Two people pay for the same AI, run the same model, hand it the same task. One gets work the other can’t come close to. Episode one is about why, and why the gap that decides your next two years isn’t the one everyone is arguing about.
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.
