AI Now Beats Experienced Accountants on Close Tasks. What Is Left for Humans?
In a Mercor test, a top AI model outscored 12 CPAs on month-end close work. The bigger shift may be that AI cuts the price of tasks before it ends any job.
In early October 2026, Mercor, an AI talent and evaluation company, published an uncomfortable experiment. It gave 12 US certified public accountants four tasks tied to month-end close. The participants averaged about five and a half years of experience. Half had worked at a Big Four firm.
The tasks required digging through multiple files to find figures, correcting misclassified items, and adjusting transaction timing to fit accrual accounting. The human accountants averaged about 37%. Claude Opus 5 ran the same tasks 20 times and scored 100 all 20 times, finishing in under 10 minutes each time.
- 37%: average accuracy of the human accountants
- 100%: Claude Opus 5, 20 runs in a row
- 49x: cost gap per correct answer, by Mercor's estimate
The cost gap was large too. By Mercor's calculation, meeting one scoring criterion cost about $0.21 with Claude and about $10.35 with a human accountant, using the US median wage. That is roughly a 49-fold difference.
The point is not whether a profession vanishes overnight. It is that the share of work inside a profession where AI does far better is growing fast.
What falls first may be the price of a task, not the job
To produce a report, people once searched for data, ran spreadsheets, wrote drafts and caught errors. AI has started to handle much of that in minutes to tens of minutes.
So the more likely first change is not that "the accountant disappears." It is that the market price of a specific task that took an accountant three hours drops sharply. Lawyers' first drafts, analysts' data gathering, marketers' research and developers' repetitive coding could face the same pressure.
Some work may rise in value: deciding which problem to solve, spotting assumptions the AI missed, signing off on a risky conclusion, and persuading clients and balancing interests.
| Skills likely to fall in price | Skills likely to rise in value |
|---|---|
| Searching and organizing information | Asking good questions and defining problems |
| Standardized calculation and document drafts | Verifying results and judging exceptions |
| Repetitive work with clear rules | Deciding with incomplete information |
| Memorizing lots of knowledge | Connecting thinking across fields |
| Producing a plausible-looking report | Owning outcomes and moving people |
The World Economic Forum expects about 39% of the core skills required in current jobs to change by 2030. It listed AI and big data as the fastest-rising skills, but also projected higher importance for analytical thinking, creative thinking, resilience, leadership, curiosity and lifelong learning.
How to work from here
1. Treat AI as a colleague, not a search box. The era of asking AI a sentence or two and judging the answer is passing quickly. Real productivity gaps increasingly come less from the model's name than from how well someone breaks work into pieces, supplies the right material, verifies results and connects them to the next step.
In Anthropic's 2026 Economic Index survey, a majority of AI users said speed, scope and quality of work had improved. What matters is that some people have begun to rebuild their whole workflow around AI, rather than using it occasionally.
2. Look for "my work that AI can't do," not "what I'm good at." A credential or years of experience is not a shield by itself. A better approach is to break your job into about 20 small tasks. Hand off quickly the ones AI already does at 80 or better. Spend more time on work where AI often errs, where accountability is required, or where trust between people matters.
3. Assume several rounds of retraining over a lifetime. The people who stay secure may be not those who find a job that never changes, but those who can redesign their work each time a new tool arrives. Speed of learning, more than job title, becomes the asset.
Education for children: back to fundamentals
Teaching elementary students nothing but prompting skills is not a good strategy. Today's popular tools may change in a few years. What lasts is the ability to think.
For investors: follow the money, not the top model
If AI starts to replace parts of professional work, money will not flow only to chatbot companies. As firms redesign work around AI, a broad ecosystem is needed: compute, cloud, data, security, business software and power infrastructure.
Still, "AI is growing" and "every AI-related stock will rise" are very different statements. The faster the technology moves, the less certain it is that today's leader holds the same margins tomorrow. Falling model prices, open-source competition, excessive data center spending and customers' return on investment are the key variables.
Insight Times Editorial Desk
