AlloyGen graphical abstract: expert agents, flexible collaboration, and adaptable team organization connected to alloy-design workflows.
AlloyGen graphical abstract from the authors’ code repository. Reproduced without changing the figure; linked preprint, version 1, posted August 4, 2026.

The central idea

AlloyGen connects materials knowledge, executable thermodynamic calculations, critique, and task-specific teams. It studies adaptation at the level of an individual agent, a collaborating team, and the organization of that team.

What the preprint demonstrates

An expert agent completes six CALPHAD exercises with retrieval and execution feedback. Further demonstrations turn an underspecified Al–Si printability goal into a workflow and assemble a team for refractory-alloy research synthesis.

My contribution

Bo Ni contributed to conceptualization, investigation, methodology, data curation, and validation, as stated in the preprint’s author-contribution record.

Scope of the result

These are computational demonstrations of adaptive workflows. They do not establish an experimentally validated new alloy or autonomous control of a manufacturing process. This work is a preprint.