Agentic AI: Transforming the Future of Chartered Accountancy in India

The Dawn of AI in Accountancy, Audit and Finance in India

Indian Accountancy has already adopted Robotic Process Automation (RPA) in completing day-to-day repetitive tasks which are based on predefined rules. However, a new era of Agentic AI has begun, which is well ahead of its time compared to RPA. Unlike RPAs, Agentic AI can learn from patterns to make independent decisions while adapting to new challenges. We have seen that RPAs work properly on structured data only, whereas Agentic AI can work on unstructured data as well, like PDFs, scanned documents, emails etc. So, in a nutshell, RPAs can auto fill responses, perform repetitive tasks, whereas Agentic AI can extract, summarize, and analyze data to fill responses. It can also give us feedback and acknowledgments in real-time while adapting to ever-changing situations. Furthermore, AI agents can be created to debug/fix errors in real-time while executing tasks.

We are presently witnessing the era of Generative AI (tools like Chat GPT, Deepseek, Gemini, Google AI Studio, Notebook.Lm, Gamma.app, Napkin AI etc.) which does help us on the basis of pre-trained models integrated with Natural Language Processing (NLP), but it lacks autonomy and decision-making. Most of these Generative AI tools involves LLMs (Large Language Models) which have interfaces for generating texts, prompts, graphs, images, videos etc. on the basis of the given input only. It also lacks the ability to think independently and generate results based on pre-trained data or reinforcement learning. For instance, ChatGPT can give us a draft reply of any notice, but it cannot actually log in to the account and submit the reply to the notice, once finalized. Further, if we prepare a PPT using Gamma.app, we still have to manually share it with our team. Whereas, AI agents can upload the prompt to Gamma.app for PPT generation and once finalized, a second AI agent, in this flow, can email this PPT to our team while composing the body of the email for team\'s review.

Therefore, the need for an AI tool is felt that can go across websites, link all workflows, act on our behalf, make decisions within the liberty we set firsthand. This is exactly why we need an AI Agent. This phase of developing need-based AI agents is called Agentic AI. So, from Generative AI to Agentic AI, the world around us is transforming, and AI is moving a step ahead.

Now imagine an AI Agent that prepares tax computations and automatically updates them with every single amendment/update simultaneously. Also, just think about an AI Agent which can download the returns from an email inbox, analyze the return data with the help of LLM, and then, if it passes the compliance test, proceed to the return filing portal and actually file the return. Sounds amazing, doesn\'t it? This is the real power of Agentic AI, which we Chartered Accountants can embrace and leverage to the fullest. We can create a series of AI agents and interlink them to work in a seamless flow. Additionally, we can integrate APIs of different AI tools within this workflow. The sky is the limit for this Agentic AI use cases.

It will give us a first-mover advantage if we start exploring Agentic AI now, because, going forward, Agentic AI is also going to be outclassed by Artificial General Intelligence (AGI). AGI is the ultimate future of AI, which we have explored up till now. In other words, we can say that if AI is the entire universe, then AGI is the Observable Universe up to which we can comment.

Agentic AI: The Need of the Hour in the Indian Context

Agentic AI works by integrating NLP, Machine Learning, Reinforcement Learning and Data Analytics so that it can make a Multi Agent System. A group of these AI agents then make an AI Agent Flow, which works as a combined package to achieve the desired result and to execute a given task. As a Chartered Accountant, if someone has their target set, then a flow can easily help them to execute the process for achieving that target. Now, this target can be data analysis or forecasting by using LLMs, credit appraisal by calling APIs of different AI tools in market trends and checking the financial health of the borrower.

Now, imagine an accounting prospect where one AI agent is using RAG (Retrieval-Augmented Generation) to extract data from unstructured scanned PDFs, then another AI agent is working at the back-end to arrange all the extracted data in a DBMS. Then comes the third agent, which is automatically posting entries from this DBMS to an ERP or Accounting Software, and meanwhile, there is a fourth agent which is continuously updating and reclassifying the data on the basis of Accounting Standards and Income Tax/GST amendments. How easy can this entire accounting process be? Simply from rough scanned images of invoices, banks statement and other documents, we are preparing a complete financial statement, that too in compliance with applicable Accounting Standards. This is the future of AI in the Indian Accounting System.

However, this evolution is not going to reduce the work, in fact, it will enhance the efficiency and accuracy of the accounting work. Therefore, CAs offering accounting services will now have their virtual assistant who will do the tedious and laborious work of data entry. Ultimately, CAs will have more time to ponder upon the correct classification of transactions, the alignment of financial statements with applicable accounting standards, and they can present the finalized results in a much better way. Therefore, Agentic AI can be a boon for Chartered Accountants and thousands of Accounting Assistants whose work is going to be very easy. Also, with the help of saved time from data entry and data extraction from scanned images, CAs majorly offering accounting services can now provide deeper insights into financial statements, can proactively detect errors in real time, and can focus on other important avenues of practice like advisory, assurance, and compliances.

As such, the truth is always as it was i.e., AI is not going to replace anyone, but people using AI will surely replace those who are not using it. So, there will always be people who will supervise the AI Agents. Also, there will always be people who will analyse the work done by AI. Therefore, CAs must learn how to give correct input to AI models so that our target is achieved. For that, CAs must initially start using Large Language Models (LLMs) to get used to interacting with AI. Once we get the basic idea of how AI responds to our queries, we can move further towards agentic AI. So, the next step in our learning should be focusing on how we can make AI agents flow. We can start with simple flows like email automation using AI. Moreover, we have a lot of online videos where we can learn about Generative AI and Agentic AI as well. ICAI is also offering AI CA Level -1 batches to train the members in using AI in our practice. Thereby, through continuous learning, attending webinars, checking out videos on AI, and allotting a few hours weekly to learn AI, it can surely upskill us and keep us up to date with this quickly evolving AI world.

Practical Use Case where Agentic AI can be used for Chartered Accountants

• Revolutionizing Tax Compliance
  Almost every CA in practice has one thing in common, and that is tax compliance work. Be it a traditional old CA Firm of \'70s-\'80s or a young, new age CA of Gen-z, all of them have tax compliance-related work in some way or another. So, Agentic AI is surely going to help a lot in tax compliance, right from the beginning of collection of data for tax return filing to summarizing and organizing the data in a proper schema for filing. Then, we can run our accuracy tests on this data with the help of another agent, and finally, when all agents pass this data to the final uploading agent, it will seamlessly upload the same and authenticate the filing on our behalf. Therefore, unlike generative AI, which can only give us information related to tax compliance, Agentic AI moves a step further by analyzing and submitting the returns as well, thereby completing the task.
• Enhancing Audit Efficiency
  In an audit environment, if AI agents are introduced, CAs can be equipped with a real time and continuously evolving risk assessment system, as these AI agents can perform compliance testing, identify anomalies and exceptions, automate control testing, and also, provide us real time exceptional reports. If an AI agent is created which can continuously validate the entire dataset in line with internal controls and predefined policy frameworks of the company, then the coverage of the audit can be drastically enhanced, instead of only relying on sampled data, and more accurate results regarding deviation from the required compliance can be found out easily. For example, during the audit under the GST Act, we can easily deploy an AI agent which will scan the entire database to detect if any ITC has been claimed on invoices involving blocked credits u/s 17(5) of the CGST Act 2017.
  
  Also, for bank audits, we can create an AI agent which can easily identify the abnormal financial transactions like roundtripping, piggybacking etc. This agent will help find anomalies and exceptions by entirely scanning a large dataset of bank transactions, thereby covering another audit aspect, and we can find out possible frauds during the bank audit using Agentic AI.
  
  AI agents can also help us automate control testing by regularly monitoring and testing internal controls and flagging deviations in real time. For example, take a payroll audit: An AI agent will continuously check the entire payroll transactions, and it can easily find out ghost employees based on historical trends and other predefined check-fields. Now, if any ghost employee exists or if any salary is authorized beyond limit, then a weakness in internal controls can be easily identified.
  
  Therefore, auditing in an Agentic AI environment will surely enable continuous risk monitoring, early detection of anomalies and deviations. Also, it will allow us to have overall more robust compliance testing.
• Staying Up to date with Amendments and Latest Regulatory Frameworks
  It is a matter of fact that a CA needs to update their knowledge every day and the very basic reason behind this continuous learning is the ever-changing tax laws, amendments in various Acts, bills etc. However, many of us might miss some of this plethora of amendments and updates. Therefore, AI Agents will come into the picture here, which will keep alerting us every now and then on the updates across Indian Laws/Acts/Bills. Now, we can design our agent in such a manner that, based on our historical chats with LLMs, the AI agent will identify our area of interest or our core fields of practice where we are more interested in getting updates. Also, there are a lot of AI tools out there in the market which even convert these amendments and updates into podcasts and audio-visuals for us to grasp this knowledge in whatever manner suits us best. Imagine a recent bill of around 100 pages, and one cannot find time to read it all. In this case, one can simply create an AI Agent which will automatically pull the latest bill/act from the web, and with the help of AI tools like Notebook.LM, create a podcast of it and start playing it whenever your phone connects to your car-play. How easily we can utilize the time from leaving home to reaching office. One can listen to the entire podcast and at least have a bird\'s-eye view of a bill of around 100 pages in just 15-20 minutes.
  
  However, there will always be a need for HITL (Human-in-the-Loop), as an AI agent integrated with Generative AI tools, indeed, can convert the information into interactive podcasts, etc. but the responsibility of interpretation of the same lies with the CA who is an expert in this domain. Therefore, AI agents can be used to streamline the process of pulling information from the web and converting the same into podcasts, etc. but interpreting this podcast and judging the content of this podcast will entirely be the call of the end user i.e. CA.
  
  Also, as of now, it is well established that AI can produce inaccurate outputs, so while interpreting the information related to any Act/Law/Bill etc., we need not to completely depend on the output provided by AI. In fact, there will always be a need for review of the same.
• Marking the virtual presence of a CA
  In the field of this rapidly evolving technological era, a CA also needs to mark his/her virtual presence. It is now the need of the hour that we should educate and inform our clients in an interactive and engaging way, which will in turn result in better input data that we get from our clients and more tax compliance. But the bigger challenge lies in the process of making the information interactive and engaging for the clients. Many of us might not be used to creating audio-visuals of the information which may engage our target audience. Therefore, now, with the help of AI tools, we can create agents which will transform this information into more engaging and interactive audios, visuals, graphics, or animations. All we need to do is to create an agent by programming it to trigger whenever there is an update which we need to pass onto our clients. It will make those ready interactive graphics, flowcharts, visuals etc.
  
  Now, once the output is prepared by an AI agent, it can be directly submitted to us by the Agent for review. After a complete review, when this information in an interactive mode seems fit for broadcast, then another AI agent will broadcast it to all our client\'s groups, social media etc. Imagine how our virtual presence is going to be impactful and right on time. Therefore, the entire process from selecting the information which needs to be passed on, to creating a script out of it, and then finally creating an audio-visual out of this script can, though, be automated using AI agents, but every single step in this entire process will require a HITL for review, since whatever information we are broadcasting will have to be well interpreted as per the law of the land, and should have been reviewed for originality and accuracy also. As such, it can be summarized that AI agents will make the process of marking our digital presence easy and fast while keeping the authenticity and accuracy of the information true to the law by involving review checks at every single step.

The list above is only just illustrative and not exhaustive, one can imagine hundreds or thousands more use cases of Agentic AI for CAs. However, it is also to be emphasized that Agentic AI, when working with autonomy, should also be harnessed at the same time. It is very evident that issues like Data Privacy, Accountability, and Bias risk are inherent in Artificial Intelligence across the globe as of now. So, these possible threats can be minimized by signing SLA (Service Level Agreements), reviewing before final authentication, and regular monitoring of AI Agents etc.

Ethical Considerations and Societal Impact

Since we have discussed how AI can be used more and more in our practice and how we can generate content and data, it is also important that we as Chartered Accountants should consider the work ethics and social impact of AI also. Following ethics and the code of conduct is not new to our fraternity, as we are already following our code of ethics issued by ICAI. However, AI gives us the liberty to freely flow in the ocean and we can dive deep into it, which may result in an adverse social and work impact, if handled negligently. Therefore, the use of AI with the highest ethical standards and the code of conduct is expected from every Chartered Accountant, so that the data privacy of our clients can be ensured. CAs are the custodian of clients, data, and we need to evaluate every possible adverse outcome, before we give anything to AI for processing. Hence, it is highly recommended that the use of AI in our practice be adopted only after taking due care of all the possible threats and risks associated with it, like a breach of data privacy, AI bias, hallucination, full autonomy etc.

Need for embracing this transformation and AI-driven future for Chartered Accountants

The transformation of AI from Generative to Agentic itself is a great shift in the field of technological advancement, but the bigger picture is yet to come when AI will further transform from Agentic to General Intelligence. Therefore, we Chartered Accountants should fasten our seat belts for this amazing roller coaster ride, where if we learn it with interest, then it will surely give us that advantage of cashing in on maximum benefits out of it. Time is quickly reaching a point where machines will ease the work for most of the CAs, but the only condition is that one should embrace this technological shift by updating themselves, so kick start today itself. Imagine the day when AGI will make decisions as a financial strategist or automate the audit process by 80-90%. In that era, we must retain the command of the AGI. Presently, ICAI is leaving no stone unturned to make its members understand that AI is the need of the hour, and every CA should start investing in it right from the Level-1 AI CA course.

Therefore, the road ahead is clear for Chartered Accountants to include the use of AI in our daily routine practice so that we become used to it initially and then we can embrace Agentic AI to start preparing projects for ease of practice. Thereby, we all shall be prepared to welcome AGI, and we will reign over it for a better AI-driven future.