The eternal complement

OpenAI publishes “The eternal complement,” a long essay on progress: human insight has outrun the capacity to realize ideas. It cites research-productivity work associated with Nick Bloom — more researchers to sustain Moore’s law, economy-wide research effort up while measured productivity fell — and the rising cost of instruments, fabs, and technician labor.

The thesis: frontier intelligence and execution capacity are economic complements. “Institutional intelligence” — the uncelebrated work of carrying ideas through bureaucracy, supply chains, and correct local actions — may be where advanced AI spends most of its time. Today AI makes execution less scarce; tomorrow a flood of AI-generated agendas could make execution the bottleneck again.

We’re glad to see a lab essay that praises the welder and the nurse’s insurer, not only the mythic genius. If AI’s near-term value is code, literature search, and turning sketches into prototypes, that’s useful AI we recognize — the support staff ambitious people never get.

The darker turn in the piece is honest: when models propose thousands of film concepts and hypotheses, taste and execution become scarcer than ideas. That’s not a reason to slow curiosity; it’s a warning against confusing a chatbot brainstorm with a civilization that can build JWST-class instruments.

I’m watching whether OpenAI’s product roadmap matches the essay — agents that reliably complete institutional chores — or whether “eternal complement” becomes soft branding while the company ships spectacle. The argument earns a pro-tech read when it keeps respect for the boring work.

Framed for OpenAI’s “Intelligence Age” content line; more manifesto than product note, but it reframes why agentic execution features matter economically.