Deep research agents answer complex questions by looping through search, reading, and reasoning. Recent results on reasoning‑heavy benchmarks such as BrowseComp-Plus show that well‑tuned lexical retrieval can surface high‑quality evidence, yet agents often fail to link those documents to the gold references. We identify the agent's first retrieval move as a pivotal design choice.\ \ Question's Gambit is a first‑move retrieval module. It first breaks the question into a set of clues, reformulates each clue into complementary search queries, merges and deduplicates the retrieved passages, and reranks the pool to create an opening context. This context supports both clue aggregation and final‑answer verification before the agent enters its iterative search‑and‑reasoning loop.\ \ Experiments on BrowseComp-Plus demonstrate that Question's Gambit improves retrieval recall and downstream agent accuracy over strong baselines. With gpt-5.5, answer accuracy rises from 83.1% to 90.5%, surpassing the previously strongest agentic baseline Pi‑Serini. Additional evaluation on MultiHop‑RAG confirms that the benefits transfer to conventional multi‑hop question formats.\ \ The results confirm that effective agentic deep research depends not only on the tools inside the loop but also on the quality of the first move. Our implementation is publicly available at https://github.com/radinhamidi/Question-s-Gambit.\ \ Review