How we rebuilt Sigma’s AI agent to be cheaper and faster for customers

Sigma rebuilt its Assistant build mode from a hierarchy of specialized subagents into a Python coding agent that uses a typed SDK to create and edit data-analysis workbooks through natural-language requests. The previous system delegated tasks such as formulas, filters, charts, and formatting to separate agents, which could lose context between handoffs and produce inconsistent interpretations, while the new system executes connected operations in a single program using reusable objects, type checking, and runtime validation. In internal evaluations covering 130 tasks across 30 suites, the new approach built complete applications three times faster at one-third the cost and improved accuracy from 71% to 86%, while incremental edits were 17% faster, 43% cheaper, and four percentage points more accurate. Its design emphasizes fewer LLM calls through program execution, progressive disclosure of SDK documentation to reduce token use, typed interfaces that clarify available capabilities, rapid error-feedback loops, and example-based coding skills. Sigma says the updated build mode, now generally available, produces more consistent and visually polished dashboards and apps while reducing customers’ AI-provider token spending.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.