Ask These Questions Before Starting An AI Startup
Jordan Fisher (Anthropic alignment team, ex-Standard AI) argues founders must plan for AGI in 2-3 years and that trust, not technology, is the defining startup problem of this era.
- Fisher predicts AGI in 2-3 years and says founders not planning around it are failing their job.
- Enterprise buy-side will also get AI-powered, potentially eliminating SaaS sales cycles and building software in-house via tools like Claude Code.
- As startup teams shrink toward semi-automation, human whistleblower checks vanish — a single bad actor can redirect an entire product with no internal friction.
- AI-powered audits (with no memory, no IP leakage) could replace human auditors and become a binding trust mechanism for companies.
- Tacit knowledge at firms like TSMC and ASML is not on the internet — frontier LLMs cannot build a cutting-edge semiconductor fab, making proprietary data defensible in specific verticals.
- Alignment progress is being driven by economic pressure: agents trusted to run for days or weeks without human review require it, not just safety research.
- Most VCs claiming to be ahead on AI are already 2 years behind; Fisher’s thesis is to invest or build for what is defensible in 2 years, not what is hot today.
- Users in-the-moment prefer sycophantic AI responses, but when asked which principle they want — honesty or flattery — almost everyone chooses honesty.
2025-10-07 · Watch on YouTube
