←
Akshat Singhal890 posts
Akshat Singhal avatar

Akshat Singhal

@asinghal_7

Founder, CEO legistify.com || BITSian || Entrepreneur || Investing || Sports Enthusiast

◷ Joined April 2013▧ 219 Photos and videos
Akshat Singhal avatar
Akshat Singhal@asinghal_7·10h

AI has made great engineers harder to spot! An average engineer can now ship as much code as a great one. Theo says AI has widened the gap between them to 100x. I mostly agree. In a regulated software space, the best engineer is the one who notices the edge case nobody has tested, or pushes back when someone wants to skip the audit trail. That never shows up on a productivity dashboard. It’s changed how I think about hiring. One strong engineer now carries a lot more weight, so getting that call right is critical. I’m also wondering how juniors will learn, since AI now does much of the work that used to teach them. We’re hiring engineers who care about getting the details right. If that sounds like you, check out our careers page. I’ve dropped the link in the first comment. You can apply here: legistify.com/careers

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Nov 16

Look at which tasks are still stuck in traditional tools. They’re the ones where someone has to be right. A16Z argues that value is moving from the system that stores data to the AI layer that acts on it. In their survey, 56% of sales teams already do account research mostly or entirely in AI tools, while forecasting and keeping the CRM clean still happen the old way. Legal has even more of that kind of work. When a regulator or a court asks which version was signed, or who approved a change, you need the actual record. I think legal teams will soon start their day in an AI feed of renewals, hearings, and obligations. That feed is useless if the data behind it is wrong. If your legal team is bringing in AI and wants its records to stay reliable, I’m happy to show you how we connect the two at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Nov 13

Finding security flaws is getting automated. Fixing them still takes manual effort. OpenAI built an AI system that went through Linux, the software behind most of the world’s servers and Android phones, and found 948 serious security flaws. Any one of them could let an ordinary user take full control of a machine. The AI also wrote a fix for each, but the tech team now has to check every one. The run cost about $1.2 million, roughly $1,280 per flaw, and covered around 7% of the code. For me, the bigger story is legal. India’s data protection law asks companies to take reasonable steps to keep people’s data safe. When finding flaws costs this little, I think regulators and courts will expect a lot more, and “we didn’t know” won’t hold up for long. If your legal team is taking a fresh look at its security duties, I’m happy to show you how we help teams track compliance at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Nov 11

Every AI leaderboard is built on someone else’s idea of what matters. Artificial Analysis just updated its Legal index to give more weight to non-hallucination. I’ve written about fake citations a lot, so I’m glad to see it. This chart shows how much those weights change the picture. The same model ranks 3rd for legal and 7th for healthcare. So before picking a tool, figure out what your team actually needs, whether that’s citation accuracy, handling long contracts, or working across your systems. Then test it on your own documents. If your legal team is setting its own criteria for AI, I’m happy to walk you through how we evaluate it at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Nov 9

The billable hour gets a little harder to defend with every new AI model. I came across a list of businesses whose margins go up as AI improves. It warns that firms selling hours of work will get squeezed on price each time a new model comes out. I’ve worked with enterprise legal teams for 10 years, and I think they’ll start asking their outside counsel for fixed fees pretty soon. More of the work is also shifting to in-house teams. The same list says accountability, checking AI’s work, and audit trails will grow in value, and in most companies, legal handles all of that. AI is already writing contracts, policies, and marketing copy, and someone still has to sign off on every piece. I expect in-house legal to have a lot more say in how the company runs because of it. If your legal team is preparing for the extra review work AI is bringing in, I’m happy to show you how we help at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Nov 6

AI agents are cheaper than offshore teams in India, but that’s not the whole story. A new a16z chart puts AI agents at about $6-8 an hour and offshore teams in India at about $10. AI agents save a lot compared with a US back-office team at $30-45. Compared with India, they save a couple of dollars an hour at best. In legal ops, one missed hearing date or wrong renewal notice can wipe out a year of those savings. I’d judge agents by how much each correct result costs you, and that includes the time someone spends checking their work. That’s why I think India’s service firms could do well here. They already know the processes and the clients trust them, so pairing agents with people who can catch mistakes makes a lot of sense. If your legal team is figuring out which ops work agents can safely take over, I’m happy to share how we think about it at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Nov 4

AI just beat human reviewers at the one thing we thought they were best at: catching mistakes. Researchers hid errors in 3 clinical trial reports. Then they asked 15 AI models to find them. The models found between 53% and 96% of the major errors. Human reviewers found just 27%. The table result surprised me most. When a table didn’t match the text, humans caught it 7% of the time, and the AI models caught it every time. Contracts are full of this kind of mistake. Think of a payment amount in a schedule that doesn’t match the clause, or a cross-reference that points to the wrong section. Almost every model missed the same subtle issue, though. So if one AI misses something, adding a second one probably won’t help. In our experience, AI is great at checking whether a document agrees with itself. The calls that depend on context still need someone who’s seen a lot of deals. If you’re curious about how contract review works on Legistify, I’m happy to walk you through it.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Nov 2

YC’s new Requests for Startups reads like a list of places AI can’t just plug into. They’re asking for AI that runs fully offline for banks, hospitals, and defense; AI for city hall, company-owned models trained on private data, and expert help for benefits, taxes, and immigration. As a YC W22 founder who’s spent 10 years building for regulated enterprises, seeing these on one list made me smile. If you’re building in one of these spaces, the product will likely be the quickest part. Accreditation, security reviews, and procurement take much longer, and sales cycles can easily run past a year. That slow stretch is also what keeps competitors out once you’re in. Question to all founders here: which of these 11 would you build, and why?

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Oct 30

The AI that wins your demo might be the one you should trust least. Take a look at this chart from Artificial Analysis. It maps how AI models communicate across 8 traits, from warmth and conviction to transparency. None of these models leads on all of them. Tbh, that matters more in legal than most people realize. After 10 years of building for enterprise legal teams, I’ve noticed something about evaluations. The tool that agrees with you and sounds confident usually wins the room. That’s a real problem in legal. When a GC asks whether their position is strong, the most useful answer is often the uncomfortable one. So test AI tools on how they handle bad news. Give them a weak position and see whether they push back or flatter you. Then match the style to the task. Contract review needs transparency about uncertainty. A notice to a counterparty needs conviction. A note to the business needs warmth. If your legal team is choosing AI tools right now, I’m happy to share what we look for at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Oct 28

Confidentiality is the reason law firm data exists in the first place. I read a post last week arguing that frontier labs would pay almost anything for law firm data, and that firms should finally build a strategy around it. I strongly disagree. Those files hold clients’ deals, disputes, and mistakes, shared on the understanding that they stay private. That data was never the firm’s to sell. The comparison to defunct startups doesn’t hold either. A startup’s Slack belongs to the startup. A law firm’s files are about other people’s businesses and lives. And once clients believe their matters could reach an AI lab, they’ll share less. The very data that makes firms valuable stops getting created. Even using one client’s negotiation to price another client’s deal raises real questions about consent and conflicts. We work with legal teams at over 300 enterprises, the same clients these firms serve, and I expect more GCs to add clear AI and data-use terms to their outside counsel guidelines soon. If your legal team is thinking through AI and data-use terms for its outside counsel, I’m happy to share what we’re seeing across our customers at Legistify.

Akshat Singhal avatar
Akshat Singhal@asinghal_7·Oct 26

The eMudhra incident affected maybe 1% of its business. But it put trust on the line overnight. eMudhra is a Bengaluru-based certificate authority. It issues the digital certificates that help browsers confirm a website is genuine. On October 17, an independent researcher found that eight of its intermediate certificates had private keys weak enough to guess almost instantly. A flaw in its April 2025 key generation ceremony had produced weak keys alongside the strong ones. The issue was reported responsibly, browsers blocked the certificates within days, and eMudhra revoked them. There’s no evidence they were misused, and the IT ministry confirmed that Indian DSC and eSign services were unaffected. Still, its shares fell sharply. After 10 years of selling to large enterprises, I’m not surprised. Its biggest customers, banks and governments, pay it to manage digital keys safely, and that’s exactly where this flaw sat. In enterprise, that kind of doubt rarely shows up right away. It shows up a year later, in renewals. What stood out to me is that the flaw went unnoticed for nearly 18 months. Following a process and checking its output are two separate jobs. One company’s incident also raises the bar for everyone. The government has set up a team to investigate and is checking with other operators, so I’d expect every vendor in this space to face sharper questions in its next security review. If your legal team is tightening how it verifies vendors and tracks compliance, I’m happy to show you how we help at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Oct 23

With AI, anyone can launch a competitor to your product in a few weeks. Even custom workflows, which feel sticky, can be rebuilt faster every year. But earning an enterprise’s trust still takes years. I mostly agree with Nikita here. I also read a great piece last week with a simple test: is your advantage hard to do, or hard to get? After building Legistify for 10 years, here are a few more moats I’d add: 1. Trust and accountability When something goes wrong in a contract or a dispute, someone has to answer for it. Enterprises pick vendors who’ve proven they can carry that weight. 2. Regulatory depth Security reviews, audits, and compliance checks take months per customer, and better code doesn’t speed any of them up. 3. Living data Court records change every day, and years of tracking them across India’s courts build an edge a fresh model can’t copy quickly. 4. Time The long sales cycles that slowed us down early on now protect us. 5. People. All the credit here goes to our team for the relationships they’ve built with customers. At the end of the day, people prefer to work with good people. If your legal team is looking for a technology partner it can count on for the long run, I’m happy to show you how we work at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Oct 21

We’re about to let AI agents spend company money. Induction Labs just ran a really interesting experiment with them. Their new FrontierML benchmark gave AI agents $10, $50, or $200 to get results on 21 complex ML tasks. Humans scored highest at every level. But some agents spent more and ended up with worse results. That pattern looks very familiar to me. In litigation, I’ve seen matters keep eating budget for years because nobody stopped to ask whether they were still worth fighting. More spending rarely fixes a weak position. The best GCs I’ve worked with treat every rupee as a decision. They set a budget per matter, check what it bought, and know when to settle. Legal ops teams have spent years building these controls for outside counsel. Agent budgets will need the same discipline: limits, approvals, and a clear record of what each dollar bought. If your legal team wants clearer visibility into where its time and spend to go across contracts and disputes, I’m happy to show you how we do it at Legistify.

Post attachment
Akshat Singhal avatar
Akshat Singhal@asinghal_7·Oct 19

Look at the green line. That’s what AI adoption in legal looks like. This chart from OpenAI tracks how much of each department’s AI work moved to OpenAI’s Codex, its agent tool. Engineering climbed slowly for months. Legal sat near zero until early 2026, then shot up to 88% in just a few months. Over 10 years of working with enterprise legal teams, I’ve learned that a flat line in legal rarely means a no. It usually means the team is still figuring out whether the tool can be trusted. Once that bar is cleared, the switch happens fast and almost all at once. I know this is OpenAI’s own team, so most companies will move differently. But I’d bet the growth curve will look similar. If your legal team is figuring out where AI fits into its workflows, I’m happy to show you how we’re doing it at Legistify.

Post attachment