Search papers, labs, and topics across Lattice.
AgentSWE is a novel benchmark designed to evaluate the fidelity of coding agents in building other agents, addressing the critical gap between demonstrated behavior and actual requirement fulfillment. The benchmark comprises 25 tasks across agent creation, editing, and optimization, each assessed via acceptance exercises on held-out cases against requirement-derived rubrics. Evaluations reveal that current coding agents consistently fail to deliver on user requirements, exhibiting significant gaps between their demonstrated capabilities and their final submissions.
Coding agents routinely fail to build agents that truly meet user requirements, with the gap between demonstration and delivery being the norm.
As the production of software passes into the hands of software itself, coding agents are increasingly commissioned to build agents fitted to their users'own workflows. This delegation carries a quiet assumption, that the delivered agent will faithfully serve the requirement it was built for, rather than merely appear to. No existing benchmark tests that assumption. We introduce AgentSWE, a benchmark for agent software engineering spanning three stages of the agent development lifecycle: Creation of a complete agent from a natural-language requirement, Editing of production agent codebases, and Optimization toward a measurable target. Each of its 25 tasks is framed as a commission rather than an exam: the builder iterates against a handful of illustrative cases, while acceptance exercises the frozen submission on held-out cases designed to span the full requirement, grading behavior against requirement-derived rubrics under hard pass/fail gates or, for Optimization, against the source benchmark's own objective. Evaluating a range of coding agents across models and harnesses, we find that none comes close to delivering what a requirement asks for: the gap between what a builder demonstrates and what it delivers is the norm, not the corner case. An accompanying empirical analysis of the failure trajectories identifies recurring failure modes, among them overfitting to the visible cases, hollow runtimes behind code that looks complete, and unsupported claims of completion, each a signal by which coding agents can learn to build the agents they are asked for.
