Is AI-Washing Over? Investors Seek Real Innovation in Startups

Why the Remote Labor Index Shows Limits of AI in Real Work Automation

Update Understanding the Remote Labor Index and AI’s LimitationsA newly published research paper from the Center for AI Safety has unveiled the Remote Labor Index (RLI), a significant benchmark designed to evaluate the effectiveness of AI agents in performing real, paid remote jobs. Although AI's advancements are undeniably promising, the results reveal a sobering reality for those anticipating a shift towards widespread automation. Current AI agents, as assessed by the RLI, demonstrated a strikingly low performance, with Manus, the leading AI, managing to automate only 2.5% of the evaluated tasks. Other sophisticated models like Grok 4 and Sonnet 4.5 were not far behind, achieving only 2.1% automation rates, while models like GPT-5 and Gemini 2.5 Pro fell to 1.7% and below 1%, respectively.The Implications of Low Automation RatesThese results indicate a significant gap between AI’s capacities and the requirements of complex, professional work. While humans excel in creativity, planning, and execution, AI is still struggling to deliver work that fulfills professional standards. Researchers found that the majority of AI failures stem from issues like incomplete submissions, quality discrepancies, and technical errors. In fact, 45.6% of submissions received by human evaluators failed due to poor quality, while over one-third were incomplete or malformed.Why AI Agents Are Not Designed for Complex TasksPaul Roetzer, founder and CEO of the Marketing AI Institute, shared insights into why current AI benchmarks may not effectively represent their potential capabilities. Specifically, the benchmark tests general agents that are not tailored to specific job functions like software development or architecture. In specialized settings, the efficacy of AI could be considerably higher. For instance, OpenAI has been actively engaging finance professionals to instruct their models on investment banking roles, pointing to a possibility that specialized agents may perform tasks more effectively than their general counterparts.Deciphering the Future of AI in the WorkforceWhile the RLI presents a talk about stagnation, it’s essential to view this through a lens of growth and evolution. As AI technology advances, there is a notable trend towards specialization that could potentially enhance performance. AI agents are notably good at executing smaller, discrete tasks but often fall short when needing to complete comprehensive projects requiring multiple skills or steps. Thus, even as we see low automation rates, the groundwork is being laid for future AI capabilities.Balancing Human and AI CollaborationDespite AI’s shortcomings, Roetzer stresses that human oversight remains critical. Automation does not eliminate the need for human intelligence—rather, it amplifies it. As AI agents become increasingly capable, their integration into the workplace is likely to lead to a reevaluation of job roles and necessary skill sets. Ultimately, the collaboration between humans and AI may enhance productivity, potentially reducing the number of workers needed to complete specific tasks, rather than replacing the workforce entirely.Final Thoughts on AI’s Journey AheadThe Remote Labor Index serves as a crucial tool to gauge the current state of AI capabilities are practicing real-world tasks. The reality shown by the data indicates that while AI is on a developmental journey, the expectation of immediate or profound shifts in the workforce is premature. As advancements unfold, it will be important for stakeholders to understand both the limitations and opportunities AI presents moving forward.