Solving Self-Improving Superintelligence

The human brain is a sponge. Today’s AI brains are brittle and rigid. At Letta, we’re building self-improving artificial intelligence: creating agents that continually learn from experience and adapt over time.

Founded by the creators of MemGPT from UC Berkeley’s Sky Computing Lab (the birthplace of Spark and Ray). Backed by Jeff Dean, Clem Delangue, and pioneers across AI infrastructure. Our agents already power production systems at companies like 11x and Bilt Rewards, learning and improving every day.

We’re assembling a world-class team of researchers and engineers to solve AI’s hardest problem: making machines that can reason, remember, and learn the way humans do.

Note that this role is in-person (no hybrid), 5 days a week in downtown San Francisco.

Your role

You will develop methods for agents that self-improve after training. At Letta, you'll work with a world-class, tight-knit team of AI researchers and engineers towards our vision of self-improving superintelligence. Advance the field through open publishing of research through papers, technical reports, blog posts, and open-source code.

What you’re Responsible For:

• Developing prompt optimization and system prompt learning methods that allow agents to continuously learn from long-horizon tasks

• Developing methods allow agents automatically generate and test out hypotheses in an environment, or seek out and improve their own weaknesses

• Improving the efficiency of learning from long-running agentic tasks

• Designing and running experiments to measure self-improvement scalably

What We're Looking For:

• Expertise in machine learning, in particular LLMs and continual learning

• Familiarity with agent frameworks and prompt optimization

• Track record of impactful research (breakthrough publications and/or open-source contributions)

• Ability to balance execution speed with empirical rigor

• Real-world impact beyond pure academic work