Jun 16, 2026 · 52 min · 10 segments
The Startup Podcast is back for its 300th episode – and it’s an absolute doozy. In today’s Reacts, hosts Chris Saad and Yaniv Bernstein break down how Anthropic became the new industry leader in AI…
Chris SaadHostYaniv BernsteinHost



Uh, and now most recently Fable, which is another interesting story, I think, and we can unpack all of these.

And, and, you know, a, a sort of quite a surprising turnaround, and I think there are two...
The first is one that's as old as time and that we've discussed on this podcast before, which is around product discipline.
"We're gonna have like apps inside ChatGPT and, you know, and we're gonna have advertising, and we're gonna make sex bots and like whatever." And I think what that meant is it took away from their focus on building the best product.
Now, the other side of it, which was maybe a little bit harder to predict, although it seems obvious in hindsight, is all about coding, right? So large language models can do all sorts of things, and for a long time the, the nexus of competition was like, you know, which one is best at doing writing? Which one's best at doing analysis and aggregation and so on? And, and it was sort of purely about the large language models.
Uh, you sort of moved from GitHub Copilot to Cursor, and it, it started to become useful.
But what wasn't perhaps obvious is that coding isn't just another domain in which large language models can be effective, but that it's kind of this meta domain, right? Because even if LLMs are not good at various things, if they're good at writing code, they can then write programs that are able to cover for their gaps in the same way that as humans we, we build tools and then we use those tools, and that's how we've become so dominant as a species.
What AI can do is by writing code, then it can become really, really effective.
How many Rs are there in strawberry?" And it says, "Oh, there are, there are three Rs in strawberry," or, "No, there's five Rs in strawberry." And like, it gets it wrong.
But yes, obviously LLMs are very bad at counting the number of letters in a word, and that's, uh, you know, if, if you go into the technicalities, that's to do with how the tokenization is done, blah, blah, blah.




Uh, and now most recently Fable, which is another interesting story, I think, and we can unpack all of these.

And, and, you know, a, a sort of quite a surprising turnaround, and I think there are two...
The first is one that's as old as time and that we've discussed on this podcast before, which is around product discipline.
"We're gonna have like apps inside ChatGPT and, you know, and we're gonna have advertising, and we're gonna make sex bots and like whatever." And I think what that meant is it took away from their focus on building the best product.
Now, the other side of it, which was maybe a little bit harder to predict, although it seems obvious in hindsight, is all about coding, right? So large language models can do all sorts of things, and for a long time the, the nexus of competition was like, you know, which one is best at doing writing? Which one's best at doing analysis and aggregation and so on? And, and it was sort of purely about the large language models.
Uh, you sort of moved from GitHub Copilot to Cursor, and it, it started to become useful.
But what wasn't perhaps obvious is that coding isn't just another domain in which large language models can be effective, but that it's kind of this meta domain, right? Because even if LLMs are not good at various things, if they're good at writing code, they can then write programs that are able to cover for their gaps in the same way that as humans we, we build tools and then we use those tools, and that's how we've become so dominant as a species.
What AI can do is by writing code, then it can become really, really effective.
How many Rs are there in strawberry?" And it says, "Oh, there are, there are three Rs in strawberry," or, "No, there's five Rs in strawberry." And like, it gets it wrong.
But yes, obviously LLMs are very bad at counting the number of letters in a word, and that's, uh, you know, if, if you go into the technicalities, that's to do with how the tokenization is done, blah, blah, blah.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.