Post Category → Formation

Startup Founders Using ChatGPT and Gemini for Legal

Related Reading: Legal AI and the Future of Startup Law

TL;DR: No matter how much it might horrify conservative lawyers, founders are going to use AI for some legal tasks, and it’s actually an upgrade from what they’ve been doing previously for low-stakes issues (just Googling and winging it). But there are a few things to keep in mind to gain efficiency while minimizing penny-wise and pound-foolish risks.

The rise of AI is unquestionably the most profound change in the legal industry of this decade, including for startup lawyers. Every single serious law firm (including Optimal) has implemented AI in their practices in a material way, and is testing other tools in beta/alpha for future use once they mature enough for safe integration.

Adjacent to the issue of lawyers using AI is the riskier proposition of clients (including founders) using it on their own, completely bypassing legal professionals. Cynics might expect all lawyers to handwave away clients leaning on AI, expecting them to feel threatened by competition or (again, cynically) to hoard as many billable hours as possible. But the cynicism is overdone. Legal is a big industry with a variety of personalities, and there are a lot of lawyers who are very much not luddites.

Young first-time founders have always struggled with soberly understanding why legal works the way it does, because the legal context of permanent contracts, high-stakes negotiations, and irreversible mistakes is so different from the iterative tech and biz environment in which founder psychology thrives. The subtext of “move fast and break things” is that it’s acceptable because when things break they can, in fact, be fixed; especially in software. The often misunderstood subtext in legal is that, actually, they can’t be fixed; so maybe you really should slow down a bit.

The short story of my perspective is that expecting founders to never use AI for legal is spectacularly naive. Is it risky? Yes. But risk tolerance is literally what makes founders founders. If anyone in business is going to lean on AI somewhere for legal, it’s entrepreneurs.

It’s naive because the reality is, pre-AI, many founders have already been doing plenty of legal-ish things away from lawyers. Lawyers – even at lean boutiques, forget BigLaw – are not cheap. Talent has to get paid.

Even well-funded young startups have to pick their battles as to when to lean on professional legal advice, and when to simply wing it, because approaching every decision with the same level of risk intolerance of a Fortune 500 company is simply not feasible. There isn’t a budget for it.

So the sober way to assess the prospects of founders using AI for legal isn’t to judge it against a non-existent parallel universe in which every decision has air-tight elite counsel, but against founders doing DIY legal based on personal instincts or a simple google search alone; because that’s what’s already been happening. Viewed that way, “GPT, esq.” is a massive upgrade.

Is using GPT Pro or Gemini Deep Think better than simply using Google? Obviously. It’s also better, speaking candidly, than the advice you’re likely to get from all the so-called “low cost” lawyers out there with minimal real-world experience in startups and venture capital.

I say this often: law is a lot like healthcare. There are specialties and niche subspecialties. Serious “Startup Law” (often called ECVC for “Emerging Companies and Venture Capital”) is a niche subspecialty of corporate law. There are hundreds of “corporate lawyers” but far fewer corporate lawyers truly specialized in ECVC, with all the knowledge of market norms and contextual nuances that entails.

Asking a non-ECVC lawyer – like a small business lawyer, or a generalist who dabbles in real estate, estate planning, and god knows what else – for advice on your VC-backed startup is a lot like asking a dermatologist (and a low-end one) for advice on a neurological issue. You are almost certainly going to get better answers from GPT Pro.

Founders are going to use AI for some legal issues, and that’s probably a good thing. However, there are a few things I’d suggest for doing it intelligently; moderating some of the risk, having a clear understanding of AI’s most likely failure modes, and knowing when it will be high ROI to loop in a lawyer even at the earliest stages.

First, use Clerky or Stripe Atlas for formation, and don’t assume that investor-preferred “standards” are non-negotiable.

These are extremely well-tested automation tools that are far cheaper than anything a law firm will produce, relying on well-respected templates produced by specialist lawyers. AI tools like GPT and Gemini are pulling from the entirety of the internet when they generate an output, but 99.999% of the data they’ve been trained on is irrelevant at best and simply wrong for your context at worst. For a couple hundred dollars you can avoid almost all the worst mistakes founders make in a DIY formation by simply leaning on Clerky or Stripe Atlas.

Even better, use them with an ECVC lawyer. Most of our own clients are incorporated on one of these tools, and we are simply in the loop to ensure no contextual nuances “bust” the standardized terms they use.

For early-stage fundraising docs, like Post-Money SAFEs, don’t assume they aren’t negotiable. Many VCs want you to think that, but it’s not true.

AI, just like prior legal automation tools, fails on unique context.

Context is one of the main reasons to keep a specialized human in the loop on legal, and one with a long-term relationship with the core team. Aside from check-the-box “compliance” sorts of issues, most of the legal issues startups face do not have a single correct answer.

What industry are you in?
Who are the founders?
Who were their prior, or who are their present, employers?
Where do they live?
Where is the company headquartered?
What’s the relationship of the founders to each other?
What are their growth and exit goals?
Who are their investors or likely investors, and how are their expectations in tension with the founders’?
What’s the distribution of leverage among the various relevant parties?
What’s the broader economic / market environment they are navigating?

To an AI tool, you are simply a user among millions. But to a lawyer with whom you have a long-term relationship, you are a specific company with very specific contextual needs, and that heavily plays into legal answers.

This is actually why, in the long-run, healthcare will be far more shaped by AI than law will be. There is an extremely higher level of subjectivity to desired outcomes in law than in medicine. Your goal in healthcare is to eliminate the disease, identify whether or not (binary) you have a condition. Goals in high-stakes legal are far less straightforward, and therefore far less addressable by algorithms alone. Personalities, relationship dynamics, and business contexts vary a lot more than biology.

Founders can reduce some of the risk of AI by incorporating more context into their prompts. Of course, this relies on founders actually knowing what context is relevant, and they’re often going to be wrong. Any generalist DIY tool has a failure mode that follows from the user, in this case founders, simply not knowing what they don’t know. The AI doesn’t know what you don’t know either.

AI is very helpful for pre-gaming discussions with, and messages to, your lawyers.

A lot of back-and-forth between clients and lawyers, which costs money, isn’t itself about devising a contextual strategy or even answering a legal question for the client but simply educating the client on certain concepts that they need to learn before an answer can even be derived. AI can be extremely useful to get that educational process out of the way.

So you might prompt the AI with something like “I’m going to ask my corporate lawyer about [X], but what are some concepts I should understand ahead of time to make our discussion as efficient as possible? What questions should I ask them?”

You can even do this with contract review. I’ve seen clients e-mail us a document and say “I ran this through GPT and here were some suggestions, just mentioning to you in case helpful. And by the way, I really don’t care about [X, Y, and Z.]”

The theme here is AI can be fantastic for the objective parts of a legal matter, like making sure you understand specific concepts, and that can really cut down on communication and resolution time with lawyers, whom you can then lean on for the more subjective or contextual parts of the project.

AI can turn a 30-minute call with your lawyer into a 10-minute one, with zero loss in output. That saves money, and many lawyers (myself included) love it when calls finish early.

AI (alone) is the most dangerous for permanent high-stakes relationships, particularly with key employees, commercial partners, and investors.

Related to the above point that context heavily influences a lot of legal issues, anything involving high-stakes negotiation – of a key new hire, a new investment or other commercial relationship – is going to be extremely dangerous to lean on AI (alone) for. The AI will not have an appropriate understanding of the negotiation context, including the leverage your counterparty has versus yours, and what the range of feasible outcomes is.

See Negotiation is Relationship Building for a deeper dive on all the subtle power and psychological games that experienced players (like VCs) – who virtually always have more experience than first-time founders – can play to sway a negotiation. Whatever output an AI tool might generate is going to be too complex for founders to actually understand and utilize on their own. It’s also likely to not encompass the true range of options because it was trained only on publicly available data, and it’s not like a public article has been written on every single negotiation tactic or legal nuance.

Once again, AI can be helpful for educating founders on relevant concepts without their lawyers’ timer being turned on, just like AI can be helpful to educate a medical patient before going into a consult with a physician. But it’s not going to make you as knowledgeable as an elite professional. You don’t have the time.

For the love of all things holy, do not use AI (alone) to negotiate your equity round term sheet.

Don’t assume using lawyers will always be unacceptably expensive.

Using a specialized boutique firm instead of BigLaw typically cuts legal bills (and hourly rates) in half with zero drop in quality, and sometimes improves quality because you’re working with more senior people. Thus don’t attach yourself to firms that are unnecessarily expensive (when leaner high quality options are available) pushing you to overuse AI.

ECVC lawyers also typically have precedent and templates you can lean on that do not require a ton of their time to generate. Before assuming it’s going to cost hours of time for a lawyer to prep a document for you, ask if they have a template to start with. That template, if it exists, will unquestionably be more useful and less risky than anything an AI tool will generate.

If not already obvious (I hope it is), pay up for the “Pro” models of ChatGPT or Gemini.

Among people who are benchmarking the models for legal tasks, the most advanced (and expensive) models have been clearly shown to be the least hallucinatory. Pay the $200 per month. This (legal) is not a game. If you are bypassing lawyers to lean on AI, which is risky, do not be so pound foolish on what is still peanuts (a couple hundred dollars) in the grand scheme of things.

Will the above completely eliminate the risks of leaning on AI for legal as a founder? Of course not. But it will certainly reduce them.

For a separate discussion on how AI is not likely to change Startup Law, though there will be plenty of shysters who pretend otherwise, see Legal AI and the Future of Startup Law. This notion of “AI First” law firms will work in very discrete compartmentalized areas, like high-volume low-stakes contracts for larger enterprises, but it will crash and burn in the kinds of high-stakes long-term representation that serious startup lawyers do.

A good metaphor is that “AI First X-ray review” (a narrow productized service) has serious legs. But an “AI First hospital” is preposterous, at least with the technology emerging in the next decade. The margins required by investors simply will not be there without VCs playing background games to cut down quality via de-skilling (eliminating seasoned senior professionals with deep contextual knowledge), while hiding it from naive clients. That’s what happened with Atrium a few years ago.

Elite boutiques have already brought dramatic efficiency (lower rates via lower overhead) to high-end law, while maintaining flexibility and Partner-level oversight. The same Partner who would be $1300/hr in BigLaw will be $650 at a boutique, without making less money. That is a big drop. Those lean boutique firms, along with BigLaw (higher rates for higher scale and ultra high-stakes), are themselves rapidly incorporating AI into their practices right now.

You’re going to somehow build “AI First” direct competitors that are cheaper, not malpractice nightmares, and have the kinds of far-larger profit margins (~3x of professional services) required for VC returns? Hope you have Nobel prize winning bleeding-edge tech. Good luck and God bless.

Seed-Stage Startups Should Shrink Their Option Pools

Background reading:

You only get 100% of your cap table to give away (or keep), and the sad fact is founders make all sorts of tactical errors that needlessly give up points to investors and other parties. Sometimes those errors are driven by bad advice offered by misaligned participants in the ecosystem.

One example I’ve written extensively about is the aggressive anti-dilution mechanism built into YC’s default Post-Money SAFE Template. YC portrays its template as a wonderful legal fees-saving “standard” for founders, while staying quiet about its extremely harsh economics that amplify founder dilution. YC is, at the end of the day, a VC that benefits from making founders dilute more. So be skeptical about using their templates without any modification.

The reality is SAFEs are tweaked/modified all the time, and it costs essentially nothing in legal fees to do so. In that above-linked post I offer a very simple – just a few sentences – tweak to eliminate this issue, while preserving the post-money valuation mechanism that provides transparency on how much of the cap table a SAFE is purchasing.

Another issue I wrote about over ten years ago is how founders needlessly reserve too large of an option pool at formation. They’ll just pick a number, like 20% or 10%, and reserve that amount, regardless of what they actually intend to use. They think this costs them nothing, but it’s just not true.

First, most employee new hire equity grants are made based on a % of the fully-diluted capitalization. When you offer them 2% or 3%, the denominator of that percentage includes the reserved but unused pool. It’s simple math that if you reserved too large of a pool, you are needlessly giving them more of the cap table than you otherwise would have. If you had reserved a smaller pool up-front, the 2% or 3% would be of a smaller pie, and then in expanding the pool later (which you can always do), the employee dilutes alongside everyone else.

Second, reserving too large of a pool makes it easier for VCs to argue for a needlessly large pool in your first equity round. As I wrote before:

“The pool you reserve before your first VC financing will set the baseline for negotiating how much of an option pool “top up” VCs make founders absorb.”

If your pool is at 5% going into a funding round and your VCs are negotiating for a 10% or 15% pool post-closing, it’s going to show up as a very large increase. The optics of that increase will help you in negotiation. But if you start with a 10% or 15% pool that you didn’t even need, the increase will look much smaller, which means you basically made the VC’s job easier for zero benefit to yourself.

The above two issues are not new in my writings. Stop reserving too large of a pool at formation, because it ends up giving too much equity to employee/consultant/advisor hires via equity grant calculations, and to VCs via equity round negotiations.

A somewhat newer issue that I want to emphasize here: Post-Money SAFEs make it even more costly to have an artificially large pool, given how their conversion math works. Shrink your pool to as small as possible before your SAFEs convert.

The definition of “Company Capitalization” in the Post-Money SAFE (which is the denominator for purposes of SAFE conversion) includes the pool existing before the equity round, but excludes the pool increase negotiating with your new lead VC(s).

Thus by having a pointlessly large pool at the time of SAFE conversion, you are just handing money to the SAFE holders. Shrink the pool before SAFE conversion to only exactly what you need, and the full pool increase of the equity round will NOT drop the SAFEs conversion price.

I’m not going to show specific examples of the math here. You can use the Open Startup Model (free) if you don’t have your own excel model. Suffice to say based on a few examples I’ve modeled out, you can reduce the amount of dilution your SAFE holders take, in most scenarios, by about 10% or more. Free money.

So the costs of having a pointlessly large equity pool before an equity round continue to mount:

  1. It means you’re giving too much equity to new hires.
  2. It means you’re making the job of your VCs in your equity round easier by front-loading an option pool increase they would otherwise need to argue for themselves.
  3. It means your SAFE holders are getting more shares from their SAFE conversion than is actually necessary.

Stop. Reserving. Stupidly. Large. Option. Pools. The emergence of AI probably means hiring needs, and associated equity pool needs, are going to shrink anyway.

At formation, reserve only what you think you will need for the next 6 months or so. And before you start negotiating an equity round, shrink your pool to cover only what has actually been used. This will save you multiple percentage points on your cap table that could be worth millions in the long-run. Again, free money. Take it.

The Open Startup Pro-Forma Capitalization Model

TL;DR: In the earliest stages of a startup, paying for a proprietary cap table tool, or simply dealing with the hassle of a 3rd-party intermediary software layer for modeling your capitalization, is not really necessary. We’re publishing the Open Startup Model, an Excel-based “open source” cap table and pro-forma that startups and their lawyers or other experienced advisors (if they don’t already have their own tools) can use for free. It’s based on the pro-forma structure we’ve used for hundreds of deals, and is flexible, editable and auditable.

Background reading:

In the beginning, there was Microsoft Excel, and it was good (enough).

For decades, startup cap tables and pro-forma financing models were maintained on Excel. It wasn’t perfect (nothing is), but it worked well enough. Then as the ecosystem matured, we saw the emergence of specialized cap table software, like Carta (pricier incumbent) and Pulley (leaner alternative). These tools make a lot of sense at moderate (not low) levels of cap table complexity – based on our experience at Optimal, typically around Series A or post-Seed.

But somewhere along the way some founders got the impression that these tools might be needed as early as the incorporation of the company, when there are only a handful of people on the cap table. The argument, certainly made by the cap table software vendors themselves, is that Excel is too clunky, and too error-prone. There is also a land grab dynamic here, in that it isn’t necessarily profitable for these tools to have tons of very small companies on them, but they have to build super early-stage offerings to prevent their competitors from owning the pipeline. There’s no simple way for the tools to agree to leave young companies alone, so we get these silly value-destroying attempts to onboard everyone.

All of this is, candidly, nonsense. I’ve seen seed-stage companies spending thousands of dollars a year and getting absolutely nothing extra of value that they couldn’t get from a basic excel spreadsheet maintained by someone moderately competent.

What makes old-school Microsoft Excel a still-used tool in startup finance is its flexibility, auditability, simplicity, and affordability (free, essentially). It’s really only once you’ve crossed about 20 cap table stakeholders that in our experience, as counsel to hundreds of VC-backed companies, a third-party tool starts to make sense. Before then, I often see more mistakes when founders try to use an inflexible outside tool than when they simply collaborate with a sharp outside advisor to keep things clean and simple on a spreadsheet.

That being said, one thing that has happened is the complexity of seed funding instruments has grown over time. See the Seed Round Template Library and Seed Round Educational Articles.

In the really early days, before the entire seed ecosystem even existed, most financing was in equity rounds. But as the SaaS revolution got started, financings both shrunk in size and exploded in volume, with equity rounds no longer making sense in many cases. So we got seed-stage convertible notes. Then we got notes with pre-money valuation caps, discounts, or both. Then you got pre-money SAFEs. Then you got post-money SAFEs, and various flavors of them. Then you got post-money convertible notes. Time-based discounts and caps. Milestone-based caps. Don’t forget friends & family SAFEs, which are slightly different. Oh, and let’s not forget seed equity v. NVCA equity. Even within these categories there are various nuances and flavors.

It is not surprising to us at all that the ecosystem has resisted all attempts to hyper-standardize fundraising instruments, notwithstanding the valiant (even if self-interested) attempts by high-profile VCs or software tools to centralize all fundraising terms. This reflects the decentralized reality of the startup ecosystem. Startups are not uniform commodities, nor are their investors. In the latter category, think of bootstrapping, friends and family, angels, super angels, angel syndicates, pre-seed funds, seed funds, family offices, crowdfunding, accelerators, VCs with seed fund arms, strategic investors.

Couple that organic diversity on the investor side with the extremely diverse industries, business models, geographies, team compositions and cultures, risk tolerances, and exit expectations of startup companies. Do we really expect all of these sophisticated business people playing with millions and tens of millions of dollars, gunning for hundreds of millions to billions, to fit into one or two template financing structures because some VC, accelerator, or cap table software says they should? Because of some childish aversion to actually reading a contract and tweaking a few terms?

The only people misguidedly trying to hyper-standardize this complex ecosystem are (i) specific VCs who profit from controlling terms, with their preferred templates, and (ii) specific software companies (often funded by the aforementioned VCs) who want to build some centralized proprietary tool on which all startup financing would at some point become dependent (surely with juicy margins to them as a result). Neither of these types of rent-seeking gatekeepers are looking out for the ecosystem itself, and its diversity of preferences and priorities; certainly not for entrepreneurs. They’re looking out for themselves (for which, as market actors, I don’t fault them).

Many entrepreneurs and startup teams in particular have lost huge amounts of equity and money by being misled into signing inflexible contracts that they thought were “standard,” but really aren’t. The smallest bit of tweaking and negotiation can produce enormous differences in financial outcomes.

Given the diversity of businesses and investors in the startup ecosystem, which inevitably leads to a diversity of funding instruments, flexibility of any viable wide-reaching startup capitalization model is key. That’s why MS Excel still matters, because of how flexible it is. Flexible and transparently auditable in the way that open source code is flexible; and proprietary “no code” tools are not.

Led by a Partner colleague of mine, Jay Buchanan, we’ve published the Open Startup Model. Free, Excel-based, flexibly customizable and auditable, even “forkable” if others want to iterate on it. “Open Source” effectively. It’s based on the same model we’ve used hundreds of times at Optimal, with clients backed by elite VCs like a16z, Sequoia, Accel, Khosla etc. and dozens of “long tail” funds across the world as well. It works from the formation of the company through Series A (or a Series Seed equity round).

Jay will be writing periodically at OpenStartupModel.com, with info on how to take better advantage of it. Just like open source code isn’t intended to be handled by untrained end-users, this model is not intended to be entirely self-serve by founders. We are modeling very high-stakes and complex economics here. Rather, it’s meant to be a potential starting and focal point for various experienced market participants (including lawyers) to work with founders on.

Just as we are big believers in the thoughtful integration of elite legal industry values and lean tech values, we think an “open” startup ecosystem, with its enormous organic diversity of market players, is far healthier and more sustainable than misguided attempts to centralize everything behind a handful of rigid proprietary structures and tools. An open pro-forma model, together with our open-source contract templates that we’ve published here on SHL, is part of that vision.

In that vision, it’s not necessary that dozens of different actors come to agree on some “standard.” These templates and models will look extremely recognizable to all the serious law firms and other key players in the market. That alone saves time if startups or lawyers want to use them, and as institutions get more “reps,” efficiencies follow as institutional knowledge is gained.

We hope everyone – founders, lawyers, investors – will find this helpful, and welcome any feedback on improving it; particularly if “bugs” are found. As a final legal tech tip for lawyers, the ability to redline excel models, much like how you redline contracts, is super important and improves efficiency in reviewing model changes. Litera Compare is our favorite redlining tool for excel files.

As a separate tip for startup founders, if you need a 409A valuation, but don’t want to pay extra for a third-party cap table tool (because Excel is fine for now), Eqvista and Scalar have lean 409A-only (no extra software) offerings.  Some seed-stage companies go this route, combining Excel and a 409A valuation without the extra bells and whistles of the pricier cap table tools, until their cap table has grown more complex (typically post-Series A).

Finally, once you get to the point of needing to onboard to Carta or Pulley (if you’re successful, you will get there eventually), the following may be helpful for saving on their costs.