AI agents speed up PhD execution work

The author argues that while AI agents significantly speed up the execution layer of AI research—such as writing boilerplate code, running experiments, and drafting text—they cannot replace human domain knowledge and critical judgment. Addressing complex problems requires understanding real-world complexities like imperfect data detection, which AI tools alone cannot resolve. Consequently, the author predicts that researchers will increasingly transition into team leads, domain knowledge will become a primary advantage, and labs will grow more interdisciplinary. Open article →