lazyabm is an early-stage exploration of how AI agents can assist scientists in the messy, iterative process of scientific model development.
Scientific modeling is nonlinear and chaotic—full of partial ideas, evolving assumptions, experimental code, and constant revision. This project explores how AI agents can meaningfully support that process beyond simple code generation.
- Context tracking and assumption management
- Design handoff protocols between model development stages: design, implementation, evaluation, iteration, etc.
- Model design/implementation/evaluation assistance
- Experiment harnesses and reproducibility tooling
- CLI tools, agent skills, and agent plugins for modeling workflows
Very early stage.
Currently focused on discovery, relevant context, and ambitious ideas.