BroadChain learned on April 28 that at the YC W26 batch Demo Day on March 24, a venture capitalist noticed that 80% of the projects on stage were vertical agents—helping lawyers organize documents, assisting customer service with ticket assignments, and aiding HR in screening resumes. Five months ago, these projects were still considered innovative, but now Claude Code has evolved from a developer tool into a universal interface. With the release of Opus 4.6, the barrier to Vibe Coding has nearly dropped to zero. These vertical agents have yet to establish commercial moats, and even ordinary engineers or investors themselves could build them over a weekend, rendering their investment value obsolete.
YC's three-month incubation cycle appears sluggish in the face of AI iteration speed. From application, screening, onboarding, to demo day, this process—which operated for over a decade in the mobile internet era—can now undergo several paradigm shifts within five months. A friend working in post-training remarked bluntly: "Silicon Valley itself has already started to fall behind."
Tens of thousands of engineers at Meta are all using competitor Anthropic's Claude Code to write code. This is not a startup team or experimental project, but a trillion-dollar company. Code security has been set aside, token budgets have skyrocketed, and efficiency trumps everything. While Google prohibits most employees from using Claude Code or Codex, DeepMind is an exception—multiple teams responsible for Gemini models and internal applications are using it. Google's self-developed Antigravity tool claims to have 50% of new code generated by AI, but DeepMind still chose Claude Code because Anthropic offered a private deployment solution.
No one dares to stop in this arms race. Meta internally developed myclaw in an attempt to build its own coding tool, but abandoned it due to being "not user-friendly and unused," ultimately lifting all restrictions: as long as it doesn't involve customer data, anyone can use Claude Code. Various departments have started holding internal meetings on "how to become an AI-native organization," conducting training and assessments. All security practices have been sidelined in favor of efficiency.