TheSequence Opinion #904: The Age of Research Is Overrated. AI Engineering Is Winning
Why AI’s next breakthroughs may come from the learning loop around the Transformer—not from replacing it.
TL;DR
- AI history can be divided into an age of research (2012-2020) and an age of scaling (2020-2025), with a potential return to research now.
- Current frontier AI models still incorporate familiar architectures like Transformers and Mixture-of-Experts.
- Headline improvements are coming from better data, longer context, stronger reinforcement learning, synthetic tasks, tool use, memory, verification, adaptive reasoning, and agent orchestration.
- This evolution is compared to Formula 1, where gains come from aerodynamics, energy recovery, tires, software, and pit strategy rather than a completely new car design.
- The field is likely in both an age of research and engineering, with research now expressed as industrial-scale engineering.
- Scaling in AI has not ended but has "escaped" into engineering applications.