A Friendly Warning About AI-Generated Contracts

We use AI every day. Claude helps us format contracts. Gemini helps with due diligence reviews. We're not here to tell you to stop using these tools. They're genuinely useful, and in many cases a rough AI-generated agreement is better than operating on a handshake.

But in the last few months, we've reviewed a steady stream of contracts that founders generated with AI and signed without counsel. On the surface, they look fine — clean formatting, professional language, the right headings. Underneath, some of them are quietly catastrophic.

This is your friendly warning.

---

What we've actually seen

Let's be specific, because "AI contracts have problems" is too vague to be useful.

The agreements look fine. Until they don't. The tell isn't always obvious on a quick read — the formatting is clean, the headings are right, the language sounds professional. The problems surface when someone who does this every day actually reads it. Last month we reviewed a contractor agreement where the service provider had the option to be paid via SAFE — on a vesting schedule. A SAFE. Vesting. To a contractor. For services rendered. Every one of those words exists in startup legal vocabulary. Together they form something that isn't a real thing and creates problems in every direction. The founder had no idea.

Equity terms that quietly hurt the people you're trying to help. We've seen offer letters grant actual stock instead of options — triggering a tax event on the recipient at grant that they had no idea was coming. We've seen equity subject to vesting with no mention of 83(b) elections, which can mean a major tax bill down the road on shares the person can't even sell yet. These aren't edge cases. They're what happens when an AI assembles equity language without understanding the mechanics behind it.

The due diligence moment. This is the one that really stings. When an investor's counsel reviews your agreements during a financing and finds broken IP clauses, off-market provisions, or equity terms that don't match your cap table — they don't just flag the issue. They draw a broader conclusion: this founding team doesn't know what they're doing. Sophistication signals compound. A clean set of agreements says you understand how this works. A stack of AI-generated contracts with amateur errors says the opposite. That impression is hard to walk back.

---

The legal debt problem

Technical debt is a concept every founder understands. You build fast, cut corners, ship code that works but isn't clean — and over time the weight of those shortcuts slows everything down. The fix is painful but ultimately possible: you refactor, rewrite, rebuild.

Legal debt works the same way, with one critical difference: you can't fix it unilaterally.

Bad code can be rewritten by your engineering team on their own schedule. A bad contract requires the consent and cooperation of the other party to amend. Your contractor who signed a broken IP assignment six months ago and has since moved on? You have to find them, explain the issue, and convince them to sign a correction. Most of the time they'll cooperate. Sometimes they won't. Sometimes they can't be found. Sometimes they'll realize they have leverage they didn't know about.

Every problematic agreement you sign today is a potential cleanup project at the worst possible time — when you're in the middle of a financing, when an acquirer is doing due diligence, when a key employee departure creates urgency. Legal debt compounds quietly and surfaces loudly.

---

Why AI gets this wrong

AI systems are not bad at law. They're genuinely impressive at producing legal-sounding text, identifying relevant concepts, and structuring documents. But there are three things they consistently lack:

Market norms. "Market standard" in startup law isn't written down anywhere. It lives in the heads of practitioners who've done hundreds of deals and know what sophisticated parties expect. What's normal for a Series A SAFE? What's the going rate on option pool size before a seed? What IP provisions does every serious investor's counsel expect to see in a contractor agreement? AI systems approximate this from training data, but they don't actually know — and they can't tell you when they're off.

Judgment about your specific situation. We see enough deals that within less than a minute we can tell if a contract was well-prepared or a mess. That pattern recognition comes from doing this hundreds of times. A contract that's fine in one context is wrong in another. An NDA appropriate for a vendor conversation is not the right template for an employee's first day. AI doesn't have that instinct, doesn't ask the questions that change the answer, and can't tell you when something feels off.

Consequences. AI generates text. It doesn't bear any responsibility for what happens when that text is signed and later scrutinized. The founder who signed the agreement bears all of it.

---

What to do instead

We're not saying never use AI for legal drafting. We use it ourselves, every day. The difference is that we pair it with judgment, market knowledge, and accountability.

The agreements that matter most — IP assignments, employee and contractor agreements, advisor grants, SAFEs, customer contracts — these are the ones worth getting right the first time. Not because the legal fees are trivial, but because the cost of getting them wrong is enormous and the window to fix them without friction is short.

A much better approach: use AI to get oriented, understand concepts, and think through what you need. Then have counsel — someone who does this every day and knows what investors and acquirers actually expect — review and finalize the agreement before it's signed.

At Flux, that's what the Foundation plan is designed for. We stay embedded in your business so that when you're striking agreements with employees, contractors, advisors, and partners, we're in the room — not called in to do cleanup two years later.

The AI isn't the problem. Signing what it outputs without a second set of eyes is.

---

If you have a stack of AI-generated agreements you've never had reviewed, our Foundation plan includes a legal health check and corporate record audit as part of onboarding. It's almost always cheaper to find the issues now than to fix them mid-deal.