
What happened
Legora used GPT-6 Astra to review 41 financial documents in a single run, completing the work in minutes instead of an evening or days. The agent found all four planted errors, including a £500,000 gap in a revenue note.
Why it matters
Legora reports GPT-6 Astra improved performance by nearly 40% over the previous model on this financial-statement tie-out workflow. The agent also completed about 50 more checks, improving accuracy and completeness while keeping the human expert in charge of final decisions.
What to watch
Legora is expanding beyond legal work into audit, tax, compliance, and risk. The success of this approach hinges on how well the human-in-the-loop model scales to these new workflows.
Summaries like this, in your inbox every morning.
Legora's deployment of GPT-6 Astra tackles a specific pain point in legal and financial work: the painstaking reconciliation of financial statements. This task, previously consuming evenings or days, is now compressed into minutes, with the AI agent checking balances and flagging discrepancies. The improvement in performance, nearly 40% for this workflow, stands out against a modest 3% average gain across all tasks, suggesting that the benefit is most pronounced for tasks requiring large-scale document ingestion and complex cross-referencing.
The company's strategy is to extend its agentic operating system beyond legal work into audit, tax, compliance, and risk. The results highlight the value of a human-in-the-loop model, where the AI handles the exhaustive comparison and the human expert retains the final judgment. This approach could define how professional services firms integrate AI, potentially shifting the focus from automation to augmentation of expert work.
Looking ahead, the broader impact hinges on whether this performance holds across other document-heavy workflows and industries. The nearly 40% improvement is promising, but the modest average gain across all tasks suggests that returns vary. Firms will need to identify where AI can deliver the most value and how to retain the trust and accountability that human experts provide.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Action State Group Inc. launched open-source EvidenceBook, Terms Packs and Countersign so companies can cross-…
Egis Group and SiPearl formally signed a strategic CPU partnership at the Egis Group TechDay Taiwan-France cer…

Startups Electric Twin, Artificial Societies, GWI and Toluna are using data from real focus groups and surveys…

DeepSeek raised around $15 billion in one of China's biggest-ever financing rounds, after twice increasing its…

US stock futures and Asian markets fell on fears the AI boom is turning to bust, as an Australian data center…

Meta is creating a new class of nonvoting stock, first reported by Semafor, so Zuckerberg can issue shares for…
