Industrial asset management workflows powered by large language model (LLM) agents promise end-to-end automation but face real-world hurdles in latency, concurrency stability, and safety, according to a research paper published on arXiv. The paper presents DynAMO (Dynamic Asset Management Orchestration), a deployment-ready engine designed to overcome these barriers through a Plan-then-Execute architecture that generates verifiable workflow graphs.
Introducing DynAMO
DynAMO, developed by researchers Kanishk Kushwaha, Vikrant Vinod Bansode, Harsh Vardhan, and Dhaval C. Patel, supports two execution modes: SequentialWorkflow, which runs tasks in topological order, and ParallelWorkflow, which leverages dependency-aware concurrency to run independent tasks simultaneously. By dynamically identifying independent tasks, DynAMO preserves structural correctness and safety while improving efficiency through controlled reasoning overlap, the authors reported.
Performance Benchmarks
The researchers tested DynAMO across six controlled experiments on the AssetOpsBench industrial benchmark. Key results include:
| Metric | Sequential Orchestration | Parallel Orchestration | Improvement |
|---|---|---|---|
| Median end-to-end latency | Baseline | 1.6x reduction | 1.6x faster |
| Latency on highly parallelizable workflows | Baseline | 1.8x reduction | 1.8x faster |
| LLM reasoning + orchestration share of execution time | >90% | >90% | Identified as primary bottleneck |
| Inference latency after structured context pruning | - | ~30% reduction | Reduced overhead |
A latency decomposition after instrumenting external tool calls with realistic delays revealed that LLM reasoning and orchestration still account for more than 90% of execution time, identifying model inference as the primary system bottleneck. The paper also demonstrated that structured context pruning reduces inference latency by approximately 30%.
Robustness and Safety
Beyond speed, DynAMO maintained correct functional behaviour—including task completion, agent sequencing, and output quality—even under controlled fault injection, exhibiting graceful degradation rather than catastrophic failure. A reproducibility analysis confirmed stable execution under repeated runs, with parallel scheduling reducing latency variance.
The authors concluded that these findings establish DynAMO as a practical blueprint for scalable, safe, and latency-aware agent deployment in Industry 4.0 automation pipelines. The code for DynAMO is publicly available at the link provided in the paper.
