How To Get AI Startup Ideas

YC’s Lightcone partners (Gary Tan, Jared Friedman, Diana Hu, Harj Taggar) argue the best AI startup ideas require going to the edge of an industry — either through deep prior expertise or literal undercover work inside target industries.

• A founder built AI medical billing software by getting a real job as a medical biller and automating his own role locally with Llama 3 on two MacBooks — legally.
• Jobs already outsourced to low-wage BPOs signal prime automation targets; Lilac Labs discovered drive-thru order-taking had been offshored before building their AI replacement.
• Datacurve’s 19-year-old founder pivoted from a ChatGPT wrapper to AI data tooling by mining a prior Cohere internship; now doing mid-to-high seven figures less than a year in.
• Salient (AI auto-loan voice agents) came from a Tesla Finance Ops engineer; Diode (AI PCB co-pilot) came from founders who spanned Apple hardware and software — both cases of narrow founder-market fit nobody else had.
• YC now treats year-long idea pivots as normal, not failure; Giga ML took ~1 year before landing an enterprise deal with Zepto that is now snowballing.
• Indeed.com is an underrated idea source: search remote clerk and analyst roles to find knowledge-work jobs LLMs can automate.
• Paul Graham’s blinders concept: founders unconsciously suppress ambitious ideas because they feel too scary, so the ideas never consciously surface for evaluation.
• A large fraction of YC’s billion-dollar companies trace directly to an internship, not a full-time job — a strong signal for where students should prioritize access.

2025-02-07 · Watch on YouTube