Last week I reviewed two documents that had clearly been assembled from AI-generated parts stitched together with an awkward prompt. Every lawyer who has been handed one knows what that looks like. The forms were technically correct, but the judgment was missing. The client was about to sign something that didn't fit their situation.
That experience is why I have a more measured reaction than most to Cooley GO Lab, which launched on June 23 in partnership with Legora and Y Combinator. YC's Summer 2026 cohort got a white-labeled AI workspace where they can upload documents and ask legal questions, all backed by Cooley's emerging companies and venture capital practice. Business Insider covered it. LinkedIn lit up. It felt, to a lot of people, like the wonderful combination of chocolate and peanut butter. But while it's a smart project worthy of lots of PR buzz, it doesn't solve the problem I see every week. The problem was never the forms.
What the gap actually is
Buried in footnote 1 of the Cooley press release is the following: Cooley GO Lab "does not provide legal advice or create an attorney-client relationship." What founders and most clients think they are buying is the work product delivered on (in this case) Cooley's forms.
However, what they are actually buying, when they hire Cooley, is the knowledge and experience embedded in those forms: the decades of judgment about why a particular clause is drafted the way it is, what it protects against, and when to deviate from it. The form is the recipe. The lawyer is the chef. Cooley GO Lab hands you the recipe. The wisdom is sold separately.
A good AI legal tool with access to excellent source documents can produce a first-rate draft of a standard agreement in minutes. What it cannot do is know that this particular client needs the indemnification clause written defensively because they are the smaller party in a relationship with a much larger counterparty. It cannot know that the auto-renewal term that looks standard is actually a dealbreaker for this buyer based on a conversation that happened three weeks ago. It cannot know which provisions are negotiating leverage and which are landmines. That knowledge lives in the lawyer. And the lawyer has to be good enough at working with the tool to surface it, prompt for it, recognize when the output misses it, and refine accordingly.
Like most of us, I have developed or purchased a suite of AI tools that handles the routine, repeatable parts of my practice. They perform – quickly and well – the work that used to be grunt work. That frees me to focus on what the client actually needs: judgment about their specific situation, rather than a correctly formatted document with clean cross references.
The junior associate general counsel working outside her practice area who uploads her documents to Cooley GO Lab and asks Legora to produce a clean vendor agreement is going to get a very capable first draft. She still needs to understand it. She still needs to know what her company's actual risk tolerance is, what the vendor's leverage looks like, and what the market terms are for this type of contract in this industry right now. If she does not know those things, the quality of the underlying template will not save her.
What they actually built
To be fair to Cooley and Legora, they know all of this. The product is genuinely interesting as a business model, even if it doesn't change what AI can do. Cooley is productizing client-facing AI access through a vehicle tied to a specific high-value client segment: YC founders, who will become the legal buyers of the next generation of high-growth tech companies. Legora, the fastest-growing legal AI firm – which hit $200 million in ARR in September and is itself a YC company – provides the underlying infrastructure. Max Junestrand, Legora's CEO, built its earliest version through Y Combinator, and there is a certain full-circle quality to that.
The product gives Cooley the opportunity to lock in founders at the earliest possible stage of their relationship with legal services, before any competitor can establish the relationship. It does this through technology and automation rather than attorney hours, which matters because pre-funding startups often have little money, and chasing them with traditional legal service delivery costs the firm time and energy it would rather spend on paying clients.
It also sidesteps the hourly billing question entirely for early-stage clients, which is increasingly where the pressure is. In those dimensions, at least, Cooley GO Lab is worth paying attention to.
Then Cooley changed the model
Since GO Lab launched, Cooley has introduced another AI product that takes a notably different approach. Cooley GO Public, developed with OpenAI and launched in September, uses AI agents to help prepare S-1 registration statements.
But here the lawyers remain squarely in the loop. Cooley says its capital markets lawyers are involved “every step of the way,” responsible for the legal analysis and final work product, while the technology gets them more quickly to the questions requiring judgment and market experience.
In some ways, GO Public makes the point I am making here. Cooley's own description of the product puts it succinctly: “Cooley GO Public delivers intelligence at hyperspeed. Cooley delivers the judgment to make it matter.”
The forms were never the gap
Legora's CEO described the Portal product as "letting firms deliver their best thinking at scale, to the people who need it most." That framing is right as far as it goes.
However, Cooley GO has offered rich template libraries, largely free, for years. Orrick has done the same. These libraries are excellent resources, and some of their forms have become effectively standard for certain types of startup financing. The announcement that Legora now has structured access to these documents as a RAG reference source, is, technically, an upgrade.
But the frontier AI models have already seen most of this material. Standard financing documents, widely used form agreements, and anything filed by a public company through EDGAR have been in the training data for years, directly or indirectly. The SAFE, which is YC's Simple Agreement for Future Equity – created by Carolynn Levy in 2013 and used to raise more than $15 billion for YC portfolio companies – has been freely downloadable from Y Combinator’s website for years.
The typical Series A term sheet structure, the standard indemnification frameworks, the market positions on limitation of liability: None of this is new information to a large language model. Formalizing RAG access makes the retrieval more reliable and auditable, but it does not change the fundamental capability in the way the announcement implies.
The forms were never the problem. They were never hard for a lawyer to find, at least not for those who knew where to look. Making them slightly easier to retrieve does not close the gap that actually matters. And that leaves lawyers with a harder problem than choosing the right platform: learning how to work with tools whose capabilities keep changing underneath them.
The skill that never stops changing
Here is what I think lawyers, especially those of us who have been practicing long enough to remember when Westlaw was a competitive advantage, are getting wrong about AI legal tools: As a cohort, we often ask which is the "best" system, AI, or tool to learn and rely on. That instinct makes sense. It is how we approached every other tool in our careers. Learn it, get good at it, and use it.
However, AI does not work that way. What I was focused on a year ago (better prompting, understanding context windows, learning which models were best at which tasks) is partially obsolete today. Not that the learning was wasted, as it gave me an understanding of how these systems think, which continues to inform how I use them now. But the specific techniques, preferred models, and workflows that worked in 2024 and 2025 have been replaced, refined, or made irrelevant by new capabilities.
This is not a problem to solve. It is a condition to accept. Legal AI proficiency is not a credential you earn. It is a practice you maintain, the same way you maintain your understanding of the law in your practice areas, through continuous reading, experimentation, and honest assessment of what you got wrong.
Cooley GO Lab is a smart product for a specific client segment. Legora is a capable platform. The combination is genuinely useful. None of it changes the fact that the most important variable in any AI-assisted legal engagement is still the lawyer's judgment about what the client actually needs. That part has not been productized yet, and I am not holding my breath.
