AI Tulips (mamund.substack.com)

"AI Tulips" uses 17th‑century Tulip Mania as a lens to diagnose today’s AI hype: explosive adoption (ChatGPT hit 1M users in five days and 100M+ by early 2023), rapid model iteration (GPT‑3 → GPT‑4 → talk of GPT‑5/6), a new high‑paying “prompt engineer” role, and massive funding ($56B in generative AI in 2024) that pushed valuations up ~200% for late‑stage AI startups. The piece flags symptom‑pulling behavior—FOMO at scale (e.g., PwC rolling out OpenAI tools to 100k employees), loose definitions like “agentic AI,” guru‑led courses, and VC froth—arguing the market dynamics look less like rational adoption and more like speculative mania. For practitioners and builders the essay stresses concrete technical and operational risks: hallucinations that create legal exposure (fake citations submitted to court), inadvertent data leaks (real incidents at companies like Samsung), brittle integrations that compound technical debt, and perverse incentives that reward buzz over outcomes. The prescription is pragmatic: treat generative models as tooling, not magic; validate on small, well‑defined problems; prioritize measurable behaviour over slick demos; avoid one‑size‑fits‑all platforms that hide complexity; and consider the second‑mover advantage—let early adopters surface the bugs while you focus on reliable, maintainable solutions. The bottom line: AI’s capabilities are real, but success depends on disciplined engineering, clear problem framing, and skepticism about marketing claims.