AI Persona Lab
Runs independent synthetic perspectives over a product decision, interface, plan, or codebase before release.
The problem
One AI review often repeats one framing. Product and engineering teams need several relevant perspectives, independent findings, and visible disagreement before deciding what to change.
What I built
AI Persona Lab gives a coding agent a repeatable review workflow for an interface, feature, README, architecture, or plan. The user names the artifact, intended audience, and decision. The host runs separate reviewer perspectives, then synthesizes obstacles, disagreements, recommended changes, and questions that still require real-user research.
The repository includes three connected surfaces:
- A host workflow for running reviews in Claude Code or Codex.
- A deterministic CLI for creating briefs and preserving personas, rosters, encounters, and reports.
- A web workspace for preparing a bounded review without claiming that the web app executed the model work.
Project lineage
AI Persona Lab is the maintained successor to the archived AI User Personas experiment. It consolidates the earlier two-agent review into one reusable workflow and catalog rather than maintaining two overlapping products.
Evidence boundary
Synthetic persona findings are hypotheses, not user validation. Their quality depends on the supplied artifact, brief, reviewer selection, recall scope, and host model. The workflow keeps disagreement visible and directs consequential claims back to real users or source evidence.
Try it
Read the installation and verification guide, then ask your coding agent to review one concrete artifact from three named perspectives.