We led Letta's seed round in early 2024. I want to explain the reasoning behind that investment, because I think it illustrates something important about how we think about infrastructure bets in the AI space — and about the relationship between research and production systems.
Letta grew out of MemGPT, a research project from UC Berkeley that explored how to give LLMs access to persistent, external memory beyond what fits in a context window. The research insight was genuine: when models can read from and write to external memory stores, and when they can reason explicitly about what to move in and out of their working context, they exhibit qualitatively different behavior on long-horizon tasks.
From Research to Infrastructure
The question we ask about every research-to-production trajectory is: where does the production system diverge from the research prototype, and is the team prepared to do that engineering work? For MemGPT, the answer was clear. The research prototype demonstrated that persistent memory changed what agents could do. The production system would need to answer the hard questions the research left open: how do you handle concurrent access to memory? How do you version memory? How do you handle memory corruption or inconsistency? How do you make memory retrieval fast enough to be practical in a real product?
The Letta team had thought carefully about all of these. They were not trying to turn a research paper into a product — they were building a proper infrastructure company, starting from a research insight about what agents fundamentally need.
The Memory Gap
The broader thesis behind the investment is that stateful agent memory is a category, not a feature. The teams building production agents are all solving the same memory problem — building custom solutions, reinventing the same wheel, making the same mistakes. The right answer is a specialized infrastructure layer, analogous to what Redis did for caching or what Postgres did for relational data. Letta is the most credible team we have seen pursuing this. We are excited to be building alongside them.