A Mercor study finds AI models now exceed licensed accountants in speed and accuracy on structured accounting work. However, no current AI system can independently complete comprehensive accounting tasks without human supervision.
AI has crossed a significant threshold in accounting work. According to recent research from Mercor, current large language models now outperform licensed CPAs on straightforward accounting tasks in both speed and execution quality. This represents substantial progress—18 months ago, the same models lagged considerably behind human accountants.
The advancement is noteworthy for routine, structured work where inputs are standardized and outputs are clearly defined. For businesses handling high volumes of basic accounting operations, AI systems now offer tangible efficiency gains.
However, the picture changes dramatically with more complex scenarios. When tested on the APEX Benchmark—which measures performance on demanding, real-world accounting challenges—no AI model successfully completes all required tasks. The gap between handling isolated structured problems and managing full accounting cycles remains substantial.
Closing the books requires judgment calls, handling exceptions, and integrating multiple data sources in ways that current AI cannot manage independently. Tasks involving ambiguous situations, regulatory judgment, or complex reconciliation still demand human expertise and oversight.
For accounting firms and in-house finance teams, the implications are mixed. AI excels at accelerating routine work and reducing human effort on repetitive tasks. This could free accountants to focus on analysis, advisory work, and complex problem-solving.
Yet complete automation of accounting functions remains out of reach. Any organization deploying AI for accounting must maintain human oversight, particularly for high-stakes decisions like financial statement preparation and regulatory compliance.
The trajectory is clear: AI will continue encroaching on accounting tasks. But the profession's core value—exercising judgment on complex financial matters—remains distinctly human for now. The future likely involves AI handling volume while humans manage complexity.
Customers armed with AI-generated information are increasingly dismissing server warnings about food allergens, creating dangerous situations in restaurants across the country.
Suno, an AI music creation platform, has launched a public beta feature that generates spoken voices from scripts or descriptions. The new capability allows users to create voiceovers and background music simultaneously.
Experts warn that artificial intelligence could amplify disinformation campaigns during elections across Southeast Asia, a region with young, digitally connected populations vulnerable to manipulation.
DeepSeek has released Harness Desktop, bringing its AI capabilities to macOS and Windows users. The application provides local access to DeepSeek's models across both platforms.