Answers are grounded in this paper and cite their sources.
When we envision the landscape of combustion in 2050, the central question is not merely what machines can do for us, but how the role of the combustion scientist must evolve to address the outstanding challenges that continue to define combustion science and technology. This topical review examines the transition from machine learning (ML) as a high-fidelity toolkit for subsystem modelling to artificial intelligence (AI) as a strategic collaborator in scientific discovery. We begin by examining the current state of ML in the field, focusing on its ability to accelerate stiff chemical kinetics, reduce dimensionality, augment sub-grid closures for turbulent reacting flows, and reconstruct sparse experimental data. Using a bibliometric corpus of 2647 records, we show that combustion research already has a broad data-driven footprint, but that the literature remains dominated by a priori studies and method development, with limited embedded, operational, and decision-support use. We then transition to the emerging role of AI as an autonomous agent, highlighting recent breakthroughs in “Agentic AI,” where multi-agent systems coordinate complex software and navigate design spaces for solving combustion problems. This evolution calls for a broader rethinking of scientific practice in combustion. We define a “Human-in-the-Loop” framework to describe how combustion scientists and AI systems may interact in future workflows for discovery, modelling, and design. We argue that the future combustion scientist will increasingly focus on formulating physical constraints, evaluating model behaviour, and guiding exploration across design spaces that exceed the reach of traditional intuition or classical simulation alone. Novelty and significance statement: This topical review presents a perspective on the transition from data-driven modelling to agentic AI in combustion science. It discusses how machine-learning methods are already used for chemical kinetics, closures, reduced-order modelling, diagnostics, control, and digital twins, and uses a bibliometric corpus to assess how far these approaches have moved towards operational deployment. The novelty lies in linking this evidence to the emerging concept of the augmented combustion scientist: a researcher who works with AI systems that can organise actions, use tools, and navigate complex modelling and design spaces. The significance of the review is to clarify where agentic AI may add value beyond current data-driven methods, while emphasising the continuing need for physical constraints, validation, traceability, and human judgement.
No takes yet. Share an insight, caveat, or question.
Parente et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: