When the One Big Beautiful Bill arrived as a massive, unstructured 900-page document lacking standardized format and official IRS forms, Intuit’s TurboTax team faced a pressing challenge: could AI shorten a traditionally months-long tax code implementation into mere hours without losing accuracy?
The solution was a groundbreaking workflow combining commercial AI, a proprietary domain-specific language, and an innovative unit test framework, designed not just for tax but for any team managing complex regulated domains.
Joy Shaw, Intuit’s tax director with decades of experience, shared how AI rapidly distilled the law’s noisy details into actionable insights, enabling coding to start ahead of official forms. They leveraged large language models to analyze and reconcile House and Senate bill versions, reducing weeks of work to hours.
While general AI tools handled document interpretation, Claude, a specialized AI, translated legal text into Intuit’s unique coding language and highlighted only the changes needing developer focus. This selective approach accelerated development.
Two proprietary tools ensured near-perfect accuracy: one auto-generated application screens from law changes, and another provided detailed unit testing that pinpointed errors and allowed in-framework corrections.
Intuit emphasized the importance of verified, deterministic code to meet consumer product accuracy standards, supported by human experts validating AI output.
This workflow’s core elements—using commercial AI for analysis, domain-aware coding AI for implementation, custom evaluation tools, and organization-wide AI fluency—offer a blueprint applicable to sectors like healthcare, finance, legal tech, and government contracting facing similarly complex regulations and tight deadlines.
Sarah Aerni, Intuit’s Consumer Group tech VP, remarked on blending AI with human expertise to deliver reliable consumer experiences.
