Join us for a conversation with Sarah Wooders, CTO & co-founder of Letta AI, as she reveals why LLMs are not there yet and how stateful agents will unlock the next generation of AI applications. From Berkeley PhD to YC alum to building the infrastructure for truly intelligent agents.
Key insights from this episode.
On the core problem with LLMs:
"It's kind of like if you've seen the movie Memento... if you have a person that just forgets every single day or forgets every five minutes... That's basically what LLMs are."
Why 2025 feels like the early Internet:
"I feel like with AI, this is our version of that. There's just so much opportunity. Everything is so undefined. There's a ton of white space... so much low hanging fruit."
The difference between real agents and marketing fluff:
"A lot of people just added LLMs into [existing workflows] and then called them an agent... that's inherently still very different from something that's like an agent that has identity, that works autonomously, that learns."
On building in the AI boom vs 2020:
"Right now is a much better time to be building a startup... the amount of value that's just being created and that you can capture is so much more."
What's still missing in AI infrastructure:
"MCP is super early... there's a ton of differences in how people implement the protocol. Auth hasn't really been figured out. There's a big lack of standardization."
Why open source matters for AI tools:
"We want to do that agentic orchestration in an open way... allowing developers to have as much control as possible over their context window, being able to see what tokens are going in and out."
Links:
Letta AI: https://letta.com
Sarah Wooders on X: https://x.com/sarahwooders
Evil Martians on X: https://x.com/evilmartians
Victoria Melnikova on X: https://x.com/vmelnikova_en
Evil Martians is the go-to agency for early-stage developer tools startups: https://evilmartians.com/devtools
#AI #Agents #LLM #Startups #OpenSource #AItooling #StatefulAgents
