Translating long-form technical volumes remains a persistent challenge in machine translation, requiring not only semantic fidelity but also discourse coherence, consistent terminology, and preservation of structural elements such as figures and equations. This paper introduces Agentic Machine Translation (AMT), a framework that treats long-form volume translation as a coordinated multi-stage process rather than a single-pass task. AMT integrates layout-aware preprocessing with OCR, segment-wise translation, dual-signal quality evaluation combining embeddings and LLM-based scoring, and document reconstruction with multimodal reintegration. To evaluate the approach, we curate a domain-diverse dataset and conduct both automatic and human assessments. Results show that AMT achieves consistently higher similarity scores and human ratings than strong baselines, with clear advantages in structurally complex and extensive technical volumes. While challenges remain in handling specialized terminology and reducing computational overhead, the findings demonstrate the promise of agentic strategies for advancing the translation of extensive technical volumes.