BAML by Example: Build a Self-Improving Agent

Give an agent a goal in plain language. In this hands-on workshop, you’ll use BAML to build an agent that creates the tools it needs on demand, critiques its own failures, rewrites those tools and improves across successive iterations.

You’ll then take the same loop toward production: mining real execution traces for failures and feeding them back into continuous tool improvement. By the end, you’ll have built a working self-improving agent, implemented a feedback loop based on failed runs, and connected that loop to a post-deployment improvement workflow.

You’ll leave with a working repository containing all workshop code under an open-source licence. BAML supports OpenAI, Anthropic, open-source models, Vertex AI and other model providers, so the techniques are not tied to one model.

About the presenter

Vaibhav Gupta

Vaibhav Gupta is the Founder and CEO of Boundary, the Y Combinator startup building BAML - the programming language for agents. BAML is a Turing-complete language designed for software written by agents but understood by humans, with the syntax of TypeScript, the correctness of Rust, and the compile times of Go. Across nearly a decade in software engineering, Vaibhav has built predictive pipelines at D. E. Shaw, Google, and Microsoft HoloLens. In his free time, Vaibhav dabbles in competitive table tennis and board games, and various aspects of compilers.