Skip to playerSkip to main content
  • 8 minutes ago
“I haven't written a line of code by hand in 8 months.”

At Fortune Brainstorm Tech, Anthropic Head of Claude Code Boris Cherny said AI now handles much of the coding process. As a result, the bottleneck has shifted from writing code to reviewing it—and Anthropic uses multiple Claude agents with different personas and perspectives to collaborate on code reviews.

“What we found is with this approach we catch pretty much every bug,” he said.

Category

🤖
Tech
Transcript
00:00I haven't written a line of code by hand in, you know, I think eight months now.
00:03Quad code, co-work, 100% written by quad code, code at Anthropic.
00:07At our biggest customers, like all the way from, you know, like Salesforce and, you know, like NASA to YC
00:12startups,
00:13it's starting to become close to 100% written by quad code and, you know, like the numbers are just
00:17kind of growing.
00:18As you kind of get rid of coding as the bottleneck, so I don't have to do this anymore, my
00:22agents do the coding,
00:23what's the next bottleneck? Well, it's reviewing the code.
00:25So for us at Anthropic, we kind of, we thought pretty hard about it, we tried a bunch of different
00:29ideas,
00:29and the thing that we came up with is using Claude to do the code review.
00:33And the way we do it is essentially we have a team of Claude's, they all have kind of different
00:36personas
00:37and they have different perspectives and they essentially collaborate to review the code.
00:41And what we found is with this approach, we catch pretty much every bug.
00:45It's a really expensive code review. So essentially you throw tokens at it, it's like a test time compute thing.
00:50But as a result, you catch all the bugs and obviously there's less expensive versions too.
Comments

Recommended