Report: First-Year Employee Turnover Plummets

The "Great Resignation" was characterized by a wave of voluntary resignations beginning in early 2021, driven by factors like wage stagnation, lack of career advancement, and a desire for better work-life balance. By mid-2023, the quit rate had largely returned to pre-pandemic levels, influenced by a more competitive job market. Recent labor market data from late 2024 and early 2025 shows a rebalancing, with quit rates declining as workers' confidence in finding equivalent or better jobs decreases. This stabilization in the workforce coincides with a surge in AI's impact on employment, with AI expected to create 133 million new jobs globally by 2025 while displacing 75 million. Some estimates suggest AI could replace the equivalent of 300 million full-time jobs by 2030, transforming a quarter of work tasks in the US and Europe. This shift is creating new roles in areas like AI oversight and robotics maintenance. For AI labs, securing high-quality human feedback for model alignment is a critical operational challenge. Techniques like Reinforcement Learning from Human Feedback (RLHF) are essential for training models to be helpful and harmless, but this process is resource-intensive. RLHF involves collecting preference data from human evaluators who rank model outputs, which is then used to train a reward model. To address the scalability issues of RLHF, some labs are turning to Constitutional AI, an approach developed by Anthropic. This method trains models using a predefined set of principles—a "constitution"—to critique and revise their own outputs, reducing the reliance on extensive human labeling. The goal is to embed ethical guidelines directly into the training process, making the AI inherently more aligned with human values. The demand for nuanced data has also created a choice between synthetic data generation and human annotation. While synthetic data offers scalability and cost-effectiveness, it often lacks the contextual accuracy that human labelers provide, especially for complex tasks. A hybrid approach is often most effective, using synthetic data for volume and human-labeled data for refining accuracy and addressing bias. As AI becomes more autonomous, evaluating "agentic" AI systems that can perform multi-step tasks is a growing field. Evaluation frameworks are moving beyond simple task completion metrics to assess reasoning, tool use, and reliability. This creates new opportunities for data labeling focused on validating the entire process an AI agent undertakes, not just the final output. For AI infrastructure startups, the fundraising climate remains robust, with AI companies attracting a significant portion of venture capital. In early 2026, seventeen U.S.-based AI startups raised over $100 million each in the first two months alone. However, investors are increasingly focused on sustainable business models and are scrutinizing the high costs associated with model training and infrastructure. A successful go-to-market strategy for selling to technical buyers at AI labs requires a deep understanding of their buying committees, which include economic, technical, and user buyers. The sales cycle can be long, and the strategy must align product, sales, and marketing around a clear value proposition for each persona. As the AI market matures, regulatory compliance is also becoming a critical factor for securing investment and building trust with enterprise customers.