Eric Schmidt, the former Google chief executive who once ranked himself among the best programmers of his generation, told a Blackstone podcast this week that artificial intelligence has now surpassed him at the craft he built his career on. The admission from a 71-year-old technologist who helped shape the modern internet marks a symbolic inflection point: the layer of software creation that once required elite human talent is being absorbed by the very tools that talent produced.
The architect analogy
On Blackstone's "Inside Blackstone" podcast, host Christine Anderson pressed Schmidt on the shifting role of the engineer. He responded with a housing metaphor: his generation had to lay every brick, while the next will simply describe the blueprint and watch agents pour the foundation. Schmidt name-checked Claude Code, OpenAI Codex and unreleased Google products as the new general contractors, arguing that fluency in code remains essential, not to write it, but to issue precise orders to the systems that do.
Inside the agent-first shop
The transition is already measurable inside AI-native companies. At Daytona, a 30-person startup, chief executive Ivan Burazin told Business Insider that each of the firm's 16 engineers now supervises an average of five AI agents and no human writes code directly. Anthropic chief financial officer Krishna Rao offered a parallel data point on the "Invest Like the Best" podcast: more than 90 percent of the company's codebase is generated by its own Claude Code tool, and the finance function has begun outsourcing statement preparation to the same layer. "Everyone kind of becomes a manager," Rao said.
The leverage question
Schmidt's framing treats the shift as a promotion for engineers, from builders to architects, but the economics of that promotion remain unproven. If a single developer can direct five agents, the marginal cost of software output collapses, which should compress pricing power across the industry. The source does not disclose whether Daytona or Anthropic have reduced headcount plans or adjusted billing models in response. Nor does it reveal what happens when the agents hallucinate a load-bearing wall: the oversight burden may scale faster than the productivity gain.
What to watch next
The next signal will come from the large platforms. Schmidt noted that Google is building its own agentic coding products but offered no timeline or pricing. If the hyperscalers bundle these tools into existing cloud contracts at near-zero marginal cost, the independent agent startups face a distribution disadvantage they cannot code their way out of. For now, the former CEO who once wrote the house has accepted that his name is on the blueprint, and the crew showing up tomorrow does not ask for a salary.