# Marcus Tran, Co-founder & CTO at Fieldset (B2B SaaS, 11 employees) — read of LexiRisk, June 17, 2026

> Nine years building software, three years as a founder. I sign maybe 8-12 contracts a year and have never once felt good about it.

## How I got here

I was Googling "uncapped liability clause vendor agreement" at 11pm because we just got a 40-page MSA from a new cloud infrastructure vendor and their legal boilerplate looked aggressive. Second result was some law blog. Third was this page. I clicked it thinking it was a lawyer referral site. It wasn't.

## What I clicked first

The headline "Stop Signing Blind Contracts" landed. Not because it's clever but because it's literally what I was doing at 11pm on a Tuesday. I kept reading. The three-column pain section: "Lawyers Cost $300+ Per Hour," "You Don't Understand The Language," "The Damage Compounds Later." That third one actually made me pause because the example they give, "You sign a vendor agreement with an uncapped liability clause. Six months later... you're personally responsible for damages you never expected" is not hypothetical to me. I've been in that conversation.

## Where I paused

The FAQ answer about accuracy: "LexiRisk flags 95%+ of common dangerous clauses." That number is doing a lot of work. I stopped and read it twice. 95%+ of *common* dangerous clauses. Common to whom? Defined how? If they have training data from thousands of contracts, what's the source? Startup founders uploading NDAs? BigLaw document sets? That distinction matters enormously. A clause that's routine in enterprise SaaS looks alarming if you trained mostly on freelance agreements.

## What I distrusted

No customer quotes. No logos. No "here's what we caught in a real contract." The page tells me what it catches, lists six clause types with emoji, and then just... stops. I have no idea if a real founder has ever used this and found it useful. The "Built by Wishdeal Studio" footer raises my eyebrows. That's a product studio shipping ideas. The bottom of the page literally says "we don't have live customers on this idea yet." At least they said it. But I noticed it, and it changes how I read everything above it.

Also: "trained on thousands of real contracts." Thousands is a number that can mean 2,000 or 200,000. I can't calibrate that. A competitor in this space (and there are a few) would say something like "trained on 180,000 contracts across 14 industries." The vagueness here feels like nobody made the team actually commit to a number.

## What would convince me

Show me a side-by-side. Take one anonymized contract, run it through LexiRisk, and show me the output: what it flagged, what it said in plain English, and then have an actual attorney comment on whether it got it right and what it missed. One real example does more than six emoji-tagged clause types. If LexiRisk caught something a founder would have missed and they can document it, that's the entire product pitch.

Alternatively: publish a benchmark. Take 50 contracts with known issues (you could synthesize these), run them through, show the catch rate. If 95% is real, prove it publicly. That kind of transparency would stand out from every other "AI legal tool" I've seen.

## What I'd ask in an email reply

1. The MSA I'm looking at right now is 41 pages from an infrastructure vendor. What does your analysis actually look like on something that dense and jurisdiction-specific? Can I see a sample output before I pay $19?

2. Your FAQ says you analyze contracts "governed by US law." My vendor is incorporated in Delaware but their parent is UK-based and the agreement references English law in one clause. Does your model flag that kind of jurisdictional ambiguity or just skip it?

3. Who trained the model? Did a practicing attorney review the clause taxonomy, or is this pattern-matching on legal text? I'm not asking if it replaces a lawyer. I'm asking if a lawyer was actually involved in building it.

## Verdict: on-the-fence

The pain is real and they named it correctly. The page is cleaner and more honest than most tools in this category. But "no live customers yet" plus no proof of output quality means I'm not paying $19 tonight. I'd reply to a cold email from this team if they led with a real example.

---
*Memo by skeptic persona, generated 2026-06-17. Studio breaks own self-grading loop.*