AI agents can do 16% of freelance work — and AI judges can't tell

The Remote Labor Index reports that Fable 5 now completes 16.1% of freelance jobs at a professional quality standard, a massive increase from the 2.5% baseline established less than a year ago. This benchmark covers commercially viable tasks like CAD modeling and graphic design, where Fable 5 significantly outperformed competitors like Opus 4.8 and GPT-5.5. However, attempts to replace human evaluation with AI models failed, as automated judges scored work up to three times higher than human experts, missing critical flaws like fake geometry. This indicates that while agents are rapidly acquiring valuable skills, current frontier models still struggle with the complex tool use and judgment required to supervise high-quality work.