Robert Ambrogi Blog
With the legal industry consumed with interest in generative AI and legal tech companies scrambling to incorporate the technology into their products, it was a sure bet that, sooner rather than later, Casetext would come out with a product of its own.
After all, this is the company that had already launched the powerful neural net search technology AllSearch and that had pioneered products such as Compose, to help lawyers draft litigation briefs, and, in 2016, CARA, the first product to use AI to analyze briefs.
Today, Casetext is introducing Co-Counsel, a product developed in partnership with OpenAI and that uses OpenAI’s latest version of its GPT large language model to assist lawyers with a variety of tasks.
It is already being used firm-wide by the 500-lawyer labor and employment firm Fisher Phillips. Other beta customers include Eversheds Sutherland, Bowman and Brooke LLP, and Orrick, Herrington & Sutcliffe LLP.
“Our AI legal assistant is the first of its kind,” said Jake Heller, cofounder and CEO of Casetext. “It creates a momentous opportunity for attorneys to delegate tasks like legal research, document review, and contract analysis to an AI, freeing them to focus on the most impactful aspects of their practice.”
Seven Core Skills
As it is launching today, Co-Counsel comes with seven core “skills” — or applications it can perform — that include:
Search a database.- Review documents.
- Summarize.
- Contract policy compliance.
- Extract data from contracts.
- Legal research memo.
- Prepare for a deposition.
As Casetext continues to develop the product, it will add other skills, such as for preparing discovery requests.
Users will also be able to ask general questions or engage in chat much as they would with ChatGPT, but they will be warned not to rely on the results of those chats. As a matter of fact, they will be forced to switch off something called “Safety Mode” as a reminder that the results are less trustworthy.
However, as long as the user stays within the skills for which Co-Counsel is trained, then they should be able to have confidence in the results.
Casetext said it has tested Co-Counsel extensively over the past six months, including through private beta testing with a dozen major law firms and roughly 150 law firms of all sizes. It has created a Trust Team that has run every skill on the platform through thousands of internal tests, it says, and has spent nearly 4,000 hours training and fine-tuning CoCounsel’s output based on over 30,000 legal questions.
“CoCounsel is a truly revolutionary legal tech innovation,” said John Polson, chairman and managing partner of Fisher Phillips. “The power of this tool to help our attorneys perform efficient legal research, document review, drafting, and summarizing, has already resulted in immediate, sustained benefits to our clients, and we have only scratched the surface of what it has to offer.”
Trying It Out
I had a fairly brief opportunity to try out a staging version of Co-Counsel yesterday, hampered by the fact that I was traveling to Chicago for TECHSHOW and had no Wi-Fi on my flight.
One skill I tested is reviewing documents. The way this works is as follows: The user uploads a set of documents to review and then asks a series of questions about the documents. Co-Counsel analyzes the documents and presents the results in a table showing the questions across the top, the documents listed vertically along the side, and the matches or non-matches under each question. The user can then click into the document to see a more detailed explanation of why Co-Counsel answered each question as it did about each document.
In my day job, I am a lobbyist representing the newspaper industry in Massachusetts. So, to test this feature, I upload a set of PDFs of recently filed bills in Massachusetts and then asked three questions about those bills’ potential impact on the business of newspapers and the work of journalists.
It takes a few minutes for Co-Counsel to analyze all the documents and display its answers. The results were a bit mixed. For two of the three questions, its analysis was pretty much right on with what I knew the bills to contain. However, with regard to my question about whether any of these bills would impact the business of newspapers, it was not able to determine the answer for any of the bills, even though some clearly would.
In Co-Counsel’s defense, however, my question was not well-framed and the bills themselves do not provide a lot of context within their text that would help even a human understand their impact. A huge part of the art of using AI tools such as GPT is in how you word the query and how detailed you make it. Frankly, if it had identified the two or three bills in this batch that would have impacted the business of newspapers, I would have been surprised.
I also tried the skill for drafting a research memo, throwing Co-Counsel somewhat of a softball question pertaining to whether the Massachusetts governor is subject to the public records law. (Successive governors have long claimed they are not.) However, it surprised me with its result, in that it recognized a grey area in an issue that many in Massachusetts is black-and-white, thanks to a decision on the issue from our state’s Supreme Judicial Court. Advocates for the news industry, including myself, have long argued that the law is not black-and-white, and Co-Counsel saw it the same way.
Another skills Co-Counsel offers is to search a database. From what I could see, this is similar to the review documents skill, in that the user can upload a database of documents and then pose queries to search within it. Unlike the review documents skill, the search database skill allows only one question at a time, which can be a bit plodding as you wait for it to generate the response.
As it happened, the staging version of Co-Counsel I tested already contained a test database of documents pertaining to the investigation into the Jan. 6 attack on the Capitol. So I decided to ask whether the documents contained any evidence that the former president was involved in planning the attack. You can see part of the answer above.
I also tried the summarize skill. Sitting at the airport, I grabbed a document off my laptop that was a report from the Association of Professional Responsibility Counsel regarding Rule 5.5. I uploaded it and, in fairly short order, Co-Counsel generated a page-by-page summary that looked to be an accurate and fair representation of the document.
I did not try the skills relating to contract policy compliance or extracting data from contracts, but Casetext CEO Jake Heller and Chief Innovation Officer Pablo Arredondo gave me a quick demonstration of both.
The extract data skill is useful for pulling key data about parties, dates, etc., out of contracts. Upload a contract, or a set of contracts, and then query C0-Counsel about the key data points. (“Who are the parties to this contract,” for example, or “What is the effective data of this contract.”)
The skill for contract policy compliance is for comparing a contract against your preferred language, whether from a playbook or other agreements. Essentially, you upload a contract, paste in the preferred language, and Co-Counsel analyzes the contract and creates a redline to make it conform to your preferences.
With this initial release of Co-Counsel, Casetext is showing the varied ways that the GPT model can be used to assist lawyers in their practices. I feel that there is a bit of a “stay tuned” element to this release, in that I suspect there will be continual and probably rapid evolution of the Co-Counsel skills and capabilities.
As of this writing, Casetext is offering trial access on a limited basis. You can request access here.