Ask people who are deploying artificial intelligence in accounting firms which processes AI agents can truly handle independently today, and you will be met with hesitation. The answers tend to be evasive. Although firms are using AI more and more, the idea of software handling accounting work from start to finish without human intervention remains a distant prospect.
These doubts are not being voiced by outside skeptics. They come from the very people who build and use these agents. When it comes to specific examples, the discussion invariably turns to the same things: automation in Excel, email sorting, dashboard creation, and standardization of repetitive tasks. Work that AI could manage entirely on its own is hard to find.
Where accountants use AI most
The best results come in areas where the work already follows clear rules. By far the most frequently mentioned word is Excel. Every day, accountants convert exports from point-of-sale systems, check formulas, prepare supporting documents, and move data into standardized templates. These are among the most readily accessible opportunities. When someone spends time working with formulas, importing CSV files, exporting data, and transferring it into templates, that is where deploying AI first pays off.
Similarly, firms use AI to keep their email inboxes organized, generate dashboards for clients, or create reusable procedures that ensure the same routine is performed consistently across teams. Large financial companies are taking the same approach. They typically start with a single operational process and only then expand AI into other areas, such as accounts receivable, accounts payable, and planning.
The challenge of understanding agents
Meanwhile, many firms have moved beyond merely testing tools to the more difficult question of how to integrate them into daily operations. And that takes time. New technology alone will not change the way people work. It will take a while for the time invested to pay off. This is often described as an adoption curve. Firms must first embed AI into their processes, and only then will it begin to reliably deliver the time savings everyone expects from it.
This is consistent with what surveys show. AI agents need far richer context than standard corporate data usually provides. Organizations that build a so-called semantic layer around their data improve AI accuracy while reducing costs. The greatest obstacle, therefore, may not lie in the capabilities of the tool itself, but in the state of a company’s data.
This is also connected to another problem: a lack of knowledge. Many people in the industry are only beginning to familiarize themselves with the basic building blocks of agentic AI before they can even decide where to deploy it. There is a sense of being overwhelmed by how to get started and how to scale everything up. People want to understand what an agent consists of, what components it has, and how to build one. There is plenty of material explaining this, but navigating it is not easy.
Agents still need to be monitored
Which tasks can AI agents currently handle from start to finish during the financial close without human oversight? The answer is: not a single one.
Accounting data, client behavior, and transaction account activity are still too inconsistent for humans to be removed from the review process. It will be a long time before accounting firms decide to eliminate people entirely from any process. Estimates suggest that AI can make seventy-five to eighty percent of such work faster and more consistent, but the remaining twenty to twenty-five percent still needs to be checked.
CFOs at large companies say the same thing. Some admit that there is nothing agents can handle completely without supervision. Others would first entrust AI with administrative approvals, while keeping strategic decisions to themselves.
The accounting industry thus remains firmly grounded in repetitive, predictable work. Spreadsheet automation, meeting transcripts, routine tasks. Agents that could connect all of this and handle it on their own remain more of a promise than a reality. Even those working on them admit it.
Source: cfo.com
