SkyDeck Dinner Series: Self-Improving Agents in Production
A lot of you are working on the exact problem we’re solving too: agents that learn and self-improve. We took that research into production, and we'd love to compare notes over dinner.
Kayba is a pre-seed startup of SkyDeck Batch 22. Our open-source agentic context engine (https://github.com/kayba-ai/agentic-context-engine) and hosted solution (https://kayba.ai/) let agents learn from their own traces.
The idea is simple: you bring the research perspective, we bring the production one. What do the latest ideas look like once they hit real systems? What holds up, what breaks, and what we can learn from each other’s approaches?
What to expect:
Dinner, on us
Small room: Kayba's founding team plus a handful of PhDs and grad students
An open conversation: self-improvement, context engineering, continual / lifelong learning, agent evals
Research meets production: how recent ideas translate (or don't) once they're running in the wild
Who should come:
PhDs and grad students from Sky Computing Lab and BAIR, or similar, working on continual learning, agent self-improvement, RL for agents, or evaluation.
Format:
A relaxed dinner, roughly an hour of conversation.
Bring what you're working on if you want to dig into something specific.