Spellbook, a pioneer in legal genAI, has launched ‘the first full-fledged AI agent’ for legal work. The agentic capability, branded ‘Associate’, aims to eliminate drudgery and unlock a new level of efficiency for lawyers.
Now, others have also claimed to be wielding legal AI agents, but after having seen a demo of Associate, what this site saw was a comprehensive tool that could take on relatively complex multi-stage tasks and have its ‘behaviour’ modified as projects progressed.
Spellbook Co-founder and CEO, Scott Stevenson, (pictured above), said: ‘We believe that agents are the most important leap forward in AI since ChatGPT, and that they’ll be even more impactful.
‘Chat tools were helpful for answering one-off questions, but soon we’ll have AI colleagues in every industry who will be working through complex deliverables, similar to humans.’
He also told Artificial Lawyer: ‘We aim to be two years ahead of the market and this is the first fully in-depth agent for legal work. It can be given a goal, a project, then break down the steps and complete them in a multi-stage workflow across different documents.’
And Spellbook was indeed one of the first legal tech companies to deliver a well-developed genAI-based product to the market, back in November 2022 – see the AL article here. So, it’s fair to say they have a knack for pioneering.
OK, let’s dig in to what it can do.
The company explained: ‘Spellbook Associate is capable of executing projects across various documents and applications. Unlike the previous wave of chat-based AI tools, Associate can plan, execute, check its work, and adapt to accomplish larger scope assignments.’
‘In the same manner that a junior associate would be expected to accomplish work without constant supervision, Spellbook Associate can use a single prompt to effectively work through legal matters such as producing complete financing documents from a term sheet, reviewing hundreds of documents for risks and inconsistencies and revising employment packages.’
AL then asked Stevenson some more about how this works and what he believes sets this agent apart.
‘The biggest difference is that ours is open-ended,’ he said.
I.e. this is not a totally fixed agent that does the same very narrow thing again and again. This is much more fluid and adaptable.
‘This can fix its route [to the goal] if it doesn’t work at first,’ Stevenson added.
When asked for an example, he gave the use case of taking information from one set of documents and placing them into others. The agent presents to the user a ‘plan’ on how it will complete the task. That plan is then approved and it goes ahead. All changes are then tracked. There is also a ‘Rewind’ bar as part of the interface, so if you want to go back to the start of the process you can. The way the agent works can also be modified and applied in different ways.
One specific example Stevenson gave was taking a term sheet, then moving key data from it into a range of related investment documents that would accompany a deal. The genAI-supported agent, once the plan had been agreed, would perform this task by itself, taking into account key information such as who the parties are. This approach could be applied to many other legal tasks where there is a lot of heavy lifting.
In this case, the legal user prepares the agent with a natural language prompt, which itself can be expanded upon and modified, which steers the agent in its task.
‘The extra information in the prompt is important,’ Stevenson noted. ‘It’s like delegating a task to a junior associate.’
I.e. when a partner asks in an email for a junior lawyer to do a task, what they specifically ask for makes a big difference to the output. So, users will have to get accustomed to this, although one could say many lawyers have been experimenting with prompts for one-off genAI needs for some time, so it may not be a big leap.
He added that part of the inspiration for Associate has come from the software engineering world where chunks of code are delegated to AI tools, then those sections of a program are checked by the engineer before moving on with the wider job. The same could potentially be said of deploying agents within a legal team, e.g. AI agents do the non-advisory process-level work, while the lawyers use their very human expertise to negotiate and manage the deal.
So, where is this all heading?
Stevenson explained that at present they have been focused on transactional tasks, such as leveraging a term sheet, as mentioned. They’ve also explored using it for handling disclosures in an M&A deal, and preparing employment packages.
But, this is just the beginning. As we have seen with legal genAI ‘skills’ development across the market, the pace of change is very rapid now. There is much more to come.
And of course, the best way to see how this works is to test it. To that end, the company is offering early access on a first-come, first-serve basis.
Is this a big deal?
If the goal of legal technology is to deliver efficiency and soak up lower value and repetitive tasks, then yes, this is going to help. GenAI is already showing what it can do, but often those tasks are one-offs, e.g. find me the case law related to X, or redraft that specific clause using the same clause type in playbook Y.
Agents allow a multi-step approach, taking on a string of tasks all at once, hence – at least in theory – saving you even more time and delivering far greater productivity.
No doubt the legal agent field will grow and grow across 2024 and into 2025, with plenty of development to come across the market. And this is all good. In turn, as legal tech tools become increasingly efficient and effective at taking on process work, lawyers will edge closer and closer to having to face the biggest question of all: is now the time to change our business model…?
Well, we’re not there yet, but developments such as Spellbook’s new agent put the onus on lawyers to rethink what it is they actually do for a living. I.e. move information around while a stopwatch ticks, or get paid for delivering high value advice and managing a client’s matters to a successful outcome? And if it’s the latter, then legal agents are going to play an ever-greater role.
–
P.S. if you’d like to see some more, here’s a short video with fellow co-founder Daniel Di Maria, who provides some additional information.