AI Agent Catches Six Tax Software Bugs Before E-Filing

TLDW: An experienced CPA used Claude Cowork AI to complete his personal tax return (S-corp and partnership LLC) in TaxAct and found the agent caught six material errors—including one that would have blocked e-filing—demonstrating practical value of AI agents in tax preparation quality control.

Key points:

• Blake Oliver, a CPA with his own media business, tested Claude Cowork to prepare his tax returns instead of manually working through TaxAct himself
• The AI agent pulled general ledgers from both entities (one from Xero, one from another system) as a starting point
• Claude Cowork identified six errors in the prepared return, including one critical error that would have prevented successful e-filing
• The case study suggests AI agents can perform quality control functions and catch mistakes a tax professional might miss when rushing through complex multi-entity returns
• This represents a practical, real-world application of LLMs in tax preparation rather than theoretical capability demonstration