Korva is a scroll of the internet’s best ideas.
Constraints made Anthropic focus on coding
Freedom in constraints
Constraints bring a kind of freedom, because they strip away excess choice and make the path clear. Anthropic began without the free cash flow or distribution of Meta or Google, so it had to choose a narrow focus and push for escape velocity, even with a technology that applies almost everywhere. A document written by one of the founders in 2021, a few months after the company started, makes the case for focusing on AI coding. That was long before anyone knew what the market opportunities were.
Why coding
Coding paid off in two ways. It was going to be commercially valuable, and it speeds up research: the best coding models help researchers, which accelerates the research loop that produces better models. Dario has seen that clearly for years, and people at other labs now talk about it too.
How does coding help speed up AI research?
AI researchers spend much of their time writing code to run experiments, process data, and build the systems that train models. A stronger coding model can write and debug that code, which shortens each cycle of trying an idea and seeing the result. Faster cycles mean more experiments per month. A 2022 controlled study by GitHub and Microsoft researchers, led by Sida Peng, found that developers using GitHub Copilot finished a programming task 55 percent faster than those without it.
Why do constraints sharpen strategy?
Psychologist Patricia Stokes argued in her book Creativity from Constraints that limits push people toward original solutions, citing the work of artists such as Monet and Picasso, who each restricted their own methods. A tight budget removes the option of pursuing every opportunity at once. Without that option, a team has to decide what matters most.
Why do broad technologies start with one wedge?
A narrow first market lets a company serve one group of customers extremely well before widening. Amazon began by selling only books in 1995, and Facebook launched in 2004 at Harvard alone before opening to other colleges. A general-purpose technology has no obvious first customer, so choosing one becomes a strategic decision.
More from Lenny's Podcast: Product | Growth | Career
