What will be left for us to work on?
AI Snake Oil · Arvind Narayanan · 2026-07-13
Princeton AI researcher Arvind Narayanan argues in an annotated ICML keynote that AI is a 'normal technology' requiring decades of economic adaptation, that agent reliability lags capability gains, and that human-AI co-superintelligence rather than displacement is the achievable future.
Extraction
Topics: ai-economicsai-labor-displacementrecursive-self-improvementai-agent-reliabilityhuman-ai-collaboration
Claims
- AI is best understood as a 'normal technology' analogous to electricity, whose economic transformation will take decades of organizational adaptation rather than arriving suddenly from a lab breakthrough.
- AI agent reliability—covering consistency, robustness, calibration, and operational safety—has improved by only five to ten percentage points over the past two years even as raw capability shot up dramatically, limiting practical automation deployments.
- Writing code is not the bottleneck in software engineering, so AI-driven productivity gains in the 'execute' layer do not proportionally reduce demand for software engineers whose work also spans planning and delivery.
- Recursive self-improvement in a lab setting will not automatically produce AGI or superintelligence because creativity and other non-verifiable cognitive dimensions cannot be improved through purely computational self-modification.
- The future of AI work will shift effort from building (verifiable, automatable tasks) to evaluation (judgment-based tasks that are resistant to automation), analogous to the shift from rowing to navigating a ship.
- Claims that AI-driven layoffs are replacing software engineers are contradicted by data; companies under financial pressure are using AI as a convenient explanation for economically motivated cuts.
Key quotes
If you believe, as I do, that this is a technology that will greatly amplify our potential, then now is the best time to build skills — especially the skills that are going to be complementary to what AI is doing and is going to be able to do.
I've learned that if I don't feel exhausted at the end of the day, I've done something wrong. I've offloaded too much to AI. I've sacrificed too much of my long-term growth in the pursuit of short-term productivity.
Effort shifts from 'rowing the boat' to 'steering the ship, navigating the ship, and figuring out where we even want to go'.