~um,~ they, they said, ~uh,~ they showed me a chart and there's like the, it just goes to the moon. Like it's been plotting around for a number of years. And it's like growing, ~growing, growing, growing, growing,~ growing. And then it goes to the moon since, since mid-December.

And, and I just, like, as soon as I looked at it, I just started laughing. 'cause I knew, I'm like,

Nabeel Hyatt: You knew

Fraser Kelton: Claude coat. Claude coat. Yeah.

Nabeel Hyatt: decided it likes you, and

Fraser Kelton: Yeah. And so, ~uh,~ there's a lot of people who are looking to vibe code the thing that they have like a great solution for, and it seems to be that this is the one that, that Opus currently cares about.

And, and it's crazy.

Nabeel Hyatt: your product agent friendly. ~It, it's one thing that~ I don't wanna get into the whole agent, SEO

Fraser Kelton: ~Yeah. Yeah, yeah.~

Nabeel Hyatt: SEO stuff. But, ~but, but ~I think there's actually two sets of this problems. One is, as a startup, you're building these things.

How do you make the thing that you are building more digestible by a agent and more, more understandable by an agent, which is often a different surface area than trying to make it understandable by a human or by a team of humans at a, at an org.

Fraser Kelton: ~Yep.~

Nabeel Hyatt: and [00:08:00] then the second thing is that, like, I actually don't, having gone through this a bunch of times, I don't know that any of the models are particularly great at picking, Providers of anything,

Fraser Kelton: ~Right.~

Nabeel Hyatt: which open source code base, which one should I use? Which framework?

Fraser Kelton: ~Yeah,~

Nabeel Hyatt: I don't have the knowledge of the space, it all sounds very credible. Oh yeah, sure.

It seems

Fraser Kelton: yeah, yeah,

Nabeel Hyatt: if, if I, but in areas where I have some knowledge of who the players are, I know all the startups 'cause we've met them all and I kind of understand the strengths and weaknesses and then I'm asking questions

Fraser Kelton: ~yeah,~

Nabeel Hyatt: it is, ~it is~ default. It is defaulting to default use

Fraser Kelton: ~yeah. Yep,~

Nabeel Hyatt: It's similar to like, it's not particularly good at like recommending ~ex~ exactly the right camera that I should buy and stuff

Fraser Kelton: ~Yep.~

Nabeel Hyatt: ~These~

Fraser Kelton: That's exactly where I was. Like, ~um,~ it also, yeah, it's going to give you the, ~the, the, the. The bland, not the bland, but like the,~ the safe choice and the safe choice, you know, the, ~the, the~ one that's most commonly referenced or, ~or, or~ looked to. ~Um,~ this is, ~uh,~ revealing a lot about myself right now in this moment to you Nabeel, you know, trust, ~trust and, and, and, and~ respect.

The Recommendation Problem: From Average to Expert

Fraser Kelton: Um, first of all, I've become a very heavy [00:09:00] Gemini user, which is like a big thing for me to, uh,

Nabeel Hyatt: Insane.

Fraser Kelton: well, listen, if you have the entirety of the internet on a hard drive like Google does, like of course they're, they're so much better with, with like working with, with URLs like webpage. ~Um, and~

Nabeel Hyatt: say, I think my, ~my, get my, my~ first response is not, look, I already know you're smarter than me in most directions anyway. So the first question is just like, what's your use case? ' cause you're,

Fraser Kelton: ~yeah,~

Nabeel Hyatt: it's probably all about like, you're not trying to write

Fraser Kelton: no.

Nabeel Hyatt: apps with Gemini.

Fraser Kelton: No, no, no, no. This is a ~to~ total tangent that I bring up because, ~uh,~ it's relevant to the example of picking the, the package or the, like, the right, the right software that, that Claude coach should use is, ~um,~ this, ~I I,~ I've, I've become very curious about, the various qualities and characteristics of different fabric and, ~and, uh,~ like materials.

And, and then so if you were to ask

Nabeel Hyatt: you knitting now? Is it,

Fraser Kelton: No, I just,

Nabeel Hyatt: Hobby Fraser?

Fraser Kelton: Wearing, I am wearing clothes.

Nabeel Hyatt: It's fair, fair.

Fraser Kelton: I don't know, like curious minds go to curious places and then you're, you wanna [00:10:00] learn about these things.

Nabeel Hyatt: Can, I can opine about backpacks for way too long for

Fraser Kelton: That's exactly, so think, think about like the backpack episode to, to ~me and like, uh, long sleeve t-shirts.~

~Um, right. And if you were to ask, if you were to ask the base model what like, long sleeve t-shirt should I use, it's going to give you like, I don't know, like the Gap or J crew, um, or, or whatever. And then like you, you have to get good at nudging and nudging, nudging, nudging, nudging, nudging, nudging, and giving it like what you actually care about and what you want to like qualify for.~

~Um, and, and then the, the, the, I don't know, maybe, listen, this is the great thing about having a podcast that, that we don't really care about total audience size. So let me just meander for a moment.~ The reason, the reason that Gemini's awesome is. You can drop in a URL and, ~and, and~ it just interacts with that page far better than any of the other models.

And so I'm like, ~I,~ I bought this shirt in this size, and you can get the dimensions on that page and it fits like a little bit in this way. Gemini is now basically like my, my bespoke tailor, ~um,~ because I'm like, I didn't like this, this thing or that thing or this other thing. ~Um,~ and then it will go and it'll find me like the best, the best thing for, for fit and finish based on the little bit of feedback that I give it.

Nabeel Hyatt: And, and your read of this, ~um,~ is that this is not because it has a better internal state of something like dimensionality of things in 3D space. So it understands your body or it's better smarter logic. It is literally that it is, better at understanding the webpage and, and

Fraser Kelton: Yeah.

Nabeel Hyatt: ~inside of that~

Fraser Kelton: ~Yep. That's, um, s strongly I feel that way. And, and it's like, it's like super bizarre. And then you're like, well what's the, anyway, the total tangent and we can, we can drop this in post, but, uh, it is,~

Nabeel Hyatt: ~ask a que I have to ask a very important follow up question~

Fraser Kelton: ~yep.~

Nabeel Hyatt: ~What long sleeve t-shirt should I buy? Frazier.~

Fraser Kelton: ~Well that's, that's like saying what piece of what backpack should I buy? Uh, you, you probably have an answer actually for that, but the, a long sleeve t-shirt you have to think about are what season where, like what geographic region do you, what, what are you prioritizing for?~

Nabeel Hyatt: ~live in San Francisco.~

Fraser Kelton: ~Yeah.~

Nabeel Hyatt: ~you know me like it's a little chilly, so it shouldn't be, uh, it should be thicker. Uh, what, what do you got on there, my friend? You got Toronto Weather? Uh, long sleeve T-shirt.~

Fraser Kelton: ~Uh,~ the Outlier [00:11:00] brand, which is a company from New York City, makes a long sleeve Marino shirt. And then, and then you like fall down the rabbit hole of curiosity around Marino and like what a micron is and like, yeah. ~Uh, where does the, this scratch come from Merino and how do you reduce the scratch? And like, uh, you end up, you end up so deep, you're like, ~.

But then you're like, okay, so if you get down to like 15 microns, you're learning about like the longevity of the fabric and Gemini's like my great t on this. And it's like, if you have a dog, I'm like, I do have a dog. It's like, you don't want to go to Micron of like 14. ~Uh, so yes, go to outlier.nyc, uh, super small brand, um, based in New York and uh, get the long sleeve Moreno t-shirt.~

Nabeel Hyatt: One, it purchase. Done. ~two, two, i I.~ This, this brings up something that I think is, is quite interesting, ~um, ~the first at almost everything is pretty average. It

Fraser Kelton: Yep.

Nabeel Hyatt: the, the kind of first pass of all things is, is a c unfortunately, I think most people are trained to hit the first google result , I

Fraser Kelton: Yep,

Nabeel Hyatt: a quick first pass.

And

Fraser Kelton: ~yep.~

Nabeel Hyatt: get average. And then the interesting nudge that AI models provide is the ability to get the [00:12:00] exact answer you want. If you put the time and energy in what we have still, what, what Google tried to solve, Reddit tried to solve, like what we went through in the arc of the early internet was we started from something that looked like the Yahoo homepage for the internet, which was the net average, right?

I go to the homepage of Yahoo and I click music and I get the five music websites that exists on the internet.

Fraser Kelton: ~Yep.~

Nabeel Hyatt: then over time, instead of talking about this as search or something else, let's just talk about it, what it means to a user. We found a better job of surfacing niches

Fraser Kelton: ~Yep,~

Nabeel Hyatt: of finding the best of the thing.

And so the idea at the end is that you're in the the Reddit thread with the other total geeks who

Fraser Kelton: yep. Yep.

Nabeel Hyatt: the whole wisdom of the crowd stuff. This is long tail stuff. This is the whole like 6, 7, 8 years into the internet. It was kind of the second wave Web 2.0 was really about not finding the average, but finding that niche.

It's funny that we got down this rabbit hole, which was completely unplanned. What we don't have a product surface area for, and maybe it's a new consumer hit product that'll come out [00:13:00] for ai, that's a AI native that just doesn't exist today.

Maybe there's an enterprise version, maybe there's a coding version for finding which database you should use. But right now I have not seen a single, a single user interface or affordance, which is about rapidly taking you from, I'm interested in a long sleeve shirt, or I'm interested in database, or blah, blah, blah, and making you an expert. To make a choice

I mean, again, all of them have capability. This is all stuff that is in the latent space of all of these models and has been for quite some time .

Fraser Kelton: You know, I, I think my, my generous take when I squint and look at the, the health release from Chat GPT was that perhaps this was their horizontal products attempt to try to do this within a vertical that is clearly important. ~Um,~

Nabeel Hyatt: Oh, well said.

Fraser Kelton: yeah. And, and, and so like, I,

Nabeel Hyatt: rash mean? Is

Fraser Kelton: yeah.

Nabeel Hyatt: a, trying to meet me an expert quickly in

Fraser Kelton: Yep. And, and it also, but like, it also is like, to go back to your example of like the Excel or like the, when you have a really horizontal [00:14:00] product, you can have a, like a little thing that's like, here are the templates that you can use that help you like educate the user as to what they can do.

And so I can imagine that like a little thing that says health that is, you turn to this, this like corner of this broad horizontal app whenever you want and they can make editorial decisions for you within that to help you like get along. And so like I could, I could get my head around that.

Nabeel Hyatt: Yeah. What, what I'm sure it isn't, is. It's not that you suddenly work out what the five dropdowns are for long sleeve shirts, right?

Fraser Kelton: Yeah.

Nabeel Hyatt: 2.0 way of solving it is

Fraser Kelton: That's it.

Nabeel Hyatt: have, you know, just like you go, um, to an e-commerce website and there's five dropdowns.

Fraser Kelton: Yep.

Nabeel Hyatt: reminds me of, there's an adage about, about real estate websites, which is that, real estate websites don't show dropdowns about a home based on what you want to buy they show you what is easily measured and listed in a UI , and that might be number of bedrooms, amount of square footage or so on and so forth.

And then inevitably what that means is that the perfect house for you, as, as we know when [00:15:00] we talk to a real estate agent. Involves the things that are cannot be selected on the left hand

Fraser Kelton: Yeah. Yeah.

Nabeel Hyatt: It is, it is. No, but I actually really wanted one that was like within four minutes walk to a dog park. And um, and the amount of light that comes in in the morning. 'cause I like to have coffee over, like whatever it is for you in your particular life,

Fraser Kelton: Yep,

Nabeel Hyatt: your job is to go out and find the thing that's not on the, on the list

Fraser Kelton: yep.

Nabeel Hyatt: then the market has underpriced because it's not on the list.

Fraser Kelton: Yeah. Exactly. . And so,

Why Gemini Works for Personalization

Fraser Kelton: the reason that I ended up on Gemini originally was not necessarily because of the pages, like its ability to handle pages. That was a, a joy that I then discovered after the fact. It was, um, this is gonna sound funny, was because of memory.

Um, I, I find that both Claude and ChatGPT, they crowbar memory in as if it's like, you know, ~uh,~ we've all come across people that have a personal CRM and they use it a little bit too blatantly, and you're like, oh, that feel like what's going on here? ~Like, um,~ 'cause I was trying, I,

Nabeel Hyatt: Charlie? Was the

Fraser Kelton: yeah.

Nabeel Hyatt: great. Next, last week?

Fraser Kelton: [00:16:00] Yeah.

Well last week it's like, was the dentist visit, did that turn out well? And you're like, dentist visit, that was nine months ago. What's going on? ~This is, so, um,~ I was trying to ask one of those two, ~um,~ about like fabric and what I should buy and then, then it came back and is like, this would be perfect because you're a VC and it will give you the sleek look in your next partner meeting.

And I was like, oh man, come on, come on.

Nabeel Hyatt: ~that's~

Fraser Kelton: ~And so,~

Nabeel Hyatt: That's called overfitting.

Fraser Kelton: well. Yeah, like, it's just, it's just actually like a terrible user experience. ~Um,~ and I'm like, ~co~ like this is where I want amnesia. ~Like, um, ~like, first of all, you have a bad sense of the taste of Spark overall, like you've stereotyped about this VC thing.

He get offered a, at first it was like, you should go to Brooks Brothers. I'm like, what? ~Uh, so yeah. Um,~ and then to, to tie this up and then I, we really have to get back Claude code. ~Um, ~

~um,~ I bought the shirt, the shirt showed up. The shirt felt amazing, like it felt exactly what I wanted. And I tried on the shirt and I'd, I was so angry I did what you would do. In that moment, I took like [00:17:00] three photos and I sent it back to Gemini in that thread. And I'm like, you goofball, like, look at, you told me to buy the medium and look at this, and, and like, it was like, I looked like a clown.

Maybe I still do. But, ~um,~ it very patiently said, it said, aha. Based on your measurements of liking this size shirt from this place previously, ~um,~ I sized you to this level because if it shrinks by that amount, you'll then be in this, this line for the, for the brand.

And I honestly was like, what's going on? I popped it into the, into the washing machine. ~Um,~ put it into the dryer because it said to put it in the dryer and I never would've put it into the dryer. And it fits so perfectly.

Trust re, trust reestablished.

Nabeel Hyatt: we are, closing off this topic because we started out with, with me complaining that these things don't know how to recommend the best thing. But obviously we are not that far away

Fraser Kelton: We're not far.

Nabeel Hyatt: ~Where,~

Fraser Kelton: ~far.~

Nabeel Hyatt: where if, ~if, if it's, if~ it's wrong, it's probably user error.[00:18:00]

Fraser Kelton: Yeah.

Nabeel Hyatt: I don't understand. It's making better decisions for my, it's like my wife. She's probably making better decisions for me than I am for myself. So I should just listen.

Fraser Kelton: It no way we're so far, like if you think the first thing, like, I'm not even joking. The first recommendation because it was like trying to like make an assumption about where I worked, ~uh,~ at least the career that I was in, and then like infer what I should buy. If I had just bought that, it would've been like, it would, I would've, ~yeah,~

Nabeel Hyatt: ~path. Frazier.~

Fraser Kelton: ~we're on the.~

Nabeel Hyatt: I'm optimistically projecting the future. We're not there yet.

Fraser Kelton: We can be, we can be excited by it. So, ~um,~ let me, ~let me, let me~ pull us back to Claude Code. ~Um,~

Submit a Prompt: Conductor's Product Innovation

Fraser Kelton: so Charlie at Conductor tweeted out this thing where he said, ~uh,~ you no longer have to submit feature requests. ~Instead~ just prompt the feature that you'd like to see if it, we will review it. And if it seems reasonable, we will, we will have Claude Code run that, and then we'll, we'll merge it.

Nabeel Hyatt: I, I, I love this feature. It makes sense that it went, uh, immediately viral and, and like crazy. For those who don't know, conductors is an [00:19:00] orchestration, um, for, for Claude Code and now for Codex where you can run a bunch of them in parallel, they built a service on top of it that that cursor has now copied and others are kind of riffing on and so on and so forth.

Um, but as, as usual for the person who invents like you, you got there from first principles, you're probably get there again and again, it's, it's another good example of, of what they've done. So yeah, the idea is like. It's funny that we go back to the web. It's, you know, I was just thinking about the Web 2.0 thing.

One of the Web 2.0 promises was this idea of the long tail. All of my customers are my community. Now we're in direct communication. It's no longer like NBC where I just like broadcast something. It's like, no, we're all interacting together this is, this is again, like if we echo that, it is, it is the 10 x version of that.

It, I'm not just getting feedback from my customers. My customers are actively writing the cloud code prompts that I will run that will write my software. Um, and then, and it's not that far away, obviously, what he's leaning towards. I don't have to [00:20:00] ask him or talk to him about that. What, that obviously the stage after that you can see is like, oh, why don't? just open up Microsoft Excel

Fraser Kelton: Yeah.

Nabeel Hyatt: or

Fraser Kelton: Yep,

Nabeel Hyatt: and I prompt my software. It gives me back something that lets me alter the surface area of this software the way I need in

Fraser Kelton: yep.

Nabeel Hyatt: of sandbox environment.

Fraser Kelton: Yep.

Nabeel Hyatt: I can test it, I can run it, we can run it internally. And then by the way, then the team at Microsoft or Figma or whoever in the future is looking at all the things that their customers are already making out in the wild and then sucking some of those in to the primary repo and pushing it to everybody.

It is,

Fraser Kelton: yep. Yep,

Nabeel Hyatt: in a way

Fraser Kelton: yep.

Nabeel Hyatt: open source community, but it is just giving every single customer on the planet who uses software, that kind of interaction with their, with, with the software they use every single day, which is incredibly powerful.

Fraser Kelton: Yeah, that's such a great analogy, right? Is it has happened for decades in the open source community. [00:21:00] You, you want to fix something that's bugging you, you want to fix it, you can just do it. Um, and now a feature's bugging you or the absence of a feature, uh, you just, you submit the prompt. ~It it is, yeah.~

~It's awesome. The.~

Nabeel Hyatt: ~I have submitted a prompt to, to, to conductor and, and I, I don't, I don't see it yet, but, but, uh, but, but no, I think he's pushed out one a day or one every other~

Fraser Kelton: ~Yeah, I've seen that he, I mean, he is doing the, the right thing, right? Is that he's now, he's now making sure that those prompts are making it into the product. The,~

Nabeel Hyatt: you can imagine from an engineering standpoint now, like he, ~I think he, that he~ tweeted back out, like he's gotten hundreds of these. You can imagine there's just now a team, ~if you build a team at~ like, there's a pod of people whose job it is to just look at what the customers are, are running and then just run them against Claude, you know? And then by the way, when ~it's, when~ Claude is cheap enough, it just should. Every prompt you submit should auto create a git tree, create a sandbox, ~auto,~ create a PRD from it, make a mockup of it, make a running, ~uh,~ example of it in software, and then it's just somebody on the product team booting up every morning and clicking through prototypes, ~uh,~ with a little brief on what the prototype is until they find stuff they like and being like, yep, push to ~lives, push to~ live. ~What? Um, what~

Fraser Kelton: it's

Two types of Engineering: Automatic vs Iterative

Fraser Kelton: not, ~I. ~It is that that's just gonna be a stepping stone. 'cause these things are just going to merge into prod in real time. Like, uh, I met a group who's building the modern dev [00:22:00] environment in the cloud where like, you're like, you look at it and you're like, well, it's not too different from what ha exists today.

And they're like, yeah, but our roadmap is like, these things are just gonna merge automatically into prod as they get written. And like users are gonna get in real time is you'll, you'll just have like a split test where it will, you don't even need the product manager yet looking at it and trying to decide if it's good or not.

They'll just, you'll automatically deploy to end percent. If it looks good, it will continue to rise automatically. Uh, if it doesn't look good, they'll roll it back. ~It will, everything looks,~ it's just gonna be such a crazy time.

Nabeel Hyatt: Look, ~look,~ I think this, ~this is, this~ is true. And this leads to the like, oh, ~well~ we ever have engineers, do we even need engineers anymore? And what goes away? And, ~and, and~ I wanna push something back to you because, ~you know, uh, I, I, I, ~I think all of that is true, and it's very possible that for a bunch of engineering work, we are ostensibly getting rid of the engineer. Um, and, and in fact it's just running itself. It's doing ab split testing. There's a QA tester, which is another engineer and so on and so forth. But I, I don't know that that means the engineers out of the [00:23:00] equation because what it feels to me. That is happening is this split, uh, in two types of engineering work that maybe got slightly muddled before and then now the fissures between these two types of ways that we are building becomes more obvious. And the first is what we are talking about right now, which probably I think Cognition and Devon were the first example of this AI engineer automation agent.

Which the idea is I give it a task and it might run for multiple days, it can QA test and do the whole thing and it just kind of does everything for you. And there's a bunch of examples today. Um, I think Cursor recently launched their attempt at long run agents. They're trying to get it to like build a browser from scratch and

Fraser Kelton: Yeah.

Nabeel Hyatt: kind of things. But then the other side of what I'm doing, say in, in, in conductor and Claude right now is, feels like almost like playing a video game.

It's like, it's constant. I like I'm plate spinning. It's, it's constant feedback cycles with the agent where it comes back to me every three [00:24:00] minutes and, and then I, and I look at some prototype of something and I'm like, not quite like that. Do it a little bit like this. I send it off and do it to the world.

And then I context switch and context switch and context switch. 'cause I'm working on five different branches on six different projects and then I go take a nap afterwards. But, um, but these two things, although we book all these things coding,

Fraser Kelton: Right?

Nabeel Hyatt: don't they feel

Fraser Kelton: Mm-hmm.

Nabeel Hyatt: to you? Like, do, do you think those are even the same model in the end?

Fraser Kelton: Um, say model, we model is now such an overused term, say model as in like the actual AI model that's working on it.

Nabeel Hyatt: You take that however you want for,

Fraser Kelton: Um,

Nabeel Hyatt: you think these two things are solved by the same sets of companies?

Fraser Kelton: yeah. Yes. I think ultimately, there's room for, there's room for it to be different, but I think that, uh, no, maybe not, maybe not.

I guess it, my answer depends on how AGI pilled and, and large model maximalist you are. And so if, if at one point, uh, Opus, uh, 6.4 or whatever is just [00:25:00] doing everything and there is no harness really on it and you're not using tools or like, um, or other models and, and like, sure, then, then like that's where I would come back.

That's the split.

I, I think in both cases you are going to have people who are engineering, , overseeing both of those. And in fact, it might be the same engineer is doing both. There's, you know, you're kicking it off and then there's one that's just doing, as you said, plate spinning with a lot of precision.

Little edits.

Nabeel Hyatt: Yeah. Yeah. This, to me feels like the, you know, like, like ideation versus production

Fraser Kelton: yeah.

Nabeel Hyatt: design where there's points where you're just like, look, I'm just arranging boxes. I'm have to lay out a magazine and laying out the magazine is like, just a lot of work.

And it is a lot of work. But, uh, it, there's the, the style has been set up, you know, it's a lot of micro decisions, but no really big macro creative

Fraser Kelton: Right?

Nabeel Hyatt: there's the like. Hey, what do we want our magazine to look like? Like, and that, that is very hard to just ask a model to get [00:26:00] there because there's this

Fraser Kelton: Yep.

Nabeel Hyatt: and iterative cycle of seeing it and then making calls and tweaking in the creative process, that does , move back into kind of humans at some point.

Fraser Kelton: yeah, if you, if I go back to like my, my t-shirt selection, uh, example for a second on my, I made a lot of decisions and put in a lot of input and, and like volleyed with the model to help figure out what is the right thing for me. And then if we talk about like the architectural decisions or the trade-offs that happen when picking even like the software package that you might want to use is there will be something that is voleying with somebody who is, um, you know, again, managing those trade-offs for the, first of all, the engineering organization, but then overall the entire organization, right.

We've all been there where you're like, Hey, if we do this, um, this is what happens to retention, but we reduce our legal exposure by why, on which access. Like these are hard, hard decisions that, that you still need somebody who's an engineer in a sense, helping to wade [00:27:00] through and, uh, I think there's gonna be engineers for a very long time. I think the, the actual nature of their job has just transformed dramatically since December.

Nabeel Hyatt: Maybe it's not just too long running agents versus short-term inflow work. I, I like, I like to think about coding right now. It feels to me like when opening up any of these products, you're basically doing. Four tasks. First you Plan. Then do the work, then Tests and then learn or extract insights

Fraser Kelton: Right.

Nabeel Hyatt: take takeaways and, and so, you know, the plan step was the first thing that got built in.

As in clawed, there's even a button for it. I'm in plan mode, so let's make sure that we, we prep properly before we do the thing. Then we have to do the thing, then we have to do evals and tests and does it work. And I feel like that's a huge area of, um, opportunity in 2026. I suspect that there will be many startups that will finally get to the points of automated testing and benchmarking and so and so, so forth.

'cause it's the, that's the bottleneck. Like now that

Fraser Kelton: Yep.

Nabeel Hyatt: code like crazy and everybody [00:28:00] can write, look at the question is like, is the code any good? Is it running well? Like blah, blah, blah. And, and so I think tests and evals become that next one. And then extracting learning from it, which

Fraser Kelton: Yep.

Nabeel Hyatt: the same thing intellectually as like, please summarize this project in A

Fraser Kelton: Right,

Nabeel Hyatt: a

Fraser Kelton: right.

Nabeel Hyatt: Like, it's very different to say what unique thing did we learn versus all the other things we've done before. And this is the problem that a lot of the context graphs have,

Fraser Kelton: Yep.

Nabeel Hyatt: to grab context about your life is they're not asking questions, deeper questions like that. Um, anyway, the question with all of agentic workflows is how many of them for the task that is being done in your area of knowledge work, whether it's law or code or whatever, is. work or takes true human creativity. Like which

The

and which ones do you need some message above?

And the only thing that's really happened in the last six months is the do step out of the four

Fraser Kelton: Right?

Nabeel Hyatt: become

Fraser Kelton: Yep.

Nabeel Hyatt: the average that pops outta the model is so

good it blows our mind. The [00:29:00] plan is still not, which is why you iterate some of the beginning, uh, tests and learn is like, of janky.

It's okay. And I think the insights step at the end where we figure out what we need to learn from this project to get smarter for the next project we do is almost nascent. Like, it's,

Fraser Kelton: yep,

Nabeel Hyatt: it's horrible.

Fraser Kelton: ~yep, yep.~ So, ~um, I, I.~ I agree. ~The, so I'm, um, hiring cafe, then we can take that out. Uh,~ I made an investment, ~uh,~ recently in a like seed stage, really early, just a very small team. And I was chatting with the founder, CEO today, and he, he said that basically since December, um, they, they've realized that as they think about growing the engineering org, they just are gonna think about it in a dramatically different way.

Because already today, they mostly spend their time now writing the benchmarks and the tests. Um, there's a little bit of planning that happens, um, but then otherwise this like just let's, let's crank out those benchmarks and those tests as quickly as we can and they're eating up their roadmap because this thing's just going as hungry as they can.

. I bring that up. I think it's an interesting [00:30:00] question is, , how to navigate the change in this moment. Um, you know, certainly, certainly in December, uh, Opus four five comes out cloud code, like it's just a phase transition for engineering. What do you think founders that we work with, founders that are listening, like people who oversee engineering teams, like what do you think the advice is for, on how to navigate this?

Nabeel Hyatt: I had a conversation with a senior executive. At Meta just last night. And, um, I started with a simple question, which was I, my son is, is graduating from college in a year. He is about to come into senior year. , He's trying to figure out whether he should double major in data science, computer science, or maybe data science and something crazy like philosophy or whatever. And, and the question was just like, he certainly can code, he's been coding forever.

He's been working in AI stuff forever. Um, uh, does the computer science major matter? Like, does, does that signal anything to anybody? And the read from this person and, and I was asking somebody at Meta [00:31:00] for good reason because I know at a startup what would happen. But the question is like, at a technology organization that is, that is fairly mature, trying to be aggressive in this market, you know, but, but it also has a big org to try

Fraser Kelton: Yep.

Nabeel Hyatt: like

Fraser Kelton: Yep.

Nabeel Hyatt: of interns and blah, blah. And the exec basically was like, oh, no, no, no. We, we wouldn't care about a comp side degree anymore at all. all

Fraser Kelton: Hmm.

Nabeel Hyatt: now is fluency in ai. Like what we would really understand ostensibly is could they write a good eval?

Fraser Kelton: Right, right,

Nabeel Hyatt: good PRD? Did they know how to instruct a model properly and understand, and you do need to be technical and you so compute you, you would need to know computer science to an extent, to, to,

Fraser Kelton: right.

Nabeel Hyatt: the things that they're trying to do, and especially inside of their infrastructure and architecture, which is complicated, but it comps side degree doesn't give us any indication of that whatsoever.

It's, it's not that it's bad signal, it's just, just like no signal. It doesn't answer any question for us. And, um, and so I'd say if meta is at that stage, like I think, I think the, the, the, at the startup stage, that's the first [00:32:00] anecdote is like you should be evaluating based on your ability to instruct and articulate a model on things that are hard to talk about.

~Um,~ 'cause

Fraser Kelton: ~Yep.~

Nabeel Hyatt: that's the

Fraser Kelton: ~Yep.~

Nabeel Hyatt: the work.

Rethinking Team Structure: From Six to Two

Nabeel Hyatt: ~Um,~ the second thing is, I think I'm convinced, the challenge I've had for several teams in the last month is ~I think I'm convinced~ all of my life in, ~in~ startups, have felt like to eight people, is the right size of pod. Like the mixture of number of engineers maybe, ~you know,~ pm , designer, , blah, blah. Like, for 50 years of software company building that is a great pod. think the right pod after Claude 4.5 is two

Fraser Kelton: Yeah.

Nabeel Hyatt: and it's only two so that you can bounce the idea off of somebody else.

Fraser Kelton: Right, right.

Nabeel Hyatt: so if you, ~it,~ so the first bit of just like if you look across your org and there's any pod that is larger than two, ask yourself why. Look, there's different companies with different architectures and different coding frameworks with different amount of AI that's going on, like everybody's in a different spot. But I, I found that [00:33:00] that very simple challenge seems to have unlocked like 12 other good questions to ask that then make you think again about what is the right process that we're going through as an

Fraser Kelton: Yep.

Nabeel Hyatt: the bottoms up view versus top down view of how we get stuff done today.

Fraser Kelton: I think if you're not dramatically rethinking your eng org structure, you're so naive. I think you're so naive. And then, and then I think if you're not rethinking the way that you structure your teams, your org, your processes, like if you're literally not going back to first principles on all of those, to the point that you just said, like, think how crazy that would've been, you said like six to eight and now you're saying two

Nabeel Hyatt: ~Yeah.~

Fraser Kelton: like that.

~That's a cr.~

Nabeel Hyatt: by the way. Six to eight for the last 50 some years across almost every size of

Fraser Kelton: Yeah,

Nabeel Hyatt: in every single industry, ~you know?~ Yeah.

Fraser Kelton: yeah. And now you're saying two and you have a good reason for it. ~Um,~ if ~you are. You're not, if~ you're not rethinking everything through like that level of change, ~uh,~ I think you're, ~you're, you're~ shortsighted. And, and [00:34:00] then, and then I think the question is like, how does somebody get to a point where they're not thinking about that?

And my only observation is if you oversee an engineering team or a technical team and you haven't been using these tools like very deeply over the past three months, I think that, ~that, like, that, ~like I think that you're totally missing that this is, this is an inflection for, as you said, that we have not seen in, in multiple decades.

Nabeel Hyatt: Question for you ? What do you do with a CEO? Who says to you, I think we're doing a good job being an AI first company. ~Now~ we're trying to transform, I'm sure literally every CEO tells every board member we're doing a good job trying to transform in the age of ai.

We've done x and y thing. 20% more prs are written by a, like whatever justification for AI success they said. What is your advice then about whether, how do they measure whether they're actually doing a great job or a bad job and their VP of engineering is telling them, ~um,~ we're transforming quickly.

~Uh,~ you know, but they're [00:35:00] not in it. Like, ~what, what, what do you,~ how do they know?

Can you AI transform without living it yourself?

Fraser Kelton: I think ~my,~ that is a serious question that requires real reflection. My from the hip answer is, ~um,~ if you're not seeing levels of rethinking. The org structure, the team structure and IC responsibilities, and therefore like incentives and, and how they get rewarded

if you don't see that level of rethinking, I think that ~that,~ that you clearly have an organization being led by somebody who doesn't understand the change that we're living through. , I think it's totally unacceptable, like, the number of prs that we have is this, or like it's gone up by 20%.

~Um,~ I think that, ~that,~ that is all so inconsequential to, to what you should actually be having in this moment in time. That if you are, if you're not legitimately planning at, I think like all three of those, ~uh,~ altitudes, entire org structure, team structure, and then individual contributor like, ~uh, uh,~ incentives. I think you're, you're like totally missing it.

Nabeel Hyatt: You know, we were at a dinner, ~uh,~ last week together and, ~and, and~ one of the [00:36:00] CEOs said, ~he,~ the phrase was, ~um, you know,~ I am, I'm the best engineer on the team, because I'm better at Claude Code. . It's a 30, 40 person company, and I'm doing it part-time while I'm also CEO eing. And by the way, I hadn't written code in years and years

Fraser Kelton: Yeah.

Nabeel Hyatt: until this age. And so I am holding my, my team to account how can I possibly be the most productive engineer on the team part-time as a rusty old programmer.

Fraser Kelton: ~Yep.~

Nabeel Hyatt: what does that mean for how you all should be working? And I know some companies that I feel really good about their AI productivity and the way that they work. I don't know that I can think of a company where the CEO isn't leading by example on that.

Fraser Kelton: Yeah. ~Yeah.~ Like I, I'm not surprised that Shopify has been very aggressive on this when you look at how active Toby is. ~Um,~ yeah, and I think, ~I think~ in a lot of these things you're, a lot of the empathy always comes. Like, it's not surprising. The empathy comes by like using and exploring and, and like [00:37:00] getting curious and trying.

~Um,~ I think this is a case where. The change is so profound that if you haven't done that, it's hard, ~it's hard~ to internalize it. I mean, you can read these blog posts from like that OpenAI team where they shipped an entire product, I don't know, however many, maybe it was a million lines of code where no engineer actually touched code.

And you can read that the, about the startup where they, they're again, like, they're not even looking at the code, I think is, is like their goal.

Nabeel Hyatt: Yeah.

Fraser Kelton: you can read those things. And I think it's like, ~uh,~ it sounds like we're going through a really big amount of change, but I think if you're, if you're Toby or, ~uh,~ your CEO friend, ~uh,~ and you're actually like experiencing this day in and day out, I, I just think it's different.

Nabeel Hyatt: Well, I think it's also hard to hold somebody to account when you're just abstractly going, well, what if you just went faster? ~Versus, versus,~ I mean, look, sometimes there's just moments in the world where their founder has to lead by example. ~Um,~ hope that isn't across the board, but it's certainly, it's funny, I have ~a a,~ a friend of mine who's a seed stage investor who we were on a walk ~a a~ [00:38:00] little while ago, and, ~uh,~ he was saying his investment philosophy right now is he asks what the per engineer budget is.

Fraser Kelton: ~Right, right.~

Nabeel Hyatt: and if it's in the top 5%, ~uh,~ then it's probably a good investment. And like, that's the simple, like they're gonna in whatever market they're in, if they're spending ~ano~ a huge amount of tokens per engineer. But actually the proto before that, that we're getting to, which might be just like, tell me what the token spend the CEO is. Um,

Fraser Kelton: yeah, yeah. Yeah.

Nabeel Hyatt: and, ~uh,~ and whether they're the person spending the most tokens on the team. ~Um,~ and for an early stage startup where you're really trying to, especially or, ~or, or~ someone that was started in ~2001, I'm sorry,~ 2021, ~it's, yeah. Or you're somebody who started in 2021~ who's trying to, ~you know, lord forbid, you know,~ transform yourself into an AI first company and all the rest of that stuff.

I, ~I, I~ don't know how you push that down to the team to go figure out for you while you, while you hold up a PowerPoint and say, ~um,~ what's our efficiency curve look like?

Fraser Kelton: Proba probably the company that I work with that's been most, ~uh,~ aggressive in embracing this is illicit. And it, it's not surprising that Andreas, the CEO is, is like probably the ~per~ [00:39:00] person that I know who, ~you know,~ is furthest out in, ~in, in~ using these technologies. And, and like he, he basically has changed again, like the way that they're structuring engineering, the workflows, the processes, not just for engineering, but ~because of like,~ for the entire company, because of what this enables.

And, ~um,~ yeah. He was giving me, this is just a funny anecdote, ~like to, to~ the sense with which he's, ~um,~ immersing himself in it. He, he was walking me through a slide. Deck and I, I was unfamiliar with the service that he was using. I'm like, this is not Google Slides. Like, I, Andreas, what are you using? And he just is like, his eyes lit up and he, like, his eyebrows bounced a couple of times and he goes, you know, and he, he didn't like the features that he was given within Google Slides.

And so he kicked off Claude code for a weekend and, and it just came with a, ~a a,~ like he now has a bespoke slide. He, not that it took him a lot of time.

Nabeel Hyatt: he built his own Google Slides. Yeah.

Fraser Kelton: Yeah, yeah. With the features that he actually cares about and like the look that he cares about. And it's

Nabeel Hyatt: ~Yeah.~

Fraser Kelton: I'm [00:40:00] not surprised that if you're doing that and then you wake up on Monday and you get that software that you're not going into the office and being like, Hey guys, I think we do need to rethink the way that we're, we're structuring engineering here.

Nabeel Hyatt: Yeah, I agree. ~agree. Um,~

Data Moats and the MCP Shift

Fraser Kelton: ~Uh,~ total different topic.

Nabeel Hyatt: who's listening to this should go write some software, by the way, ~that that's the,~

Fraser Kelton: ~Yeah.~

Nabeel Hyatt: there's one thing you should take away, go, ~go, go~ make stuff.

Fraser Kelton: ~The, the, the, yeah, it's crazy. Um,~ I was using, ~uh, I was using~ granola in a way that I didn't anticipate that I was going to be using it. ~Uh,~ I was using it through a third party service because I, I used the granola MCP, and I'm going to tell you, it was exceptionally delightful. I was running into a meeting and I, I asked granola, I'm like, give me the high notes that I need to know for, for this meeting based on, ~you know,~ the conversation I had with this founder six months ago.

And it, like, came in, came into SMS perfectly with that. ~Um,~ the, the reason that I bring this up, Nabeel is like, even, even only a handful of ~years ago, handful of~ months ago, I don't know what type of timeline we're on. This would've been like business 1, 0, 1 of the exact thing not to [00:41:00] do.

Nabeel Hyatt: ~Mm-hmm.~ Yeah. ~Yeah. The,~

Fraser Kelton: ~and, yeah.~

Nabeel Hyatt: the version of, of granola that you pitched two years ago is. Hey, we have this unique access to context. We have all of your transcriptions from all your meetings if we win. And by the way, that gives us the ability to build a better agent for you. And so we're gonna then have this surface agent that does stuff for your life, and we'll know more about your life than anybody else because we know.

So, but of course, if it's just an API or MCP call away, then ~what you just, what,~ what are you doing? You're giving it away to the next guy to build their surface area

Fraser Kelton: ~Yep.~

Nabeel Hyatt: of trying to lock in that context. ~Yeah, I understand.~

Fraser Kelton: Yeah. ~Um, in the,~ was it a hard decision? ~How, how did you,~ how did Chris get to ~get to~ releasing this?

Nabeel Hyatt: Well, obviously it was a thought process because, ~um,~ granola did not open with an MCP surfer a year ago when it

Fraser Kelton: ~Yeah.~

Nabeel Hyatt: it was, it was obviously a thought process to be frank. and I can say it's, I just literally just came up in another board meeting with another company yesterday. I think it's a regular question. ~Um. W we,~ I think the first two years post ~JBTA~ lot of founders and a lot of VCs around [00:42:00] data moats were the only moats that mattered.

Fraser Kelton: ~Right.~

Nabeel Hyatt: ~um,~ and, ~uh,~ I need access to proprietary data. 'cause proprietary data is what makes things special for me. I think as we're realizing over time, there's no such thing as proprietary data.

First of all, ~like, like~

Fraser Kelton: ~Right.~

Nabeel Hyatt: does not. to be locked in. ~Um,~ fundamental truth of what a transformer is, is it transforms content from A to B quite easily.

Fraser Kelton: ~Hmm.~

Nabeel Hyatt: that means is that even if I tried to lock in your data, the ability for a consumer or a another startup to build a tool to rip out that data will always, ~uh,~ be stronger than your ability to lock in that data.

And it can go as simply to, by the way, like even if I tried to lock down a bunch of stuff in my startup, ~um,~ I'm just gonna build a client. If it was really that valued to you as a customer, you just run an agent on your desktop that takes

Fraser Kelton: ~Right, right,~

Nabeel Hyatt: then we're at a point where it can like OCR your screen all day, like

Fraser Kelton: ~right, right,~

Nabeel Hyatt: then put that in structured data, build a custom database from scratch, ~it's gonna take,~ I mean, like, ~there's just, it's,~ this is a fallacy.

This is the total fallacy. And so if it really is a fallacy, then who are you kidding? ~Uh,~ [00:43:00] and, and quite frankly, the more interesting thing that happens in the market right now is that everybody. Is building their own interfaces to their own data.

Fraser Kelton: ~right.~

Nabeel Hyatt: told a story about illicit building custom side software over the course of a weekend because they didn't like one or two pieces of the interface,

Fraser Kelton: ~Yeah.~

Nabeel Hyatt: wanted their content their way. This is just the immutable truth, so you just have to ask the question, Hey, they do granola still does a times better job taking a transcript of a meeting and summarizing it properly. And you can see this, ~there's ev,~ there's a million copies of granola at this point. Frankly, you can one shot a copy of granola this weekend.

It just won't be as good at the nuanced stuff. And in particular, that means the, ~the, the, the~ summarization and note taking part. And the other part that's also really, really hard that people will find is. Turns out if you have an org with a hundred employees or 20 employees and thousands of hours of transcripts, it turns out that's not like a greep away.

It turns out that actually, like finding information [00:44:00] specifically in transcripts and

Fraser Kelton: ~Yep.~

Nabeel Hyatt: through all that data and surfacing it really properly to get insights out of that

Making Context Ubiquitous

Fraser Kelton: ~Right.~

Nabeel Hyatt: a very deep and very hard problem. So you

Fraser Kelton: ~Yep.~

Nabeel Hyatt: that the company is gonna be good at that, or you come from a ~straight, a~ place of weakness where you think you're bad at it. And if you think you're good at it, then let the information be free. And if that means that there is some unique use case that somebody needs to build with cloud code for their dentist office or for, we're back to dentist again, by the way, for their, their dentist office or in New Zealand or whatever, like, then they should have access to the data.

It's their data anyway. And so, ~no, I I,~ it is. It is a little anathema to us Web 2.04 folks. ~Um,~ but I think the truth of this particular market is that the best interfaces, ~um,~ with the best customer-centric promises are gonna win. And the people who try to play petty are just gonna lose 'cause there's too many other ways around it.

Fraser Kelton: Yep. ~Make,~ I mean, ~uh.~ I don't think I would've used a granola app in that use case. ~Like a, uh,~ I, it's an infrequent thing that I would turn to it for that purpose, and it made my [00:45:00] affinity for granola 10 times higher just because of that one moment I was like, I, I absolutely love that I'm getting the information that I need right in this moment.

And I don't, I don't think, you know, I think it would be a really long roadmap for, for Chris to get that place for me if they, if they could even carve it out. ~Um,~ but it was, it was great. It made my ~my, it made my~ experience with granola much better.

Nabeel Hyatt: This is, ~you know,~ a little bit of the way I think about. Context and information in this AI age now is, ~um, you know,~ our, our partner Natalie Vais says like, ~uh, it, you know,~ sometimes commodification is the good thing. Like sometimes being the standard is the good thing, and everybody talks about it as the bad thing. But in, in dev tools and databases and so on, like being the default, like, yes, you're commodity, but the commodity is like ubiquitous and everywhere. it can't be universally true of startups, but it certainly be true of the startups that have the highest level of ambition.

Fraser Kelton: I think so. I think so

Nabeel Hyatt: Yeah.

Fraser Kelton: Should we be done?

Nabeel Hyatt: I think we should be done. This was a wonderful conversation. Let's [00:46:00] do it again . Before next model update. ,