Adversarial model and critiques

Getting sharp, in-character adversarial critique of code, documents, or leadership communication — manually soliciting feedback from domain experts is slow and inconsistent

User submits code, document, or communication and names a critic persona (Andrej Karpathy, Sam Altman, Yann LeCun, Fei-Fei Li, or custom). Load the persona's system prompt, pass the content through Claude with the persona context, and generate sharp, specific feedback in that critic's intellectual style. Can run as one-off critiques or scheduled daily digests in email/Slack.

Get consistent, high-signal adversarial feedback grounded in a specific thinker's public work and style — no generic feedback, always specific to what was submitted

System prompt template: {PERSONA_PROMPT} + content type + user submission. Five built-in personas with detailed trait-based prompts trained on their public writing (essays, Twitter, talks, books). Custom personas can be added by grounding traits in public record and capturing their signature tone and phrases. Scheduled digest mode outputs structured format with observations tied to specific files and one concrete action item.

Personas are only as good as their public record — requires accurate sourcing to avoid inventing characteristics. Custom personas need careful grounding to avoid drift from actual person's intellectual style.