TheSequence Opinion #904: The Age of Research Is Overrated. AI Engineering Is Winning
TheSequence Jesus Rodriguez
Ilya Sutskever sketched three eras of modern AI: research-focused (2012–2020), scaling-dominated (2020–2025), and research returning with bigger computers from 2025 onward. Frontier models in 2026 still use Transformers with Mixture-of-Experts routing, with improvements coming mainly from data quality, context length, reinforcement learning, and system-level engineering rather than novel architectures. The field is blurring the line between research and engineering, with scaling continuing but disguised as engineering optimizations across the entire system.
Why it matters
Why AI’s next breakthroughs may come from the learning loop around the Transformer—not from replacing it.
