Claude Code Psychosis: How SemiAnalysis Is Token Mogging Meta | Ep. 008

Summary

The SemiAnalysis team discusses their extensive deployment of Claude Code AI agents for financial research, modeling, and conference analysis, demonstrating a significant productivity edge. They explore the trust, reliability, and cybersecurity implications of AI-generated analysis and consider the broader market adoption and competitive landscape for AI models.

• Detailed demonstration of an agent swarm called 'Wags' used to initiate coverage on companies like AOI, automating financial modeling and data ingestion.
• Discussion of the cost-benefit analysis: spending a few dollars on agents versus days of analyst work, highlighting high ROI.
• Exploration of how AI agents transform the role of interns and analysts, shifting focus from data entry to analysis and review.
• Insights into using AI for conference analysis, indexing thousands of presentations to enable targeted research queries.
• Consideration of trust and reliability in AI outputs, noting a shift towards default trust with verification for critical errors.
• Debate on the pace of broader AI adoption in large enterprises, citing cultural and incentive barriers despite clear productivity gains.
• Analysis of recent AI model releases, including Anthropic's Mythos and Meta's model, touching on cybersecurity capabilities and competitive dynamics.
• Speculation on the future of AI model differentiation, focusing on ecosystem, user experience, and persistence in long-range tasks.