The Sequence Opinion #868: Recursion Is the New Scaling Law

The technique pushing breakthroughs in agentic computing.

For most of the modern AI era, progress has had a deceptively simple recipe: make the model bigger, train it on more data, and spend more compute. This formula produced the Transformer era, the foundation model era, and the current wave of large language models. Scaling laws gave the field an almost industrial rhythm. Loss curves became roadmaps. Compute budgets became strategy. The frontier could often be described by a single question: how much larger can we go?

But the most interesting recent progress in AI is beginning to feel less linear. It is no longer just about building a larger model that gives a better answer in one pass. Increasingly, the frontier is about models and systems that can revisit, revise, search, simulate, critique, and improve. The important unit of computation is shifting from the forward pass to the loop.

That shift suggests a provocative idea: recursion may be the next scaling law.