Post Highlights
- LLMs can generate, summarise, and converse using advanced text prediction capabilities.
- RAG retrieves external data in real-time, reducing errors and improving accuracy for evolving tax legislation.
- AI agents automate multi-step workflows, combining retrieval and refinement to handle complex tax tasks.
- Use GenAI does not replace professional expertise which continues to be vital for compliance and intricate legal judgment.
Introduction
Below is a short, practical guide for anyone looking to leverage Generative AI (GenAI) for Tax purposes.
This guide briefly introduces the concept of Large Language Models (LLMs), retrieval-augmented generation (RAG), and AI agents which are key concepts at the time of this post. TO give further context the guide makes references to OpenAI’s GPT‑o1 model (a hypothetical next-generation AI system) before giving you real actionable steps for incorporating GenAI into your daily tax workflows.
If you already under the basics of LLMs, RAG and AI Agents, click here to skip to the guide.
Quick Overview: LLMs, RAG, and AI Agents
What Are LLMs?
LLMs—such as GPT-o1 or DeepSeek R1 are advanced AI models trained on vast bodies of text. They learn to generate human-like text by predicting the next word in a sequence, allowing them to replicate human cogitative approaches to:
- Draft coherent paragraphs
- Summarise complex documents
- Converse in natural language
Retrieval-Augmented Generation (RAG)
RAG extends an LLM’s abilities by allowing it to pull real time information and data from external sources, such as a local database or knowledge base. This drastically reduces errors and enhances the model’s capacity to handle niche or frequently updated topics (e.g., new tax rulings or regulatory updates) and is core to being able to use LLMs for tax law.
AI Agents: Going Beyond Single Prompts
AI agents take GenAI to the next level by automating multi-step processes instead of providing a single response to a prompt. An AI agent can therefore rationalise complex to:
- Break down a task into sub-steps
- Retrieve necessary data using RAG or API calls
- Verify and refine its own output
This iterative process makes AI agents extremely powerful for complex tasks—such as conducting an end-to-end analysis of a tax scenario or automatically drafting, reviewing, and even emailing a set of compliance documents.
How GPT‑o1 Model Works
GPT‑o1 is a hypothetical example of a next-generation LLM that combines:
- Large-Scale Training: It’s been exposed to an even broader range of public and industry-specific texts, enhancing its general knowledge base
- Domain Fine-Tuning: It’s fine-tuned on specific tax and legal data—making it more adept at interpreting and summarising Australian tax legislation or giving preliminary assessments of regulatory changes
- RAG and Agent Capabilities: GPT‑o1 is designed to easily integrate with retrieval systems and AI agent frameworks. It can query documents in a corporate database, run a series of logic steps, then produce a precise, context-driven final output
In short, GPT‑o1 stands as an example of how AI tools are evolving beyond static text generation to become dynamic assistant capable of orchestrating multiple information sources and tasks without constant human intervention.
Practical Steps to Using GenAI (With or Without AI Agents)
Follow this step-by-step guide for better results and a deeper understanding of the AI process

Conclusion
Generative AI tools particularly those featuring retrieval-augmented generation and AI agent capabilities offer unprecedented efficiency and versatility to research and synthesis tax legislation and material for the Lawyer to apply in practice. It does not replace experience and the ability to deliver complex theories to clients gained through years of experience earnt by the Tax Lawyer.
Therefore, in its current state, treat AI as a collaborative partner, a tool that can handle mundane or time-consuming tasks that is programmable by the intelligent use of prompts and experience by the tax lawyer. By crafting precise prompts, using agents to automate multi-step processes, and rigorously reviewing AI outputs, you can harness GenAI’s full potential while safeguarding quality and compliance
Further Reading & Resources
- OpenAI – Guide to Prompt Engineering
- Hugging Face – Introduction to Retrieval-Augmented Generation
- LangChain – Building AI Agents and Workflows
- Australian Taxation Office – Legal Database
- OECD – AI Principles