The standard output of generative models is regression to the mean. Without strict constraints, they produce the same layouts, the same predictable colour ramps and the same synthetic aesthetic. The web is filling with identical artefacts devoid of intent.
Artificial Human Design (AHD) is an evaluation framework that enforces mechanical constraints on generation. It applies rigorous rulesets, runs linters over output and scores designs against human-authored tokens. It is not an agent. It is a defence mechanism.
ahd eval-live swiss-editorial --brief briefs/landing.yml --models <specs> --n 10
What ships
- Brief compiler
- 28-rule linter
- Eval harness
- Vision-critic scaffold
- MCP server
- Editor plugins
- Eight tokens
What is gated
- Live frontier-model calls (needs API keys)
- Live vision critique (needs multimodal key + screenshot pipeline)
- Standalone npm packages for the editor plugins
- Additional tokens