The Australian Taxation Office is adopting a measured approach to implementing agentic AI, influenced by its organizational culture and the sensitive nature of its data holdings. According to Assistant Commissioner for Strategy, Design and Architecture Jimmy Tzimopoulos, who spoke at an IBM-run session during Gartner’s IT Symposium/Xpo, the office is currently focusing on preliminary architectural preparations while taking a cautious stance on the use of agentic AI in end-to-end processes.
Re-engineering processes takes precedence over autonomous agents
Tzimopoulos noted that while agentic AI can be effective in certain use cases, it is not always the most suitable solution. “What we’ve found, especially with agentic AI, [is that] you can achieve a lot of the use cases, but just because you can, doesn’t mean that you should,” he said. Tzimopoulos emphasized the importance of assessing each use case individually, suggesting that up to 95 percent of them could be addressed through process re-engineering, business automation, and machine learning, with agentic AI playing a secondary role.
The ATO’s approach to agentic AI is centered on understanding its potential applications and limitations. Tzimopoulos stressed the need to create a safe learning environment for staff to explore the capabilities of agentic AI. “You’ve got to create an environment for people to learn with this capability,” he said. This cautious approach is reflected in the office’s decision to make Copilot Chat available to staff while keeping its agentic component disabled due to concerns about sensitive data.
Ensuring human oversight for sensitive data
The ATO has made Copilot Chat available to a wide range of staff, but the agentic component remains inactive due to the office’s sensitive data holdings. “In the ATO, we use Copilot Chat, so we’re making that available to as many people as possible to use,” Tzimopoulos said. “We haven’t turned on the agentic component of that because we’ve got sensitive data, and we want to make sure that we understand how we use that.” The office has shifted its approach from “human in the loop” to “human at the helm” to ensure accountability for AI-driven outcomes.
This change in approach is driven by the need for accountability, as Tzimopoulos explained. “I know it’s only a slight tweak of language, but it goes to the accountability piece,” he said. “Excuse the language – someone’s arse has got to be on the line for what you’re trying to achieve because in the end, we’re accountable for the outcomes to all citizens and residents of Australia, and therefore, that human really needs to be accountable for that.” As a risk-averse organization, the ATO is mindful of its responsibility to protect sensitive data and is taking a cautious approach to agentic AI.
Tzimopoulos emphasized the importance of aligning AI deployment with the organization’s culture and risk appetite. “You’ve got to architect your AI based upon the culture of your organisation,” he said. “If you’re risk-averse, make sure that you’ve got that reflected in how you deploy AI. Don’t be someone that you’re not.” The ATO is currently undertaking preliminary architectural preparations for agentic AI, with a focus on ensuring that its design and progress reflect the office’s culture and risk appetite.
Future accounting roles in the AI era
The ATO is exploring ways to present the costs and benefits of AI-driven services in a clear and transparent manner. Tzimopoulos suggested that a balance sheet could be used to illustrate the total cost of ownership and help business lines understand the value of these services. “For us, we’ve got an idea of potentially a balance sheet,” he said. “But if we can actually present to a business line a balance sheet that shows who’s using it, and what the TCO is, and then they can correlate that back to how much they’re actually paying for the service, you’re starting to actually get them to think about what is the true value of this end-to-end process. You’re also getting them to think, ‘Do I need to do [a process] holistically using agentic, or can I actually compartmentalise it?’”
