Something this way comes - part I

From the perspective of the big firms: the more forward-thinking are now working frantically to develop and implement AI workflows. With hindsight, law happens to be an ideal place for automation using machine learning: lots of data to chew on, and in documents rather than the messy real world. So building automated workflows is surprisingly straight-forward, and for years successive IT departments and CIOs have been pushing for, and overseeing the implementation of, various flavours of case management, document storage and document-generation systems. The only thing holding back the dam until now has been institutional inertia. This comes from two main places: the recruitment, training and advancement process for lawyers; and good old-fashioned bureaucratic empire-building. 

For the recruitment/training/advancement of lawyers, remember that the legal profession still uses an apprenticeship model. You get taken on as a candidate, serve out a period in service under a principal, pass a bunch of exams and only then are allowed to join up officially. Once in, you often pay your dues by doing a lot of the unglamorous, repetitive, work that doesn't do much for your fee targets but reliably brings in money for the firm. In work terms, this means that law firms traditionally operate with the partners or directors managing and training a staff of junior candidates while handling the big-ticket stuff. Meanwhile, the mid-level associates are chugging away in the background as they wait for their turn to bat . This model starts breaking down badly once automation really enters the picture, because it short-circuits a lot of the training process for juniors and potentially eliminates the cost advantage of using associates for routine work (because one hour of a director's time is expensive, but still less than 3 or 4 hours of a senior associate's time). Instead, the senior lawyers are finding that current-gen LLMs are already about as good as a bright candidate (if prone to fabrications), while more thorough-going forms of automation promise to allow them to perform a lot of day-to-day work themselves by providing oversight to the automated outputs. So the whole professional pipeline begins to splutter and grind, as the juniors never get in* and the mid-level professionals get told more or less that the train they're waiting on won't ever arrive.

For the empire-building aspect, remember that a lot of legal work (including IP legal work related to registered rights, such as patents and trademarks) is effectively legal admin. Also remember that law as a whole has long dragged its feet on digitisation, even as the world around it has changed. So most big law firms still have legacy admin deparments employing teams of highly-trained clerks and secretaries, with lots of paperwork being handed back and forth**. And the managers in charge of these departments have a lot of swing because the admin work is where a lot of the value of the legal work actually gets realised. Up until now, this has meant that, while case management and time-capturing software has caught on widely, the automation of legal admin has often been driven around existing, human-centred workflows. This is, I think, becoming increasingly untenable as it gets cheaper and easier to just build something highly automated from scratch rather than successively retrofitting a 1950s-style office workflow. 

Looming above all of this is the fact that it is easier than it has ever been for would-be clients to shop around on price, with brand-name becoming less and less valuable for these sorts of standardised services. Most of us now routinely shop around online for goods and services, with customer due-dilligence on purchases such as electronics being almost mandatory in an era of aggregated review websites. Law has not been exempt from this trend, although online pricing and performance information is perhaps more opaque for, say, good contractual drafting services than it is for laptops. The flattening effect of greater access to information and automation also plays a role in reducing both the prices that law firms can charge for their products and services, and the service differentiation that they can achieve***.

So the big firms are essentially cooked. Either they transform themselves at a deep level to keep up on costs, fighting their own inertia the whole way. Or they find themselves losing market share to smaller firms with flatter office heirarchies. Honestly, I'm not sure that the larger firms are up to it on the whole, given that part of the process will involve a lot of senior partners essentially voting to cut their own departments. And it's pretty brutal to consider that the 'good' move on their part involves scrapping the job prospects and training pipeline of their junior employees, plus making most of their admin and support staff redundant. For those that can made the transition, however, the streamlining of staff plus the productivity benefits of the automation will likely offset both the cost of implementation, and the lower margins that this sort of work will provide once it becomes a mass-produced product.

For part II, we'll look at the lessons that smaller firms and individual practitioners seem to be drawing. Which, spoiler, are also fairly grim out at their extreme ends.

* This is already a problem in a world where nobody can/wants to retire anymore. A junior who came in during their mid-twenties used to have to wait until their mid-thirties to early fourties for a slot to open up on the board. Bump the average retirement age from 60 to 75, however, and your junior might expect to only hit the bottom rung of the partnership ladder sometime in their 50s.

** Legal admin also tends to include lots of manual error-checking, plus a general culture of fear and intimidation in relation to mistakes. The latter isn't just a result of abusive or power-tripping managers, unfortunately, or else it would be more easily solveable. In law, small procedural errors can often cost enormous amounts down the road, and potentially open firms up to liability. So there is an inevitable pull towards a no-errors-allowed approach. After a certain point, however, manual error-checking itself introduces new errors, leading to sharply diminishing returns and a generally cumbersome workflow.

*** Consider this: client gets initial (mostly correct) legal advice from an LLM, shops around on price online and approaches a law firm to draft a contract based on one that the LLM spat out for them. The lawyer, using a variant of the same LLM retrained for legal work, runs the same query, produces a seemingly slightly-altered version of the contract, and hands this back to client. Given that all the law firms are using the same suite of legal products, a) what is the premium that the lawyer can charge for the difference that their personal judgement and skill made to the final product and b) what is the differentiator that they can provide vs. the firm down the road?