By @AnthropicAI
Publication Date: 2026-03-24 12:00:00
Written by Prithvi Rajasekaran, a member of our Labs team.
Over the past few months I’ve been working on two related problems: getting Claude to create high-quality frontend designs and getting Claude to build complete applications without human intervention. This work grew out of previous efforts on our front-end design capabilities and the long-standing coding agent harness, where my colleagues and I were able to improve Claude’s performance well beyond baseline through rapid engineering and harness design – but both eventually reached their limits.
To break through, I looked for novel AI engineering approaches that span two entirely different domains, one defined by subjective taste and the other by verifiable correctness and usability. Inspired by Generative Adversarial Networks (GANs), I designed a multi-agent structure using a generator And Reviewer Agent. In order to build an evaluator that evaluates results reliably and with taste, it was first necessary to develop a set of criteria that…
