agentsfactbearishProduction AI agent failures are more often due to inability to manage reasoning context (conversation histories, prompts, tool outputs) rather than inability to reason wellArtificial Intelligence29 Jul 2026http://arxiv.org/abs/2607.21503v1