INTENT Research Series: Frontiers of Adaptive Intelligence - Continual Learning, RSI, Self-Improving Harnesses, and more

Luma Event Link Most agents are static once deployed - trained once, frozen, and brittle outside their training distribution. This session brings together researchers and builders working on systems that keep learning, decompose problems into reusable pieces, and improve their own reasoning over time. Talks are 15-20 min each, followed by open Q&A. We're covering these threads at the frontier of adaptive AI: • Continual learning — training models and agents on live production data without forgetting or drifting • Self-improvement / RSI — agents that critique, rewrite, and optimize their own prompts, tools, and harnesses • Frontier evals and benchmarks — how leaderboards break down for agents in the real world Agenda: • 6:00 PM: Doors Open, Food & Mingling • 6:15 PM: Kevin Gu, CTO of Thirdlayer, Self-Optimizing Harness & Context Systems • 6:35 PM: Ronak Malde, CEO of Trajectory, Scaling Continual Learning Algorithms • 6:50 PM: Matthew Sargent, CEO of Vmax, Stabilizing Asymmetric Self Play • 7:10 PM: Open Discussion + Q&A ―――――――――― Co-hosts & Speakers: Susa Ventures: Susa Ventures is a venture fund helping founders build the next billion-dollar company. Since 2013, we've partnered with over 100 startups – and 10% have become unicorns. ThirdLayer builds customized intelligence for operations — turning production traces across your people, decisions, and outcomes into an evolving model of your company. Our open-source library AutoAgent hit 3.8M views on X and 4.5K GitHub forks within days of launch. Trajectory is a research and product lab creating the platform for continual learning that allows companies to continuously train models and agents from production data to be faster, cheaper, and better. Used by Decagon, Clay, Harvey, Mercor, and Rogo. Vmax is a research lab focused on open-ended learning and novel ways to perform human labor. Our recent work includes population-based approaches to setter-solver games for automated training curricula, and mech interp techniques for generating intrinsic reward functions. ―――――――――― Intent Series Intent is a recurring speaker series for founders, researchers, and builders working towards human-AI interaction. Capacity is limited. Location will be shared upon RSVP approval. ―――――――――― SF Stuff To Do Do not edit this event! changes will be overwritten