Alibaba Cloud has developed DualLane, a system that handles technical support inquiries by deploying large language models only when necessary, using pre-validated solutions for common issues. Described in the paper “DualLane: Fast and reliable LLM agents for interactive AIOps via dual-path planning,” the research was presented at SIGKDD 2026 and prompted by problems with autonomous AI agents, including wrong tool selection, incorrect parameters, misprocessing of previous steps, and errors in final responses. DualLane processes every support query via two paths simultaneously: a fast path classifies known issues and halts slower processing, delivering template-based solutions at a cost of only a few tokens; a slow path allows an LLM agent to use up to approximately 3,000 tokens for less common issues. Alibaba estimates the cost at about $0.001 per ticket. The system detects recurring patterns in slow-route questions, enabling employees to create verified templates for future fast handling. Now in production, DualLane achieved an accuracy of 96.5 percent in offline tests, with lower latency than alternative approaches such as LLMCompiler and ReAct. Alibaba is using AI to automate support tasks and to determine when it is more efficient to keep a large language model out of the process.