A Model For Everyone: How Ronak From Trajectory Is Building Products For Continual Learning

By Sequoia Capital

Categories: VC, Startup

Summary

Ronak from Trajectory argues that true continual learning requires connecting architectural innovations (like DeepSeek's fast/slow memory) to real-world product loops—currently only labs can do this, requiring $100M+ R&D and 6-9 months. The opportunity: democratizing model training so every company can update models from live production data, not batch updates.

Key Takeaways

  1. Current continual learning is 'pseudo'—companies must do expensive batch updates to retrain models. Real continual learning works like Google Maps: every user interaction updates the model in aggregate, requiring architectural changes already happening in frontier models (DeepSeek v4.1, skip connections, memory layers).
  2. The missing layer is translating 'what good looks like' in companies' minds into actual model improvements. This requires connecting data-to-intelligence in production, not just labs—the core product opportunity at Trajectory.
  3. Current vertical AI companies (Windsurf, Cursor) needed 6-9 months, hundreds of millions in R&D compute, and large research teams to train frontier models. This gatekeeping is what needs to be democratized for continual learning to scale.
  4. Apple's product philosophy applies here: combine training, eval, and deployment in one place, making it 'dead simple' for every person. Close the loop so new production traces automatically upload models back, creating the virtuous cycle.
  5. Three-stage maturity model for continual learning: AAA-era (research labs only) → MapQuest (democratized offline training) → Google Maps (real-time production updates). We're still mostly in stage 2 waiting for architectural support.

Related topics

Transcript Excerpt

A lot of our philosophy in product actually comes from Apple. In like the early 2000s, you had a computer that had access to the internet and you had a phone that you could make phone calls. And I think no one really thought to put the two together until Apple said, "Okay, what if you actually put the internet on a handheld device and combine the two and I think by doing this of of basically putting training, eval, and deployment in one place, making that just dead simple for every single person to do. And then you can close the loop and basically now upload that model back to production, gather new traces. It suddenly is this magical paradigm that I think every single company is going to adopt very soon. [music] [music] Reach, thank you for joining us. It's a pleasure to have you. We woul…

More from Sequoia Capital