As finance and accounting professionals continue to integrate AI into their daily workflows, the industry has largely focused on automation: using technology to process documents and transactions, identify patterns, and generate financial reports from available data.
While automating routine tasks has been invaluable for professionals who used to spend a significant amount of time and resources manually processing financial data, AI in accounting has now introduced a new feature that is designed to further optimize an accountant's daily workflow: AI agents.
Rather than automating simple or predefined tasks, AI agents embedded in accounting software are now designed to work towards an objective. What does this mean? They can potentially interpret information, determine a sequence of actions, and immediately modify the next steps based on what they encounter.
The main difference in a nutshell is: traditional automation (even with AI) follows a workflow. An AI agent can potentially participate in managing the workflow.
For accounting teams, that distinction matters.
An AI agent could monitor incoming financial information, determine what needs attention, complete permitted actions across connected systems and escalate exceptions to an accountant. Multiple specialized agents could eventually work together across parts of accounts payable, accounts receivable, reconciliations, reporting and other finance processes.
But giving AI greater agency also introduces an important question: How much of an accounting workflow should an AI agent actually control? At D&V Philippines, the answer begins with a clear principle. The future of accounting should be human-led and AI-enabled, not AI-driven.
While AI agents can expand what technology does within an accounting function, it should not remove human accountability, professional judgment, or appropriate governance from it.
AI agents are AI-enabled systems designed to pursue specified goals and perform actions on behalf of a user or another system.
A conventional generative AI tool is largely reactive. You give it a prompt; it generates an output, and the interaction generally ends there unless you provide another instruction. An AI agent can operate differently.
Depending on its design and permissions, an agent can receive an objective, assess the available information, determine what needs to happen next, use approved tools to perform actions and continue working until it reaches a defined stopping point or encounters something requiring human intervention – this is Agentic AI.
Suppose an accountant asks a traditional AI assistant to identify overdue invoices from a spreadsheet. The AI might analyze the information and produce a list.
An AI agent could potentially go several steps further. It might access an approved source of accounts receivable data, identify overdue balances, categorize them according to established parameters, prepare appropriate follow-up actions and flag unusual accounts for a person to review, helping your team not only identify risks but solve them way before they become problems.
There are several possible applications for Agentic AI in accounting. As an industry with various interconnected processes – which consist of both repetitive activities and data-driven analysis, having a tool that can perform tasks on an advanced level can be significantly advantageous.
The opportunity is not necessarily to create one AI agent responsible for "doing the accounting." A more realistic model involves specialized agents supporting clearly defined portions of a workflow, for example:
An accounts payable agent could potentially monitor designated channels for incoming invoices and begin processing them without waiting for an employee to manually initiate every step.
Within defined permissions, an agent might:
The important word is exception.
A well-designed agentic workflow should not assume every invoice can be processed autonomously. An unusual supplier, unexpected amount, missing documentation or discrepancy should trigger the appropriate human review.
The objective is not to remove the accountant. It is to reduce the amount of routine processing that reaches the accountant in the first place.
AI agents could also support receivables management.
Instead of an employee repeatedly reviewing aging reports and deciding which accounts require routine follow-up, an agent could monitor the relevant information against established parameters.
It might identify newly overdue accounts, organize them according to age or other approved criteria, prepare routine communications, and alert a professional when an account requires individual attention.
For a professional services firm, that distinction is particularly important.
Not every overdue client should receive identical treatment. Client relationships, disputed invoices, engagement circumstances, and other commercial considerations can require human judgment.
An agent may recognize that a balance meets a rule.
A professional needs to understand the relationship behind that balance.
Reconciliations contain another combination of repetitive comparison and exception-based work.
An AI agent could potentially gather records from approved sources, compare transactions, identify matches, and organize discrepancies for investigation.
Straightforward items could move through a defined workflow while unusual differences are routed to an accountant.
Over time, this could change where accounting professionals spend their effort. Instead of manually checking every item, they could concentrate on explaining and resolving the transactions that do not behave as expected.
This illustrates one of the most practical applications of AI agent accounting: allowing technology to handle more of the predictable workload while professionals handle uncertainty.
The month-end close involves numerous recurring activities, dependencies, and deadlines. That makes coordination almost as important as the accounting tasks themselves.
An agent could potentially monitor the status of assigned close activities, determine whether expected information has been received, identify outstanding dependencies, and notify the appropriate people when action is required.
More sophisticated systems could help prepare supporting information for reconciliations or reporting.
However, the close also demonstrates the limits of autonomy.
Accruals, adjustments, unusual transactions, and other accounting matters can require context and professional judgment. An agent might assemble the information required to make a decision, but the decision itself may need to remain with a qualified professional.
AI agents could eventually support the workflow surrounding recurring management reports.
An agent might gather information from approved systems, organize it into a predefined reporting structure, compare current and prior periods, and identify changes that meet specified thresholds.
The resulting report could then be presented to an accounting professional for validation and interpretation.
This can make reporting more efficient without confusing automated commentary with financial insight.
An AI system may be able to detect that an expense category increased.
A finance professional can investigate why it increased, determine whether the movement matters and explain what management should understand about it.
While the individual use cases seen above is currently being employed by various accounting software such as Xero's JAX – an AI agent that has accounts payable, accounts receivable, data entry, and bank reconciliation capabilities which finance teams can readily use to optimize their workflows, the possibility of accounting workflows themselves becoming agentic is not far-off.
Today, your finance team may move between an email platform, document management system, accounting software, expense platform and reporting tool to complete one process. Each system may already contain automation, but the professional still connects the workflow.
As AI continues to evolve, AI agents could potentially become another layer connecting these activities.
One agent monitors incoming invoices. Another validates available information. Another prepares the transaction for the appropriate workflow. A monitoring agent checks for exceptions. The accountant supervises the process, reviews exceptions, and approves activities requiring human authorization.
This is sometimes described as a multi-agent approach: several agents with specialized responsibilities working together toward a larger objective.
For finance teams, this could represent a much bigger change than simply adding generative AI to existing software.
The ability to take action is also what makes AI agents more sensitive than many earlier AI applications.
If a generative AI tool produces an inaccurate draft, a user can review and correct it before anything happens.
If an AI agent has permission to act on inaccurate information, the consequences can move beyond the output itself.
For example, an incorrectly configured or insufficiently controlled agent could potentially route information to the wrong place, apply an inappropriate classification, overlook an exception, or initiate an action that should have required review.
The appropriate level of autonomy should therefore depend on the nature and risk of the task.
A useful way to think about AI agents in accounting is through graduated levels of responsibility.
Only for carefully defined, low-risk activities should organizations consider allowing an agent to act within established boundaries without case-by-case approval.
As AI agents become more capable, maintaining a clear distinction between technological capability and professional responsibility becomes increasingly important.
For example, an AI agent may be able to:
This is why D&V Philippines approaches AI from a human-led, AI-enabled perspective.
AI should enhance the ability of accounting professionals to perform their responsibilities. It can support teams by helping them:
At the same time, we train our team to remain responsible for:
This approach becomes even more important with agentic AI, as the technology is no longer limited to producing outputs.
As AI takes on more tasks, organizations need to clearly define:
While the future accounting function may contain more agents, more automation, and increasingly connected workflows, its strength will still always depend on the professionals supervising those systems and making the decisions technology cannot responsibly make on its own.
For professional services firms, combining appropriately governed AI with capable internal and outsourced accounting professionals can help maximize the benefits of AI in accounting while still having necessary systems in place to ensure accuracy, confidentiality, and compliance in handling each clients' finances.
D&V Philippines provides finance and accounting outsourcing solutions designed to help professional services firms strengthen their accounting operations and expand their team's capacity. Read our Harnessing the Power of Business Analytics whitepaper to learn how combining skilled accounting professionals with technology-enabled processes can help you build the foundation needed to scale your finance capability alongside your business.
You can also talk to our team here here at D&V Philippines to learn how our accounting professionals can support your firm today!