​Overview

​A practical engineering session on what goes into a reliable coding agent harness for data workflows, how data tasks differ from conventional software engineering tasks, and how tools, agent skills, semantic context, and trace-level analysis can improve agent performance.

​Session Details

​Coding agents are often designed around software engineering workflows: navigating repositories, editing files, running tests, and fixing code. Data tasks introduce a different set of requirements. The agent must understand schemas, business definitions, relationships between datasets, available tools, and the context needed to interpret results correctly.

​In this session, with Josh from Snowflake - we’ll break down the components of an agent harness for data workflows and demonstrate how agents can use tools, reusable skills, and semantic context to complete tasks more reliably. We’ll also examine execution traces to identify where errors occur and how those findings can be used to improve the agent, its context, and the surrounding harness.

​What you'll learn

• ​The core components of a coding agent harness

• ​Why data tasks require a different approach from software engineering tasks

• ​Using tools and agent skills to execute multi-step data workflows

• ​Building semantic context around schemas, metrics, and business definitions

• ​Using trace-level error analysis to identify and fix agent failures

​Who Should Join

​This session is for AI engineers, data engineers, analytics engineers, platform engineers, and technical leaders building coding agents for data analysis, data operations, internal tooling, or other data-intensive workflows.

​About Future AGI

​Future AGI is an open-source self-improving layer for agentic systems — it lets teams simulate, evaluate, guard, observe, and automatically improve their AI in one closed loop, so agents get more accurate and safer over time instead of silently breaking in production.