Notes from building AI agents and researching how they can be manipulated.
7 min read · 23 Nov 2026
2,850 simulated negotiations, five models, twenty persuasion tactics, $24 in API costs. Agents caught the tactics they could name, but not a high opening price, which raised what they offered on every model I tested.
9 min read · 23 Nov 2026
Nine language models each ran a $10,000 paper portfolio through the same research pipeline: 62 candidate positions, 186 analyst reports, $10.95 in API costs. The field beat the S&P 500, but one bet did almost all the work, and the only analyst whose scores tracked returns was the bear.