Scaling domain expertise in complex, regulated domains

Source: OpenAI Date: 2025-08-21 URL: https://openai.com/index/blue-j

Summary

Case study featuring Blue J, a Canadian tax law AI company, on deploying GPT-5 in a heavily regulated professional domain. Blue J builds AI tools for tax professionals that need to reason about complex, jurisdiction-specific legal text with high accuracy requirements and audit trails. The post likely covers how Blue J achieves reliable answers in a domain where hallucination is legally consequential.

Implications

Platform thread. Blue J is among the more technically sophisticated professional AI deployments — tax law is adversarial (regulators look for errors), jurisdiction-specific (can’t train away edge cases), and requires explainable reasoning (professionals need to cite authority). Success in tax law is a template for other regulated professional domains (medical coding, patent law, financial compliance). The “scaling domain expertise” framing positions fine-tuning and RAG over prompting alone.