Over the past year, artificial intelligence has fundamentally reshaped discovery in mathematics and theoretical physics, transitioning from academic skepticism to a proven catalyst for human ingenuity. Rather than displacing researchers, AI systems are augmenting analytical capabilities, accelerating discovery cycles, and uncovering patterns previously inaccessible to human scrutiny. Recent developments demonstrate AI expanding utility across theoretical domains. Software tools now execute line-by-line proof verification, identify counterexamples to unproven conjectures, and propose intermediate logical steps to bridge known and unknown mathematical territories. Unlike experimental sciences constrained by physical limits and data noise, theoretical research benefits from digital, reproducible datasets. This environment has enabled several AI systems to achieve significant milestones. Aristotle, developed by Harmonic in Palo Alto, California, has resolved longstanding problems originally posed by mathematician Paul Erdős. Axiom Math, a Palo Alto startup, has independently solved multiple research-level problems previously unsolved by human mathematicians. Concurrently, models from OpenAI in San Francisco and Google DeepMind in London have successfully addressed challenges from the First Proof Project, demonstrating capacity for generating novel, verifiable results. The integration of AI into the scientific workflow maps across three primary phases. Agenda setting remains predominantly human-driven, as AI currently lacks the contextual intuition required to identify timely research questions. However, emerging systems are being trained to scan databases and preprint repositories, such as arXiv and the On-Line Encyclopedia of Integer Sequences, to prioritize overlooked connections and suggest promising directions. Formalization, the translation of informal reasoning into machine-readable logic, is undergoing rapid advancement. Projects like the Xena initiative at Imperial College London are systematically digitizing undergraduate proofs, while researchers at Chalmers University in Gothenburg, Sweden, are leveraging large language models to formalize complex topology theorems. This process exposes latent logical gaps, as demonstrated when Terence Tao at UCLA verified his own work using the Lean4 proof assistant, and builds higher-quality datasets for future model training. Conjecture generation has evolved from experimental algorithms to practical research tools. Historical systems like Graffiti and the Ramanujan Machine pioneered pattern discovery, while contemporary AI applies similar methodologies to theoretical physics, uncovering hidden symmetries and exact formulae that accelerate breakthroughs. The cumulative impact is a substantial reduction in routine analytical labor, allowing researchers to focus on conceptual design. AI systems are executing tedious verification, systematic data exploration, and hypothesis generation with unprecedented speed. Nevertheless, the deepest theoretical advances still require human creativity and judgment. The emerging paradigm is collaboration rather than replacement. As formalization pipelines mature and systems gain broader scientific context, the synergy between machine computation and human insight will likely redefine the pace of discovery across mathematics and physics. Researchers are increasingly encouraged to integrate these tools, recognizing that computational augmentation represents the near-term reality of theoretical science.