In this episode

* The two stories you heard this week, and why both are true

* Why “the truth is somewhere in the middle” is the expensive wrong answer

* The distribution gap: where the gains are actually clustering, and why code sprints while writing, search, and advice crawl

* The split nobody covered: inside the paid tier, the people who use these systems versus the people who build the harness and drive them

* Why this isn’t just another early-adopter curve that closes on its own

* What “driving” actually looks like, and why most people never cross over

* The one new muscle to add if your career was built on judgment, not code

Referenced in this episode

* Andrej Karpathy’s April thread on the two groups talking past each other

* Jason Lemkin (SaaStr): “how awed you are by AI is almost perfectly correlated with how much you actually use it to build”

* Anthropic’s report on how agents actually get used (software engineering was nearly half of all agent activity)

* The roughly one-million-conversation study on how people use ChatGPT (coding around four percent of messages)

* Gary Marcus calling Claude Code the single biggest advance in AI since the large language model

* 2025 McKinsey, Gartner, and IT-leader surveys on stalled AI projects and agent sprawl

From the book

This is chapter one of Builder Leader: The AI Exoskeleton That Crosses the Gap. Each episode takes one idea from one chapter and talks it through. Subscribe to get the next one.