This chapter defines translation agents as workflow‑driven, modular systems that supersede single‑prompt large language model usage for complex, domain‑sensitive translation. It presents a layered architecture (task awareness, context management, translation execution, terminology management, quality evaluation, feedback interface) that enables strategic decomposition, controlled information flow, and adaptive refinement. Core design principles include semantically informed segmentation, dynamic context bridging, multi‑strategy generation, and orchestration of primary and auxiliary models for terminology, style, and structural validation. The chapter details task chains covering ingestion, structural decomposition, style/register detection, summarization‑assisted comprehension, terminology extraction and active termbase integration, long sentence simplification, multi‑model generation patterns, and document‑level consistency mechanisms. Iterative optimization—drafting, diagnostic evaluation, variant comparison, back translation, and targeted revision—is framed as an autonomous feedback loop grounded in established quality frameworks (Multidimensional Quality Metrics, Dynamic Quality Framework) while recognizing limits of legacy reference metrics. Tooling ecosystems (e.g., LangChain, Haystack, AutoGen, emerging low‑code platforms) are assessed for orchestration, memory, and monitoring support. The chapter argues that translation agents operationalize a shift from ad hoc prompting to traceable, optimizable pipelines, advancing scalability, explainability, and human–machine collaboration toward adaptive, knowledge‑integrated language services.