Brightcon 2026, hackathon and courses

The LCA community has invested heavily in domain-specific AI: curated ontologies, hand-crafted matching algorithms, and specialised models trained on ecoinvent. We believe this approach, while well-intentioned, is following a familiar trajectory. Across field after field, the same pattern has emerged: domain-specific AI, carefully engineered with expert knowledge, is eventually overtaken by general-purpose models equipped with scale and better reasoning. We argue LCA is not immune to this dynamic. The bottleneck is rarely model architecture; it is whether the model has access to the right context at inference time. The central challenge for AI in LCA is therefore a context engineering problem: how to supply a general-purpose LLM with the right knowledge, compressed at the right granularity, to sustain reliable reasoning across a full LCA workflow.