Tax practices are entering a new powered era of artificial intelligence (AI). Moving past simple query responses into a large language model (LLM), autonomous AI agents now represent a new and significant step forward as we enter stage three of OpenAI’s five-level roadmap toward Artificial General Intelligence (AGI).

Stage three focuses on creating sophisticated AI agents capable of independently performing complex tasks, making informed decisions, and substantially streamlining tax workflows to support tax professionals in their roles.

Agentic AI Tax Framework

One key advancement is the “Agentic Framework,” which replicates internal tax teams’ operations. Specialised AI agents within this framework conduct detailed legislative research, review transactional documents for tax implications, and calculate taxable income efficiently. Automating these time-intensive tasks significantly reduces manual effort, enhances accuracy, and accelerates advisory processes.

AI agents also integrate seamlessly with existing tax software, efficiently extracting and analysing data. This allows tax professionals to generate precise, timely reports, freeing them to focus more strategically on client interactions.

Example of a Multi-Agent Framework

Leveraging the sophistication of large language models (LLMs) combined with specialised AI agents, tax teams can significantly enhance their productivity. A proposed multi-agent framework includes:

Partner Agent

Initial intake of user queries, applying prompt engineering through iterative feedback loops. It confirms specifics such as the relevant area of tax law, legislative references, and desired outputs before forwarding queries.

Manager Agent

Receives refined prompts from the Partner Agent, then delegates tasks to specialised agents, ensuring the results commercially address the original query.

• Tax Technical Agent: Ensures technical accuracy by engaging sub-agents:
  • Legislative Agent: Connects via API to AustLII, sourcing up-to-date legislation. Incorporating a definitions library enhances accuracy.
  • ATO Agent (Technical): Interfaces with the ATO legal database accessing rulings and practice compliance guides to ensure ATO information is considered.
  • In-House Agent: Accesses corporate databases to incorporate internal tax positions and historical advisory context to ensure outputs factor in house positions.
• Tax Compliance Agent: Validates the accuracy and compliance of financial data through sub-agents:
  • Software Agent: Interfaces with corporate accounting and CRM systems via APIs to ensure data accuracy.
  • ATO Agent (Compliance): Ensures appropriate tax forms and instructions from the ATO database are integrated into compliance processes.

Together with other components such as LLM response evaluation, template libraries and Agent Workings databases, this multi-agent framework illustrates a future where AI and tax professionals collaborate seamlessly, enabling comprehensive, informed responses to complex tax issues. AI thus acts as a powerful ally, enhancing the depth and accuracy of tax advisory processes.