We all become seniors: Agentic workflows for code dev
by
OVGA/200
By now, agentic coding assistants are used heavily in daily work in the lab. The output quality of these assistants has reached a point where most code by the AiiDA team is now written by an agent. These agents still require clear instructions on what code to write and in which style, so that contributions are consistent with the code base requirements. At the same time, the quality of a language model's output degrades as the context window fills up with information. This has led to the development of "harnesses" for language models that structure what information the model, how, and when. Because code is now so cheap to generate, the quality assurance process for introducing new code into an existing codebase needs to be adapted to these new constraints. In this talk we cover techniques in agentic engineering to build workflows that produce code aligned with ones intent.
Laboratory for Materials Simulations (LMS)