The multi-model vibe shift
Real talk: back in the day, picking an AI model felt like choosing your college major. You picked one path and hoped it didn't brick your business. But at TechCrunch Disrupt 2026, experts are saying that era is mid. Instead of committing to one architecture, modern AI startups are increasingly using multiple models for different jobs, allowing for more flexibility in performance and cost.
Who's breaking it down
There are four major sessions at Disrupt, happening October 13–15 in San Francisco, that are going to deep-dive into these stacks:
- The Multi-Model World: Mo Jomaa (CapitalG), Vipul Ved Prakash (Together AI), and Zuzanna Stamirowska (Pathway) are hitting the Builder’s Stage to explain why companies are mixing and matching models rather than going steady with just one.
- To Build or Rent? Manos Koukoumidis (Oumi) is tackling the 'rent vs. build' dilemma. He’s providing a framework for when it’s actually worth the headache to build your own model versus just calling on a frontier API.
- The VC Perspective: Nvidia’s Nader Khalil and Sydney Sykes will break down what startups should actually be betting on, looking at the trade-offs between frontier APIs and open-weight models.
- AI Designing Hardware: Anna Goldie and Azalia Mirhoseini (Ricursive Intelligence) are getting meta. They’re exploring how AI is literally starting to design the chips that power it, a move that could speed up how fast new tech hits the market.
Why it matters
This isn't just nerdy architecture talk—it’s about survival. Founders have to balance costs, talent, and control. If you lock into one model too early, you might get left behind when something better drops next week. Staying flexible isn't just a strategy; it’s basically the main character energy these startups need to survive the current AI gold rush. For those attending, it’s all about learning how to keep your stack agile before the industry moves on without you.
