We’ll build a SQL Agent from scratch using Aily’s tech stack. We’ll use PydanticAI to create the agent harness, define its behavior, prepare the right context, and connect the tools it needs to solve SQL tasks.

From the beginning, we’ll make observability part of the build process by instrumenting the agent with Langfuse so we can trace executions, inspect failures, and iterate with real visibility into what the system is doing.

The session follows the full development loop from first prototype to systematic improvement. After getting the agent working, we’ll use Langfuse again to run production-ready evals, compare iterations, and measure quality in a structured way.