Research Without Researchers

The university lab has served many purposes, but two have always been tightly linked: producing knowledge and training researchers.

That link made economic sense because intelligence was scarce, expensive, and embodied in people. A professor who wanted to pursue five ideas needed roughly five researchers. The work trained those researchers, and the researchers produced the work.

Mythos-class systems such as GPT-5.6 Sol and Claude Fable 5 break that arrangement. They can survey the literature, derive proofs, write code, design experiments, analyze results, criticize conclusions, and draft papers.

For research whose inputs, tools, and outputs pass through a computer, current AI can usually perform the work better and faster than the graduate researchers who previously did it. Wet labs, fabrication, field research, clinical work, and complex hardware will move more slowly because they still require physical access.

For most computer-executable research, the next dollar now produces more output when spent on compute than on junior researchers.

Academic funding sources overlap, but their primary objectives differ.

Government training programs will continue supporting students because education is the result they are purchasing. Industry may continue sponsoring labs for recruitment, relationships, or access. Public-interest work and physical facilities may remain inside universities even when they are not maximally efficient.

But funding whose purpose is research output will move toward whoever has access to the strongest systems and the most compute. Once a funder can buy more results from a compute-rich institution than from a university lab, paying for apprentice research labor stops making sense.

Universities will specialize in producing researchers. Compute-rich institutions will specialize in producing research.