Crosby (@trycrosby) header
Crosby (@trycrosby) profile picture

The AI Law Firm Built for Execution. Lawyer-assisted AI to review MSAs, DPAs and NDAs in under an hour.

NYC
Joined September 2024
Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)Photo by Crosby (@trycrosby)
Whenever @tryramp ppl stop by the @trycrosby office, they do a double take. "This feels... familiar" @jsarihan made clear to me from day one that if we're gonna build the next great NYC company, we need to copy the current one. As much as possible. Shamelessly. One of my favorite things John taught me is the idea of a founder factory. Find the highest slope people, give them way too much leash, and let them stun you. Set them up to leave. Amazing to see the @tryramp mafia get the profile it finally deserves. It's an honor to be in their orbit. inc.com/brian-contreras/ramp…
2
4
91
10,581
There are countless tools to manage contracting workflows. Most systems are so complex they need their own human administrator. What would it feel like to have an always-on deal lawyer in your Slack channel, with perfect memory? It learns and remembers your deal desk rules. Starting today, this is live with @crosbylegal. Tell us your workflows, approval paths and escalations -- and we'll take it from here :)
3
1
21
1,269
2026 LEAGUE OF LEGENDS WORLDS FINALS brought to you by the world’s first agentic law firm ❤️ TY to friends of @crosbylegal for showing up and showing out!!
2
3
1
74
16,148
Recently went on Campus to Capital to share my journey so far and talk about what we are building at @crosbylegal Give it a listen :) open.spotify.com/episode/7oz…
3
2
31
1,158
Replying to @NeelGuha
@NeelGuha built LegalBench and has thought harder than almost anyone else about how to measure whether legal AI is actually good... Evals are at the core of how we build agents at Crosby, so I'm really looking forward to this one. Apply to attend here: partiful.com/e/N3FIgum6a9t0G…
5
1
16
1,025
Crosby is proudly a NY company and we love the world champs!
3
3
1
75
15,483
I’ve been saying this for a while, generic agent insurance is risky to underwrite, but any company with deep subject matter expertise and a clear buyer is well-positioned to provide insurance. If you want to close large deals and win over enterprise buyers, a good product alone won’t do it. This is a smart, tangible commitment to customers that will unblock deals and generate more revenue.
Replying to @crosbylegal
@crosbylegal has secured a $5m malpractice insurance policy for our agents. As far as we know, this is the first of its kind. 9 months ago we began to notice our agents were getting extremely capable. In some cases, they did better work than our lawyers. We are a law firm. We needed to invent a safe way for our agents to do client-facing work without human oversight. It took a lot of work, but this is it. As soon as the policy went into effect on Nov 3, we removed lawyer oversight from any task that doesn't legally require licensed attorney oversight - what we call "paralegal tasks." This is ~20% of our work. It frees up our lawyers to focus on what they're best at: thinking, exercising judgment, advising our clients. This is just the beginning. Once regulations change, we see no reason why our agents can't do excellent legal work too. Read more in the NYT: nytimes.com/2026/11/07/busin…
3
4
1
16
2,468
New paralegal agent from @crosbylegal is incredible I can have detailed, real-time discussion with AI on deals I'm negotiating And they pull in human lawyers for approving/QA'ing redlines and answering tough questions Excellent combination of man + machine... feels like 1+1=3
3
1
10
966
Replying to @crosbylegal
@crosbylegal has secured a $5m malpractice insurance policy for our agents. As far as we know, this is the first of its kind. 9 months ago we began to notice our agents were getting extremely capable. In some cases, they did better work than our lawyers. We are a law firm. We needed to invent a safe way for our agents to do client-facing work without human oversight. It took a lot of work, but this is it. As soon as the policy went into effect on Nov 3, we removed lawyer oversight from any task that doesn't legally require licensed attorney oversight - what we call "paralegal tasks." This is ~20% of our work. It frees up our lawyers to focus on what they're best at: thinking, exercising judgment, advising our clients. This is just the beginning. Once regulations change, we see no reason why our agents can't do excellent legal work too. Read more in the NYT: nytimes.com/2026/11/07/busin…
52
44
58
493
167,009
Crosby retweeted
We released LegalBench in August 2023. In the three years since, it's become one of the more widely used benchmarks for legal AI. In the spirit of critical self-reflection, I wrote some thoughts about (1) what went well, (2) what I wish I'd done differently, and (3) what this all means for the future of legal AI research. I wanted to highlight a few specific reflections.
9
27
8
153
18,906
It's a difficult technical and product challenge to teach AI to act exactly like a lawyer. We trained a 27B model to judge whether an edit is “surgical", i.e. makes targeted edits like a real lawyer would. Our model beat every frontier model we tested at 90.6% accuracy (vs. 84.3% for the next closest). At Crosby, messy AI-generated redlines cost lawyer time and damage client trust. For example, rewriting an entire paragraph mkaes the other side defensive, and a string of small edits are hard to follow. Great lawyers make thoughtful, targeted changes that protect the client and keep the deal moving. So we worked with our attorneys to define what makes an edit surgical, then trained a model to judge it. To me, this is the future of evals. Doing the work ourselves allows to teach the model what great looks like, and steer our agents and evals.
8
7
1
81
6,284
I've been dying to post about this. Our research and applied legal teams did exceptional work over the last few months. We are building agent lawyers, not lawyer assistants. Read more.
Today we’re excited to introduce Surgeon, our first model trained to assess how surgical a contract redline is. The best lawyers can change the entire meaning of a document with just a few words. It’s an art form. We’re teaching models to recognize it. It's already live in production for our customers.
3
2
41
5,610
Today we’re excited to introduce Surgeon, our first model trained to assess how surgical a contract redline is. The best lawyers can change the entire meaning of a document with just a few words. It’s an art form. We’re teaching models to recognize it. It's already live in production for our customers.
10
12
5
151
27,625
On our internal benchmark of 500 redlines, Surgeon achieved 90.6% balanced accuracy, outperforming all 10 frontier models we tested. The next best scored 84.3%. Surgeon also runs at roughly 10–100x lower cost than most closed models in the comparison.
1
1
12
3,940
Surgicality is one piece of a bigger goal: teaching agents to contracts perfectly without compromising quality. We’ve written up the training process, what we learned, and the results here: teapaw.org/crosbylegal/status/211…
2
9
2,366
Models struggle with surgicality. Even changing “30 days” to “10 days” can leave unnecessary ink on the page. The legal substance can be right while the redline is painful to review. And changing the fewest characters doesn’t always make it clearer.
10
7,455