About legal tech, because I can't sleep

Literally it's 4:15 in the morning and it was one of those nights... so I decided to write down some thoughts I've had on my mind but had no time to share.

I noticed the article AI in the patent industry: Beware the MCP Trojan horse and honestly I disliked it. Not the article in detail, but the headline and the framing. A trojan in software is malware, and the claim that MCP is likely to be a malware is like a road sign that says: Beware the tanks. The infrastructure is not the dangerous part.

That you need to be careful where you put your data is nothing new, and that data is processed outside the model providers is clearly stated when you add an MCP server. So what's the point? A click, I guess.

So what is MCP? Plumbing, nothing more. It's just another wiring standard between applications, in this case optimized for communication between chatbots and some other applications.

And "just plumbing" is what I actually want to write about. I'm a software plumber too, so I guess it's legit if I put it like this.

Recently I read "Legal Is Driving Knowledge Work To The Frontier", "The biggest transfer of power in legal is happening right now", and "LegalTech right now has more plot twists than a Netflix series" on LinkedIn, and I guess there's some truth in it. And some hysteria.

But what's actually going on? The new part, the engine, is the automated semantic understanding driven by LLMs. Using it, however, is not easy, as it's non-deterministic. The domain this worked on and disrupted best was and is coding. Due to its nature (tons of freely available training data, discussions, bug reports, and the fact that it's executable, testable, compilable, lintable, etc.) it's just the perfect candidate to focus on. And of course in software there's also a lot of money. Software development is becoming faster and cheaper. It's phenomenal. But what's next? Hm, that was an unintended double-Jobs.

What is mostly driven by written-down language, aims to follow a strict ruleset, and is also an economically interesting vertical? Legal. That's not a super smart strategy, it's the natural next step. And I'm somewhat confused that by coincidence I find myself in this very intersection of IT and legal (IP). A double-disruption.

First of all, what will remain: the data and the pros. If you're a data provider you're lucky, because automation needs to happen on some basis, right? For example, a patent database. And somebody needs to be responsible and accountable in the end: the professional. The new thing, the LLM, supercharges the professional.
It's here to stay.

However, LLM providers need to monetize their product to gain investor trust as they try to become the next monopolist providing a form of AI that can't be cheaply self-hosted on prem. So LLM providers try to provide the LLM wrapper, like Anthropic did with Claude Code in a fantastic way. Now they're going after legal:

For the record, I'm a huge Anthropic fan! So take the following as an observation, not a grudge.

I've been trying for months now to get officially listed with my patent connector at Claude (ChatGPT took some time, but I'm listed; the Microsoft listing worked too, but it was a journey).

Three months ago Chris Parsonson wrote: "In partnership with Anthropic, Solve is one of 12 legal plugins integrated within Claude, and is the only one specialized for patent workflows." (https://github.com/anthropics/claude-for-legal)

Meanwhile no MCP pull request gets merged - not mine, not anyone's.

Speaking of Claude legal, there's not much happening in the main repo. But it invited (coming from Anthropic) 1.8k people to fork and continue working on it. That would make a good source of inspiration and a market-response test without oversharing, right?

But they're not the only players in the wrapping game, which is funny, because they (and their investors) are currently paying for the gift. There are so many established and startup software providers trying to integrate LLMs into their workflow and trying to sell their product as the powerhouse, when in fact it's actually just wrappers with a different flavor around the LLM powerhouse. And as coding became cheap, this is just escalating. They're investing and hoping they can stay in the game, but they're just plumbers, because the actual work is done by the professionals. It's not the tool that does the job, even with LLMs. It's the professional. They know best.

And here too the wrapperization happens, where the actual service business becomes a semi-automated LLM-driven service offering, sometimes using some LLM-coded internal tooling.

Speaking of LLM-coded internal tooling, let's wrap it up (.....):

• The LLM is the actual invention. That is where the value sits.
• The wrapper is necessary, but cheap to build, and it can be LLM-coded.
• Building the wrapper well still takes the pro's domain knowledge.
• The pro carries the knowledge and the responsibility.
• Smart-enough LLMs can be deployed on prem.

I argue that future corporate process digitization should and will involve 1-x skilled, LLM-supported software developers directly in the department.

Let's do this and let's get hyper-productive.

That was my brain dump. I'll now paste it into an LLM to polish while making a coffee. It's 5:51.