Almost every YC AI company alive today repositioned at least once between November 2022 and mid-2023. The AI Founder Pack tracks every YC AI company across that fault line. The split between 'rebuilt on top of the new model' and 'kept fighting the old roadmap' was decisive.
What repositioning looked like
| Company type | Pre-ChatGPT pitch | Post-ChatGPT pitch | Outcome |
|---|---|---|---|
| Customer support AI | Trained-from-scratch intent classifier | GPT wrapper + RAG over docs | Multiple unicorns |
| Sales tooling | Custom NLP scoring engine | LLM-generated outbound + analytics | Survived |
| Writing assistant | Fine-tuned small model | Thin GPT wrapper, lost moat | Acqui-hired or shut |
| Legal AI | Document classification | Drafting + review + agent workflows | Series A+ |
| Voice | Custom ASR pipelines | Realtime API + agent orchestration | New unicorn cohort |
The 10-day pattern in repositioning
- Day 0–2: founder uses ChatGPT on the team's own use case.
- Day 3–5: prototype rebuilt on top of the API.
- Day 6–8: shown to 5 existing customers; pricing rethought.
- Day 9–10: roadmap and team meeting; old roadmap killed.
The companies that did NOT survive
Across the pack, the companies that didn't survive shared one thing: a year-long custom model roadmap that the founders felt protective of. The protective instinct is the killer — sunk-cost thinking turns six months of model training into a moat that no longer exists.
Key takeaways
- Survivors rebuilt on top of foundation models within ~10 days.
- The killer pattern was sunk-cost protectionism of custom models.
- Voice + legal + ops AI emerged as the strongest post-ChatGPT cohorts.
- Thin wrappers without a workflow moat got acqui-hired or shut.