Question 1

Why review contracts with AI before paying a lawyer?

Accepted Answer

AI reads a 20-page agreement in under a minute and flags auto-renewals, indemnities, and one-sided clauses for a few dollars, while a lawyer bills hundreds per hour for that same first pass. Running every contract through AI first means you walk into the lawyer conversation already knowing where the risks are, so you pay for judgment instead of reading. The pattern that works: AI-only for routine low-stakes paper, AI-plus-lawyer for anything touching equity, investors, or big money.

Question 2

How do founders handle NDAs, ESOPs, term sheets, and compliance questions with AI?

Accepted Answer

Founders now run NDAs through AI checkers for one-sided terms, ask AI to explain every term-sheet clause in plain English against market standard, and use curated guides plus AI chat to understand ESOP mechanics before the lawyer meeting. AI is excellent at explaining, comparing, and flagging deviations; the final signature on anything that binds equity or investors still deserves professional review. India-specific rules (ESOP perquisite tax, SEBI changes, FEMA) need India-specific sources, not generic US answers.

Question 3

Where is AI legal help dangerous, what still needs a human lawyer?

Accepted Answer

AI makes things up with confidence: in 2024 Stanford found legal research tools wrong on about 1 in 6 queries and general chatbots on 58-82%, and Damien Charlotin's database lists over 2,200 cases of courts facing AI-made-up citations. Anything with real downside (fundraising documents, equity, disputes, regulatory filings, litigation) needs a human lawyer as the final check. And a public chatbot may not be confidential: in United States v. Heppner a US court held a defendant's chats with the public version of Claude were not privileged.

Question 4

Which AI legal tools actually work for startups, and how do you choose?

Accepted Answer

Use one general assistant and one tool built on the law you practise. Claude, ChatGPT or Gemini explains a clause, writes a first draft and summarises a judgment, but check every citation against the source before you rely on it. For Indian law, Lexlegis AI answers research questions citing the judgments and statutes it relies on, translates a document into 23 languages including 8 Indian ones, and works inside Word, with a 7-day free trial; jhana's agents have been used by more than 16,150 lawyers and 650 judges, and it built the Karnataka High Court's paperless system; Manupatra adds AI search to its Indian research database. For contracts, Spellbook works inside Word with a 7-day free trial, and SpotDraft runs a company's contract workflow from request to signature. Harvey and Legora serve large firms and in-house teams. Rule: one assistant plus one India-trained research tool, and a lawyer signs anything that goes to court or to a counterparty.

Question 5

What prompts get reliable contract review out of ChatGPT or Claude?

Accepted Answer

Treat the model like a sharp junior associate: give it a role, the jurisdiction, your risk appetite, and a required output format, then work clause by clause instead of asking 'is this contract OK'. Ask for risk tables, plain-English summaries, market-standard comparisons, and specific negotiation asks. Verify anything that sounds like a legal citation, and never send AI output anywhere important unreviewed.

Question 6

Is it safe to paste confidential contracts into ChatGPT?

Accepted Answer

Not on a consumer plan: Free, Go, Plus and Pro chats can train models unless you opt out, and in United States v. Heppner a US court held a defendant's chats with the public version of Claude were not privileged. Use a business or enterprise plan (or the API), which keeps data out of training by default, strip names and numbers first, or use a legal tool with contractual zero data retention. Assume anything you paste could be read back to you in discovery, and treat NDA-covered documents accordingly.

Question 7

How do I incorporate my startup without burning cash on lawyers?

Accepted Answer

Standardized platforms handle formation for a few hundred dollars with battle-tested documents: Stripe Atlas or Clerky for a Delaware C-corp, Razorpay Rize or a good CS firm for an Indian Pvt Ltd. AI is for understanding what you're signing, not for inventing formation documents from scratch. Get the boring things right on day one (founder vesting, IP assignment, clean cap table) because fixing them during due diligence costs far more.

Question 8

What goes into a founders' agreement, and can AI draft it?

Accepted Answer

A founders' agreement locks down the equity split, four-year vesting with a one-year cliff, IP assignment, roles, decision rights, and what happens when someone leaves. AI is genuinely good at generating a first draft from a solid template and explaining every clause, but have a lawyer localize the final version: vesting mechanics and share transfers work very differently under Delaware and Indian law. Investors will check for exactly these documents in due diligence.

Question 9

How do I protect my IP: assignments, trademarks, and patents on a budget?

Accepted Answer

First make sure the company, not you or your contractors, owns everything: signed IP assignment agreements are the first thing investors check, and pre-incorporation work legally belongs to the individual until assigned. Trademarks are cheap insurance; run an AI-assisted clearance search before you fall in love with a name. Patents are slow and expensive, so most software startups should file provisionals selectively and spend the savings on shipping.

Question 10

Can AI handle my privacy policy, terms of service, and data-protection compliance?

Accepted Answer

Generators like Termly produce solid baseline policies and compliance platforms like Vanta and Drata automate SOC 2 evidence collection, but a generated policy you never read is a liability, not protection. India's DPDP Act now applies to startups of every size (rules notified November 2025, full compliance by May 2027, penalties up to Rs 250 crore), and the EU AI Act adds obligations if you ship AI features to Europe. Map what data you actually collect first, then generate, then read every line.

Question 11

What should I check before signing a SAFE, iSAFE, or convertible note?

Accepted Answer

Know whether it's pre-money or post-money (post-money SAFEs stack: Waveup reports clients who signed three or four reaching Series A with 35-45% of the company already committed), what the cap and discount imply for dilution, and what happens on a down exit. In India the iSAFE is structured as CCPS and behaves differently from a YC SAFE, with FEMA rules layered on top. Model your cap table after every single SAFE you sign, not just at the priced round; AI chat is genuinely good at walking you through the conversion math.

Question 12

How do I get due-diligence ready with a clean data room?

Accepted Answer

Investors expect 50-70 documents across incorporation, cap table, IP assignments, key contracts, and financials, organized in a logical folder tree with a read-me index. Start collecting from day one, and run an AI pass over your own contracts to find the problems (missing signatures, unassigned IP, expired agreements) before an investor's associate does. In India, legal due diligence digs hard into ROC filings, FEMA compliance, and ESOP paperwork.

Question 13

How should founders negotiate customer and vendor contracts with AI's help?

Accepted Answer

AI collapses negotiation prep from hours to minutes: paste the contract and ask for the other side's likely pushback, market-standard fallback positions, and a redline you can actually propose. Enterprises already run AI on their side of the table, so walking in without it means negotiating blind. The Krafton case is the standing warning: a CEO who followed ChatGPT over his own lawyers into a $250M contract fight lost badly, so AI prepares the negotiation, humans close it.

How founders use AI for Legal & Contracts

The questions

Also in Starting Up