• The widespread adoption of AI-assisted software development and emerging agent-based tools is creating new ways to automate both structured tasks and more complex workflows.
• This raises some very important questions for tax leaders responsible for automation strategies, particularly whether RPA still remains the most effective approach. 
• Take the processes involved in preparing a return. RPA can be configured to translate different formats to a standardised output, and can even be built to deal with known edge cases or errors that users see in practice. However, as ever with data challenges, there are invariably issues that users can’t foresee when setting up processes. In contrast, an autonomous AI-driven agent can instead interpret the goal and work out how to achieve it without requiring help, if the user chooses.
• Trained users can easily generate additional prompts and automation workflows, enabling agents to apply rules or carry out tasks across multiple entities.

Despite what many AI headlines would have us believe, automation has been around much longer than the last few years of the GenAI revolution. For instance, Robotic Process Automation (RPA), which uses software to automate tasks typically performed by people, is a well-established and successful niche of the technology industry, valued at over $27 billion annually. 

In the tax sector, for example, this might include everything from copying data between spreadsheets, extracting information from systems, updating datasets or any number of other repetitive computer-based activities necessary to complete compliance work.

A move away from RPA towards AI tooling

However, what once made RPA attractive compared to legacy macros and the like is now, in turn, tempting people to use AI tooling. The widespread adoption of AI-assisted software development and emerging agent-based tools is creating new ways to automate both structured tasks and more complex workflows. This raises some very important questions for tax leaders responsible for automation strategies, particularly whether RPA still remains the most effective approach. 

For example, across core tax functions and processes, AI tools are increasingly reshaping how certain tasks are carried out. Progress has been extremely rapid, with current systems capable of supporting a wide range of important, albeit routine, processes, such as analysing trial balances, flagging unusual line items and generating more tailored client queries, among many others. For busy tax teams, this means processes that previously took hours can now be completed in minutes. In these situations, the role of the tax professional evolves into one of reviewing and refining outputs rather than starting from the beginning of a given task.

The list of advantages is growing all the time, including the automation of in-depth research through tools that can identify, collate, and cross-reference information from a wide variety of sources. For tax professionals, these capabilities enable them to get up to speed more quickly on complex issues, such as regulatory changes or cross-border structuring.

Clearly, key AI outputs almost always need to be reviewed by a human expert, but even so, the underlying efficiency gains are significant, particularly for those handling high volumes of advisory work. The point is that this isn’t about replacing professional expertise. Instead, these tools help communicate it across stakeholders more efficiently so experts can spend more time being experts and significantly less on processes such as document formatting and version control.

From RPA to agentic AI 

Adding to this picture is the arrival of advanced AI tools, particularly autonomous agents, which, instead of executing a rigid sequence of steps, are designed to achieve a defined outcome on their own.  

Take the processes involved in preparing a return, for example, where information must be standardised from a plethora of different source formats, from a variety of source systems. RPA can be configured to translate these formats to a standardised output, and can even be built to deal with known edge cases or errors that users see in practice. However, as ever with data challenges, there are invariably issues that users can’t foresee when setting up processes.

In contrast, an autonomous AI-driven agent can instead interpret the goal and work out how to achieve it without requiring help, if the user chooses. The potential is enormous because it can make progress, without human input, towards completing the task rather than simply executing predefined steps. This can include complex processes, such as carrying out additional research or preparing responses for review. 

As a result, the point at which processes must be escalated for human intervention is moved further along the chain of expertise, meaning tax professionals can spend more time using their knowledge and experience on more complex and nuanced challenges, where human insight is most valuable. 

Trained users can easily generate additional prompts and automation workflows, enabling agents to apply rules or carry out tasks across multiple entities. Implementing complex automation is not only quicker than ever, but it is no longer just the domain of scarce software specialists. 

Maintenance is also becoming easier because automation logic can be updated and re-tested very quickly should requirements change. As a result, many tasks that previously required an RPA platform can now be handled through lightweight, custom-built AI automation tools or modern AI products such as Claude Cowork. 

Clearly, these developments are ongoing, and innovation is extremely rapid. But as more agentic AI tools come to market, tax technology investment decisions need to consider their potential to transform legacy processes and workflows. Don’t forget, this isn’t about replacing human expertise; it’s about positioning it at the right stages in vital decision-making processes so highly qualified experts can leave the manual, repetitive tasks to the technology. 

The real question now is which technology will define the next phase of automation in tax. Increasingly, the answer is AI.

Russell Gammon is the chief innovation officer at Tax Systems. 

What is SEIS tax relief and how to claim – What is the Seed Enterprise Investment Scheme (SEIS), how much can you raise and what does it offer entrepreneurs and investors who back them?

What is corporation tax? – What is corporation tax, what are corporation tax rates and how can you reduce your corporation tax bill? Metric Accountants has the answers