Claude Code's multi-agent code review feature, which traverses an entire codebase simultaneously, only became production-reliable with Opus 4.5, Opus 4.6, and Sonnet 4.6.

“It was only with like Opus 45 and 46 that we and Sonnet 4.6 that we felt like okay we are now able to like run multiple code review agents simultaneously to traverse the entirety of the codebase and to synthesize a set of like real issues that an engineer needs to address before merge.”

Over 90% of Anthropic's own code is now written by Claude Code.

“90 plus% of our code is actually written by cloud code.”

Anthropic worth watching: run rate revenue tripled in one quarter, net dollar retention above 500%, and over 90% of its own code now written by Claude Code

“Run rate revenue went from 9 billion to 30 billion in one quarter.”

Claude Code enabled a non-technical CFO to build a sophisticated portfolio tracking tool replacing Addepar in roughly one week, a striking demonstration of vibe-coding for finance professionals.

“He's never coded in his life. Like when I told him about Cloud Code and that he should check it out, he's like, 'I don't know how to code. Like I'm not technical.' And literally within like a week, I'm so lucky that he's the kind of person that would do this. But he built this and it's just it's insane.”

Cat Wu on the shift from chat to action as the defining product paradigm change.

“The big shift is that the 2024 generation of products were chatbased and the quad code generation of products is action-based.”

Claude Code grew from zero to approximately $2.5 billion ARR, and Cursor reached approximately $2 billion ARR, each in a matter of months.

“Cloud Code going from like zero to 2.5 billion in ARR, Cursor going to like 2 billion ARR in just like the last few months.”

Cursor: an AI coding tool that reached approximately $2 billion ARR in just a few months, signaling explosive enterprise adoption.

“Cloud Code going from like zero to 2.5 billion in ARR, Cursor going to like 2 billion ARR in just like the last few months.”

A parallel sub-agent architecture fans out research tasks to five specialized sub-agents (Perplexity, Exa, Context 7, FireCrawl, XMCP) and aggregates results back to a main agent to produce best-practice outputs.

“I'll tell Codeex like hey or Claude Code, hey spawn five sub agents. one sub agent for perplexity, one for exo, one for context 7, one for fire crawl, one for XMCP because I like to pull from different resources and then those all come back to the main agent and we get the best practices.”

Co-work, a Claude-powered product showing faster user adoption than Claude Code did at the same point post-launch, worth tracking as an agentic workplace tool.

“Then you also see something like co-work come along and start to unlock that faster than claude cloud code was if you index them to the same point in time.”

Over 90% of code inside Anthropic is written by Claude Code

“Over 90% of code inside Anthropic is written by Claude Code.”