Ep. 218: Anthropic IPO, Trump AI Executive Order, Rising AI Costs & OpenAI Merges Codex Into ChatGPT

TL;DR

• 1

  Anthropic says AI is starting to build AI: In its "When AI Builds Itself" essay, Anthropic claims Claude writes more than 80% of code merged into its codebase and that autonomous task length is now doubling every 4 months, pointing toward possible recursive self-improvement.

• 2

  The IPO race is becoming real: Anthropic confidentially filed to go public after a recent $65 billion Series H round at roughly a $965 billion valuation, setting up a likely market showdown with OpenAI and giving investors a cleaner read on frontier lab economics.

• 3

  Washington is moving from watching AI to trying to shape it: Trump's new executive order creates a voluntary 30-day early access process for "covered frontier models," while talk of government ownership stakes in AI labs shows how quickly AI has become a bipartisan power issue.

• 4

  AI costs are no longer theoretical: Uber capped employee spending on tools like Claude Code and Cursor at $1,500 per month per tool, Microsoft called Anthropic "extremely expensive," and Sam Altman admitted cost concerns emerged "quite suddenly" in early 2026.

• 5

  The real prize is not faster work, it's radically smaller companies: Paul points to Anthropic's scenario where a 100-person company can do the work of 1,000, then 10,000 or more, arguing this is the practical threat every legacy business should already be planning for.

• 6

  OpenAI is turning ChatGPT into an agentic work platform: Codex is being folded directly into ChatGPT with role-specific plugins, internal tool building via "sites," and a path from coding assistant to general-purpose work agent for sales, design, analytics, and finance.