# Crashing Waves vs. Rising Tides: Preliminary Findings on AI Automation from Thousands of Worker Evaluations of Labor Market Tasks

**Authors:** Matthias Mertens, Adam Kuzee, Brittany S. Harris, Harry Lyu, Wensu Li, Jonathan Rosenfeld, Meiri Anto, Martin Fleming, Neil Thompson  
**arXiv:** [2604.01363](https://arxiv.org/abs/2604.01363)  
**Date:** 2026-04-01  
**Summary page:** https://www.summarizepaper.com/en/paper/2604.01363v1/

## Key points

- Rising tides are the primary form of AI automation
- Steady improvement in AI performance across a wide range of tasks
- Models successfully completing tasks that take humans approximately 3-4 hours with a 50% success rate in 2024-Q2
- Projected success rate to increase to about 65% by 2025-Q3 and reach higher levels by 2029
- Gradual improvement in AI capabilities referred to as rising-tide automation
- Workers likely have more visibility into changes and time to adjust compared to crashing waves of automation
- Contrast with METR's work showing a shallower relationship between task duration and AI success rates across most job domains
- AI capabilities growing steadily over time, providing workers with a smoother transition into an increasingly automated workforce

## Summary

Our study on AI automation explores the continuum between crashing waves and rising tides in terms of AI capabilities across various tasks. By evaluating over 3,000 text-based tasks from the U.S. Department of Labor O*NET categorization, we have found evidence that rising tides are the primary form of AI automation. Our analysis shows a steady improvement in AI performance across a wide range of tasks, with models successfully completing tasks that take humans approximately 3-4 hours with a 50% success rate in 2024-Q2. This success rate is projected to increase to about 65% by 2025-Q3 and reach even higher levels by 2029. This gradual improvement in AI capabilities, referred to as rising-tide automation, suggests that workers will likely have more visibility into these changes and more time to adjust compared to a scenario characterized by crashing waves of automation. Our findings also contrast with recent work by METR which showed a shallower relationship between task duration and AI success rates across most job domains. This indicates that rather than sudden bursts, AI capabilities are growing steadily over time, providing workers with a smoother transition into an increasingly automated workforce. Overall, our study sheds light on the trajectory of AI capabilities and their implications for labor markets and the economy. It underscores the importance of understanding the gradual nature of AI automation and its potential impact on various industries as organizations adopt these technologies.