VC's experiment: ChatGPT and Claude flip-flop on retirement investing, showing why vertical AI wins
MartinGTobias · x · 2026-10-07
Investor Martin Tobias argues vertical AI beats general LLMs based on his own experiment: asking ChatGPT and Claude for alternatives to VGT in his IRA, he found both models gave wildly different answers over one conversation, shifting with what they thought he wanted to hear.
His core points:
- General LLMs are pattern matchers that optimize to please users rather than deliver hard truths
- They optimize for "best fit" to the most recent context, so positions drift in long conversations
- Retirement investing has proven frameworks, backtested strategies, and objectively good/bad answers — a vertical agent with a harness built around that decision process would do far better than general chat
He says he's talking to teams building vertical investment agents today.
More from coding & agent
- LLMs lack 'idempotency': rechecking their own large codebases yields contradictions — StephanSturges · 2026-10-07
- Open-source skill turns podcast transcripts into publishable articles, cutting 35% with zero detail loss — oran_ge · 2026-10-07
- Epic Games' raddebugger, a native multi-process graphical debugger, hits 7.7k stars — EpicGames · 2026-10-07
- cmux, a Ghostty-based macOS terminal built for AI coding agents, nears 28k stars — manaflow-ai · 2026-10-07
- Should AI coding reuse open source or start from scratch? Weaviate Podcast explores — CShorten30 · 2026-10-07
- 44 prompt-injection runs, zero bypasses: CLIM Agent Guard enforces a deterministic tool-execution boundary — Slight_Analysis_5414 · 2026-10-07