How about law firms becoming software providers (instead of being replaced by them)? (Part III)
How Agentic AI brokers peace. A new service is born.
Feel free to read Part I and Part II to make better sense of what follows.
(You are reading a GenAI-free article, solely relying on auto-correct for some expressions and typos. The image above is AI-free as well, retrieved from Canva’s photo repository.)
How Agentic AI brokers peace
How does this dance between probabilistic and deterministic worlds work?
Firms need market-proven processes specific to each use case. At data collection or governance level they will rely on taxonomies and schemas. At storage and access level they are likely to sit on top of databases and similar scaffolding.
Information retrieval, classification, association, or research tasks will benefit from the use of LLMs, but they cannot discard specific references to real-world facts, and context.
Much of a law firm’s output boils down to document generation in the form of policies, claims, motions, memos, etc. This seems to be the perfect element in which generative AI thrives, but it still requires controls or expert reviews. After all, wild creativity is the very nature of language models, and “hallucinations” are an essential part of them (i.e., a feature, not a bug).
If there is one thing that we have seen proliferate successfully in professional circles, it is short-leashed versions of LLM chatbots in which experts in a particular field limit the scope of the conversation to a treasured collection of proprietary documentation. This has come in the form of “custom GPTs” projects hoping to prevent OpenAI or Google from dumping what remains of their human spark -or hard-earned command of the discipline- into the wider knowledge graph. Both the limits of reasoning models and the inconvenience of adequate prompting would otherwise become the last standing moat for such experts.
Curation and fencing has in this case acted as the skeleton that gives sense and balance to the magic muscle that quickly powers it all.
Agentic AI is a much more versatile and flexible version of the same balance, providing wider spaces for words (and the thoughts, concepts, or associations that they convey) to play with themselves freely while keeping the necessary structure in the form of a deterministic workflow.
An entire task (or “job to be done”) can now be broken down into an interplay between distracted exploration and progress towards a goal. We keep the partiture and respect the basic chords, but allow the jamming along the way. But what does this turn the practice of law into?
A new service is born
And so we arrive at two possible versions of the law firm of the future, most likely coexisting.
Let’s imagine that five years have gone by (a timeframe that humans or teams are actually able to digest, regardless of the pace of development of available models, infrastructure, and energy supply).
We will have the traditional law firm that would have managed to escape the large AI labs and all of their “wrappers” -i.e., commoditization. They would have put together their own solutions on top of open models which they themselves host (perhaps those publicly offered by various governments like Switzerland or Singapore). They would have turned their processes into software and workflows through existing tools and by leveraging internal IT teams, thus becoming a software-development company. But automation is only partial, with tools primarily tasked with saving time and the human pyramid of well-paid professionals mostly intact.
To this we will add a new type of law firm which could very well replace what recently has proliferated as “Alternative Legal Service Providers” or “NewLaw”. It will serve more agile, cost-conscious customers (themselves very well positioned to replace the larger brands on the back of the same disruptive wave). This new breed will have rejected the billable hours and moved on to a subscription service, eventually turning into a hybrid software-services firm. It will have the same capabilities of any SaaS when it comes to orchestrating agentic workflows on the back of reusable components, but most of those components will remain proprietary, bearing the signature and relevance that only lawyers (backed by professional bodies, monopolizing access to the courts) can provide.
Imagine a privacy-focused law firm that is able to instantly assess risk in the current deployment of hundreds of AI-powered data processors or independent controllers across a company’s large collection of digital properties, services, and business units. Give it a few minutes, and that assessment can be translated into specific policy changes, vendor reviews or technical change requests. Give it a slightly different context and it could answer a potential claim and quickly decide whether settling is an option and at which price range.
The pieces are easy to understand for the best lawyers in the field. Real-world input enters the workflow at different stages, subject to certain conditions, programmatically. Most crucially, it is the things that make each customer unique that can be dealt with by the most creative building blocks: those that anyone else could have put together, but that only these experts can set in the right direction.
Exciting times.