"Manisha has been a Chartered Accountant for 30 years. She runs her own practice in litigation, traditional CA work, auditing, and systems reviews. She is a Certified Information Systems Auditor and a Certified Insolvency Resolution Professional.
She is not someone who adopts new tools carelessly. She approached AI with curiosity AND caution. And because of that, her account of using it is one of the most honest and useful I have encountered.
Here is both sides of her experience.
WHERE IT GENUINELY HELPED:
Income tax litigation involves serious, time-intensive research. When a client receives an assessment order and wants to appeal, the CA must:
✦ Understand the assessment order in depth
✦ Identify legally defensible counter-arguments
✦ Find applicable case laws with accurate citations
✦ Draft the grounds of appeal in correct legal format
✦ Ensure everything is factually and legally defensible
Before AI, this took Manisha 10 to 15 days per case. That is not inefficiency — that is the thoroughness that good legal work demands.
After she learned to use AI properly — specifically how to upload documents, give context, and build responses through iterative conversation — the process changed dramatically.
She would upload the assessment order as a PDF. Ask for a summary. Identify key points to contest. Ask the AI to help draft grounds of appeal. Refine through further questions.
Over time, the AI began anticipating her needs — suggesting related provisions, offering to draft in specific formats, flagging arguments she had not considered.
The same process now takes 2 to 3 days.
That is a 5x to 7x productivity improvement on some of the most complex, high-stakes work in her practice.
WHERE IT COMPLETELY FAILED — AND WHY THIS MATTERS MORE:
AI halluccinates case laws. This happened to Manisha directly.
The AI cited cases with legitimate-sounding names, courts, and citation numbers. She cross-checked on actual legal databases. Those cases did not exist. They were invented.
Around the same time, she read a newspaper report about a legal professional who submitted AI-generated case citations in a court filing. The judgment noted that the cited cases had never existed. The professional consequences were severe.
Manisha's response was not to stop using AI. It was to build verification into every step of her workflow. She now treats AI output as a starting point for research — never an endpoint. She cross-checks every citation. She reads actual case summaries. She applies 30 years of domain knowledge to assess whether a case is genuinely applicable.
HER SUMMARY — IN HER OWN WORDS:
""Unless you know the subject — if I don't know what 1+1 is and AI gives me 3 — I should also know whether it is correct.""
AI amplifies expertise. It does not replace it.
In fields with real professional consequences — law, medicine, finance, accounting — the human with domain knowledge is not optional. They are the essential check on a tool that is powerful but not infallible.
Use AI to accelerate. Use your expertise to verify. The combination is formidable. Either one alone is incomplete.
▶️ Watch Manisha's full story: https://youtu.be/9uIr9kNoKCk"
Top comments (0)