New Study Says AI Agents Face a Computational Ceiling

A study authored by Vishal Sikka and Varin Sikka argues that large language models cannot perform certain computations or multi-step tasks beyond a certain complexity, effectively placing a hard ceiling on what ‘agentic’ AI can achieve and challenging claims of near-term autonomous AI or artificial general intelligence. While LLMs may improve, the research suggests they won’t exceed their computational limits or replicate true human-like intelligence in the foreseeable future.