Doing Okay-ish with AI at the AI-proof competition (blog.greg.technology)

In the recent ICFP Programming Contest 2026, a solo participant utilized AI, specifically the Fable framework, to achieve a commendable top-25 finish among approximately 200 teams. This demonstration highlights the growing integration of AI in competitive programming, where complex tasks traditionally requiring human ingenuity are increasingly being tackled by automated systems. While the participant mainly oversaw the AI's operations due to its unpredictable behavior, this experience brings to light both the potential and limitations of current AI technologies in handling programming challenges. Significantly, the contest introduced a 2D esolang, which presented unique obstacles that tested the capabilities of AI models. The participant noted that while AI can efficiently solve linear programming tasks, it struggles with more complex, multidimensional problems. Throughout the contest, the AI displayed notable tendencies, such as losing focus and misunderstanding its own directives. This case illustrates the necessity for robust testing and validation mechanisms when using AI to avoid common pitfalls in programming tasks, ultimately igniting a conversation about improving orchestration and tooling for better AI-driven outcomes in the field.