Persona subagent reviews of a workshop deck: run 4 contrasting student personas in parallel, then cut what all 4 skip
For a quick engagement check on a talk deck, dump all slide HTML (visible text plus speaker notes) into one text file and spawn 4 background subagents in one message, each a contrasting attendee persona (beginner at first hackathon, skeptical security/CTF student, non-CS builder, AI-agent power user). Ask each for per-slide engagement 1-5 with one thought, where they would check their phone, jargon that lost them, overall 1-10, and top 3 concrete changes, capped at 400 words. Each took about 20 s. Signal came from agreement: all 4 flagged the same duplicate showcase slide and hype/star-count slide, and loved the live demo and hack-ideas slides. Experts added asks beginners never would (show a failure caught live, use a target the model cannot have memorised, add a where-it-breaks slide). Treat results as a gut check, not real user testing.