We explore how artificial intelligence is reshaping computational science. We contrast the scaling paradigm—bigger machines and faster algorithms—with a potential evolution paradigm driven by AI agents that reason, simulate, and interpret results autonomously. As AI begins to write code, design experiments, and analyze data, the role of the computational scientist may shift from executor to orchestrator. We also present JutulGPT, an AI agent integrated with a differentiable reservoir simulator, illustrating both the promise and pitfalls of agentic simulation. Key lessons include the need for hybrid human–AI workflows, better documentation, and transparent validation. Ultimately, the field must confront methodological, educational, and ethical challenges to ensure that computational scientists shape AI’s impact—rather than being shaped by it.