How agentic AI can reshape your tax function

While individual AI agents can automate and accelerate tax processes, the full transformative power of agentic AI becomes evident when teams of agents work together and are “managed” by an AI-powered orchestration layer with human-in-the-loop oversight.

Agentic teams feature individual agents working together, rather like the division of labor within human teams. Each has its own focus, skills and capabilities, with the goal of automating tax processes from end to end. These systems can also have an AI “manager” agent, which ensures activity is coordinated, correct and compliant. Using the trial balance use case as an example, an extended agentic team could feature agents designed to:

• Extract data from tax and finance systems
• Correctly categorize line items
• Upload reformatted data to tax provision software
• Identify new accounts and ensure they are correctly categorized
• Compare trial balances year-on-year to identify unusual or unexpected variations
• Review outputs to ensure accuracy and data compliance

“The goal of agentic AI is to create a series of micro agents that are all working together to automate a process from end to end, so the tax practitioner need only deal with the orchestration agent rather than dozens of systems,” Campbell says. He notes that the technology may be relatively immature right now, but like all spheres of AI, capability is evolving fast.

For example, agentic teams, with humans in the loop, could also be used for compliance and tax-assessment purposes. EY, which already uses natural language processing and machine learning to track tax regulation changes, is exploring how agentic AI could automate and extend this capability. A successful agent would not only detect new regulation, it would also determine the impact on a company’s tax position, alert relevant stakeholders, schedule stakeholder meetings at mutually convenient times and even generate draft analysis with recommendations for stakeholders to discuss and develop. Agentic AI is not infallible, however, so human oversight at key stages of such workflows will be critical.

“Some tax teams already use machine learning to automatically identify regulation change, but most of the subsequent activity is still highly manual and repetitive,” Campbell says. “The agentic AI approach is transformative because it automates and accelerates more of the process, saving precious time and resource while enabling tax practitioners to focus their skills on innovation and other value-added areas.”

The technology could also be used to autonomously process tax notices, with AI recommendations passed on to human tax practitioners for validation and consideration, explains Darren Beardsley, EY Americas Tax AI Leader.

Agentic AI systems can receive, interpret and classify tax notices from various jurisdictions with human oversight. This includes:

• Identifying the notice type (e.g., audit, late filing, payment demand)
• Extracting key data (e.g., dates, amounts, deadlines)
• Mapping the notice to internal tax obligations or previous filings

Agentic AI can also be trained to understand local regulatory language and formats, apply jurisdiction-specific logic and route a tax notice to the right team or system, accelerating workflows.

Beardsley points out that agentic AI acceleration of tax processes is likely to be particularly important in jurisdictions in Europe and South America where near-real-time tax processing is coming to the fore.

Aiken highlights the increasing demand for near-real-time tax reporting in Latin America and the EU as another potential use case for agent-to-agent communication. He envisages a time when taxpayer and tax authority will use standardized AI agents rather than highly configured APIs to request and provide tax data. “The issue with API calls is that they are very rigid,” Aiken says. “If there are any grey areas, they simply don’t work. AI agents, however, would be particularly effective in dealing with unforeseen circumstances and instances where there is incomplete or inconsistent tax data.”

Aiken and Campbell are clear that the technological challenges associated with agentic AI development and agent-to-agent communications are likely to be relatively straightforward to overcome. Barriers to adoption caused by internal data silos, regulatory compliance, ethics and an organization’s ability to trust and effectively integrate the technology, however, may prove more challenging.

Internal data silos are limiting AI performance. Agentic AI thrives on the ability to access and analyze large volumes of structured and unstructured data across business units. However, tax-relevant data is often scattered across disparate ERP systems, spreadsheets and local databases, each with its own format and level of accessibility. Without a unified data architecture and cross-functional collaboration, tax departments struggle to feed AI agents the information they need to reason effectively and deliver value.

“Right now, we have this huge V12 engine in the form of agentic AI,” Campbell says. “It’s amazingly powerful and ready to perform, but the tax function is slowly feeding the fuel and the AI engine is spluttering. We need to open the data pipes so agentic AI can achieve its full potential.”

Regulatory compliance presents another complex challenge. Tax departments are constantly adopting to changes in global tax legislation and meeting filing obligations and disclosure requirements. While agentic AI can help monitor and interpret these changes, companies should ensure that any AI-driven insights align with local legal frameworks. Compliance teams should also validate that agentic AI outputs meet the auditability standards required by tax authorities, adding another layer of scrutiny and control to AI deployment.

The ethical implications of AI in decision-making cannot be overlooked. Agentic AI systems that operate with a high degree of autonomy must be designed with safeguards to prevent unintended consequences, such as biased risk assessments or non-transparent tax positions. Without clear oversight and governance frameworks, organizations may face reputational risk and/or regulatory censure. Establishing ethical guidelines and accountability structures is essential before granting AI systems greater independence in tax assessment or compliance activities.

Building trust in agentic AI is equally critical. Tax professionals and stakeholders must have confidence not only in the technology’s output but also in its ability to explain how decisions are made. Many existing AI systems operate as “black boxes,” offering little visibility into the rationale behind recommendations. To overcome this, organizations should prioritize explainability and incorporate human-in-the-loop approaches that combine the efficiency of AI with the oversight of experienced tax practitioners.

Organizational readiness and mindset play a major role in adoption. Agentic AI requires more than just technical infrastructure. It demands a cultural shift. Tax departments should evolve from being primarily compliance-focused to becoming strategic, data-driven advisors. That evolution hinges on upskilling talent, redefining roles and fostering cross-functional collaboration with IT, legal and finance. For agentic AI to succeed, it should be embraced not just as a tool but as a transformative capability embedded within the broader tax approach.

As agentic AI begins to reshape the tax landscape, successful adoption by tax functions will ultimately depend on how well it is aligned with human insight, oversight and purpose. While the technology has the potential to automate complex tax tasks, respond to change and even make decisions, its greatest potential lies in how it empowers tax professionals to focus on higher-value, strategic activities. By integrating agentic AI into the heart of the tax function, guided by human expertise and ethical governance, organizations can unlock a new era of productivity and precision.

Rather than replacing people, agentic AI should be viewed as an intelligent collaborator, one that complements the judgment, creativity and contextual understanding of human tax professionals. Early adopters are already building this partnership by embedding explainability, accountability and transparency into their AI strategies. By prioritizing trust, auditability and human-in-the-loop decision making, leading tax functions are ensuring that AI enhances, rather than disrupts, the integrity of their work.

Ultimately, the future of agentic AI in tax is not just technological, it is human-centric. The organizations that will lead in this space will view AI as a means to elevate human capability, not eliminate it. By designing systems that augment tax professionals, foster collaboration and put people at the center, tax leaders can ensure their function is not only smarter and faster, but also more resilient, ethical and future-ready.

Summary

Agentic AI is rapidly reshaping the tax function, offering the ability to automate complex processes, enhance accuracy and boost productivity. Unlike traditional automation, agentic AI adapts to new information and collaborates with human teams. It excels in tasks such as GST product categorization and trial balance reformatting, while agentic teamwork enables end-to-end tax process automation. Key challenges remain, however, especially around data silos, compliance and trust, but success depends on a human-in-the-loop approach. Organizations that integrate AI with ethical oversight and expert judgment will gain a strategic edge and future-proof their tax functions.