Princeton Keynote at ICML: What's Left for Us to Do?
14 Jul 2026
A provocative question at ICML Seoul
At the International Conference on Machine Learning (ICML) in Seoul, a researcher who leads a Princeton University team focused on AI agent evaluation delivered a keynote with a pointed title: "What will be left for us to work on?" The talk was given the week before this article was written.
Beyond the title itself, the specific content and arguments of the keynote were not detailed in available reporting. Likewise, no information is available on what conclusions the speaker reached, or how the audience reacted. The speaker's name was also not disclosed in the source material.
The 'AI as Normal Technology' framework
The same speaker co-authored an essay titled "AI as Normal Technology" with Sayash Kapoor. According to the speaker, the essay offers a way to think about the medium-term future of AI and how individuals and organizations might adapt to it. The essay runs 15,000 words and is reportedly being expanded into a full book—though no publication date or timeline has been shared.
Why founders should care
For early-stage founders, this keynote and its underlying framework carry a few plausible implications, though the details remain thin:
- Task relevance may shift. The keynote's framing suggests founders may need to consider how advancing AI capabilities could reshape which tasks remain valuable for human or startup contribution—though it's unclear what specific guidance, if any, the talk offered on this front.
- Adoption as a gradual process. The "AI as Normal Technology" framing could signal that founders should plan for AI adoption as an incremental, adaptable process rather than a sudden, disruptive shift. This is a reasonable interpretation of the essay's stated purpose, though the essay's full arguments are not detailed here.
- Agent evaluation may matter more. The speaker's focus on AI agent evaluation at Princeton may point to the growing importance of rigorous assessment methods for founders building agent-based products—an area likely to see increased scrutiny as agentic AI systems proliferate.
What's still unclear
Several gaps remain in the available information: the keynote's actual arguments and conclusions, the speaker's identity, the book's publication timeline, and any audience response. Founders interested in this space should watch for the forthcoming book version of "AI as Normal Technology," which may offer more concrete guidance on positioning startups relative to evolving AI capabilities.
For now, the keynote's title alone—posed to a room full of machine learning researchers in Seoul—serves as a useful prompt for founders to keep asking the same question of their own roadmaps.