AutoSurrogate: An LLM-Driven Multi-Agent Framework for Autonomous Construction of Deep Learning Surrogate Models in Subsurface Flow

AutoSurrogate is a classic academic reference implementation (0 stars, 4 days old) that applies the currently popular multi-agent LLM paradigm to a high-value niche: subsurface flow simulation (oil, gas, carbon storage). While the problem it solves is technically difficult and domain-specific, the project itself lacks a moat. The defensibility score of 2 reflects its status as a brand-new research prototype with no community or ecosystem. Its primary value is demonstrating how agents can orchestrate ML pipelines, but it is highly vulnerable to 'Agentic Scientist' tools (like Sakana AI's 'The AI Scientist') or simply the rapid improvement of frontier models in code generation and data science reasoning. Frontier risk is medium because while OpenAI won't build a 'subsurface flow tool,' they will build the underlying reasoning capabilities that make this project's custom agent logic redundant. In the commercial sector, incumbents like SLB (Schlumberger) or Halliburton are more likely to integrate these capabilities into their existing proprietary simulation suites than adopt an open-source framework with no established traction. The 1-2 year displacement horizon is driven by the speed at which generalized AI coding agents are becoming capable of handling domain-specific libraries like Firedrake or Open Porous Media (OPM).