Google is Not a Serious Company

TL;DR

• 1

  Google's AI stack looks strong on charts but weak in real use: The hosts say Gemini 3.5 Flash loops, mishandles tools, and feels "two generations ago" in behavior despite benchmark optics, with one citing a long-running engineering benchmark where Gemini 3.1 Pro scored around 10% versus leaders in the 60s and 70s.

• 2

  Google Cloud's reliability issues are not theoretical anymore: They point to Railway allegedly having its entire GCP account deleted by an automated Google process, using it as evidence that Google is failing at the unglamorous but crucial part of AI infrastructure, which is keeping systems dependable.

• 3

  China's AI relationship with the West may be hardening fast: The Meta-Manis acquisition being undone by Beijing after employees were already onboarded is framed as a major signal that China is no longer willing to let promising AI companies or open-weight models become stepping stones for US firms.

• 4

  Anthropic's economics look brutal even by AI standards: They cite a reported $1.25 billion per month SpaceX compute bill alone, arguing that Anthropic, OpenAI, and xAI are barreling toward IPOs while still operating as massively unprofitable companies with valuations that only make sense if AGI actually arrives.

• 5

  The next software wave is not better prompts, it's better glue: Their main builder rant is that deploying, auth, databases, env vars, and dashboards are now the bottleneck because apps can be built in 40 minutes instead of 40 hours, which is why they built Lakebed, a "slop cloud" that has already hosted 400-plus apps.

• 6

  Coding agents are making infrastructure abstractions more important than languages: The hosts argue that markdown files, CLIs, and agent harnesses now behave like software runtimes, with memorable lines like "Claude Code is the new Node" and a broader point that the application layer is collapsing into cheap, disposable AI-generated software.