Correct isn't enough: the three tests of legal advice
Legal advice can be right on the law and still fail the client. Three tests every business lawyer's advice should pass, and why AI raises the bar on all three.
Every piece of legal advice for business clients should pass at least three tests. Is it legally correct? Is it what this client needs right now, given everything the lawyer knows? Is it enough for the decision the client has to make, especially a business decision that has to be made well and made fast?
Law school and continuing legal education teach the first test. Clients need the other two, and that’s where most of the problems I’ve been hired to fix began, and where the ones I’m hired to prevent would start.
One mistake I’ve seen more than once starts with correct advice. Investors want a Delaware corporation, so the company becomes one, and the advice to incorporate there is correct. Nobody mentions that Delaware charges an annual franchise tax, or that unless the company reports its issued shares and gross assets, the state bills it by the number of authorized shares. For a company with the 10 million authorized shares most startups carry, that method produces a bill of about $85,000. The other method Delaware allows, assumed par value capital, usually brings the same company down to the $400 minimum. The founder opens an $85,000 notice for a company with no revenue. The advice passed the first test, and nobody checked whether it was what this founder needed to know.
The 83(b) election fails the same test. The statute gives a founder 30 days after receiving restricted stock to file it, and after that the election is gone. Advice that mentions the election but not the deadline passes the first test and fails the second. In Puerto Rico, even that rule plays out differently because federal income tax generally doesn’t reach a Puerto Rico resident’s Puerto Rico–source income.
Why it matters: Clients now expect legal work that’s faster and better at the same time, and AI has raised that bar for every lawyer. The old trade-off between speed and quality was always a process problem. Lawyers who fix the process, with AI working under their judgment, stop being the roadblock to informed and legally sound business decisions.
Why good advice fails the second and third tests
1. The billable hour pays for time. The client pays for every inefficiency in the process and can’t predict the bill until it arrives. I’ve seen what it does inside firms, too. When hours become the measure of a lawyer’s worth, they shape who gets credit, who gets the work, and how partners and associates treat each other. Zack Shapiro’s essay on running a two-person firm with AI drew millions of readers this year. He argues that billing junior lawyers’ time for work clients never valued is collapsing as AI makes that work cheap. I agree, and I’d add that clients were never well served by it in the first place.
2. Many lawyers never got the repetitions. Startup work is its own discipline. SAFEs, seed and Series A rounds, vesting schedules, cap tables, and franchise tax filings come with a culture that runs on speed and collaboration, where the lawyer on the other side is a counterparty to close with. Puerto Rico shows how that happens. For years, few companies on the island raised outside money at all, not even seed rounds, so few lawyers got the practice. More startups raise money there now, though whether that has produced the economic impact Puerto Rico needs is a separate question. Many mainland markets have the same problem.
3. Clients’ expectations changed. People track a package and a ride in real time. Then they hire a lawyer and hear nothing for two weeks or a month. Not every legal service has to work like package tracking, though some should. Still, silence is one of the complaints I hear most from clients. Together with inefficiency, it makes a lawyer look like a roadblock to running the business, even when the work itself is sound.
4. Sometimes the law itself lags. Statutes don’t always match how startups actually operate, and founders who assume the familiar rule applies get caught. In Puerto Rico, for example, an LLC is taxed as a corporation unless it elects otherwise, while a multi-member LLC defaults to partnership treatment under federal rules. A founder who assumes the federal default finds out at tax time. The lawyer’s job is to know where the law diverges or falls short and to plan around it before the client runs into it.
The client already has AI
Clients now test their lawyer’s advice against AI. They arrive with drafts already written and questions already researched. That makes an answer that sounds correct enough cheap and fast, and it raises the bar on everything else.
What a client can’t get from AI alone is the second and third tests: advice fitted to their situation, and complete enough to act on. A lawyer who competes with AI on speed alone will lose, but a lawyer who uses it to do better work faster gives the client something AI can’t.
How I know a different model works
I’ve sat across the table from large firms on the same deals, working from the same forms. I’ve held my own, and I’ve caught and fixed mistakes larger firms made on my clients’ matters, from flawed formations to routine compliance tasks nobody completed. More lawyers on the email chain didn’t prevent any of them; one lawyer’s judgment, applied in time, would have.
Large firms train excellent lawyers, and some of my best counterparts came from them. But the leverage model wraps that judgment in cost and delay, and makes clients pay to train new lawyers on tasks AI can already do. Process discipline and structured data, along with generative AI, may be able to fix that.
How each test gets met
1. Correct. I generate every document from my own forms, which I’ve researched and backed with playbooks. When I form a company, the founders’ details go into one record, and I confirm that record once. The certificate of incorporation, bylaws, board consents, stock purchase agreements, and 83(b) elections are all built from it against my own forms, with a redline against the templates for anything that changed. A name or a share count can’t differ between page 4 and page 19. AI helps with research, and I verify what it finds against the primary source before anything relies on it.
2. Right for this client, now. Judgment is the product, and the way to apply it consistently is to build it into the process. Some call it encoding your judgment into the tools; others call it process design. Either way, the improvement survives every matter instead of living in one lawyer’s memory.
In practice, that means the 83(b) deadline goes on the calendar as a hard date the day the stock is issued. In due diligence, AI pulls each fact from the client’s documents along with the exact quote it came from, and I confirm each one before it goes into a schedule. For Act 60 clients in Puerto Rico, the decree’s annual obligations and the residency count stay on the calendar after the decree is granted, so the client doesn’t lose the benefit in year three over a missed filing.
3. Enough to decide, fast. My job is to get the client to the business goal. Ethics and quality are the conditions that make speed trustworthy.
In a SAFE round, the founders review dilution at the valuation cap before anyone signs. Each investor closes on their own track, and each wire is checked against the approved amount before the SAFE is released. Before a priced round, the founders see a conversion schedule showing what every SAFE becomes, holder by holder, so the dilution doesn’t surprise anyone at closing.
The division of labor
AI reads, extracts, compares, researches, drafts what isn’t built from a form, and helps me build the tools themselves. I decide, verify, approve, sign, and answer for the result. Nothing leaves my practice without my approval.
AI handles the reading and the first pass, so my time goes to the decisions that need a lawyer. That speed is only worth something if clients can trust what produced it, and that trust depends on limits I keep on every matter.
What I won’t do
- I won’t let AI draft a client document that no one reviewed.
- I won’t treat your AI-drafted contract as wrong because AI wrote it, or as right because it reads well. I’ll review it like any other draft.
- I won’t bill you for re-entering what you already told me.
- I won’t send you a bill that surprises you.
- I won’t put your confidential information into an AI tool whose terms let the provider train on it, and my engagement letter tells you how I use AI on your matter.
What you should expect from any lawyer
- You should know what the work will cost before it starts, and the bill shouldn’t surprise you.
- You should be able to see where your matter stands without having to ask.
- Your lawyer should know the mechanics of your company’s stage, and should say so plainly when they don’t and bring in someone who does.
- You should give each fact once, and every document should reflect it correctly.
- Deadlines that outlive the closing should be tracked by someone, and you should know who.
- If your lawyer uses AI, you should know how it’s used and who checks its work.
- You should understand what you’re signing, in your own language.
These are the standards I hold myself to. Hold every lawyer to them, including me.
The bottom line: A correct answer is becoming the cheapest thing a lawyer sells. What clients pay for is the answer that’s right for them, in time to use it.
Iván Ríos Mena is an attorney licensed in the District of Columbia and Puerto Rico and is not admitted in California. Ruta.Legal, ivan@riosmenalaw.com. This essay is general information, not legal advice for any specific situation.