China’s AI Revolution: Practical Lessons for Legal Tech
I was listening to the Big Technology Podcast with guest Grace Shao when something clicked about why we still struggle with legal tech adoption. Her insights on China’s AI ecosystem challenged several long held assumptions we have in legal.
The legal sector’s relationship with tech is still awkward. We buy systems that gather dust while lawyers stick with email. We overengineer for perfection and ignore what users actually need. Worst of all, we keep building isolated tools when it’s obvious that integration is the key to getting anything adopted.
What stood out about the Chinese approach was how relentlessly practical it is. Here’s what I took from it and how we might apply the same mindset.
Challenging the Resource Assumption
DeepSeek, a Chinese AI startup, delivered state‑of‑the‑art results with far less compute than its competitors. It didn’t just squeeze more from less; it thought about if bigger is always better.
In legal tech, we have our own baked‑in constraints: partner capital, billable hours and risk‑averse cultures, but how much of that is just stuff we’ve learned to live with rather than challenge?
So rather than trying to fix everything at once, we should pick the highest‑friction problem and solve that. That does not mean abandoning longer‑term platforms entirely, they have their place for tackling underlying issues but it does mean making smaller wins your default. Just focused, surgical tools that slot into existing workflows and remove the biggest irritant today.
Good Enough Is Good Enough
Chinese firms push “good enough” AI into real products quickly while many Western companies keep polishing models in the lab. They learn from real‑world use and because of that they get feedback fast.
In law, perfection is the default for good reason, but that instinct carries over into tech decisions and can kill momentum. Tools get rejected because they are not 100 per cent accurate, even when 80 per cent accuracy would already save hours.
Reframing is key. That AI tool does not replace your review; it prioritises it. It flags where to start, which saves time rather than making final calls. When you frame it like that, people are much more open, especially if they get to keep control.
Integration Over Innovation
Tencent integrated DeepSeek’s models directly into WeChat because that is where users already are and it’s a lesson legal tech teams keep learning the hard way: users adopt what is frictionless, not what is clever.
In legal I’ve seen technically impressive tools lik clause analysers, contract visualisers, ML tagging systems, all completely ignored. Meanwhile, a simple templating toolbar in Word has people genuinely excited. Not because it’s groundbreaking, but because it saves time on something annoying.
All that said though, we should think bigger. Small wins are vital, but we also need to join them up. Tools that remove five clicks are helpful. Tools that reshape how knowledge flows across a matter, a practice group or an entire firm, that’s where the real long term impact comes from.
Integration isn’t just a delivery mechanism. It’s a strategic opportunity. When systems talk to each other, when data moves without friction, when AI shows up where it makes sense and we stop thinking in tools and start thinking in outcomes.
Building Connected Ecosystems
The podcast described the “Six Dragons” of Hangzhou, a tight AI cluster built on shared resources, academic links and overlapping expertise. It shows that ecosystems thrive not just on talent, but on structure and design. You need the right conditions for collaboration.
Legal tech remains fragmented. Law firms, in‑house teams, courts, vendors and researchers often operate in silos. Even within firms, knowledge management might not sync with innovation, and IT might not speak legal ops. Everyone ends up solving the same problems independently, wasting time and money.
Engineering disciplines work differently. Open‑source communities share, remix and improve. Models publish publicly, benchmarks get shared. You build on what came before and move faster because of it.
Legal tech is beginning to shift. The NosLegal taxonomy is an open framework for structuring legal concepts that works across firms, platforms and tools. It’s neutral and practical, so adoption follows. It does not compete; it enables.
The Manchester Law and Technology Initiative does similar work, by anchoring projects in a neutral academic environment, it lets vendors, firms and universities collaborate without triggering competitive alarms. That is where real progress happens: in shared spaces that are not trying to own the outcome.
No single firm will ever serve every client. There is room and need to build on each other’s work.
The Risk of Our Risk Aversion
Caution is deeply embedded in legal culture and for good reason, though when it comes to innovation that culture often turns into inertia. We worry so much about what could go wrong that we ignore what happens when nothing changes.
We focus on short‑term risks such as data breaches, compliance issues or botched rollouts, but the long‑term risks like losing talent, frustrating clients, falling behind are just as real. We just do not measure them as clearly.
In my recent article, I wrote about the rise of “vibe coding,” where legal professionals build tools under the radar because formal routes do not support experimentation. These people are problem‑solvers showing that innovation will happen, with or without official support.
Firms should embrace that energy by creating spaces for safe, supported experimentation. Regulatory sandboxes are one option, internal working groups with permission to test low‑risk tools are another. Whatever the form, the goal is to shift from “we cannot afford to get this wrong” to “we need a way to get this right.”
Learning to Build With Constraints
Chinese companies have developed sophisticated AI tools despite significant headwinds such as chip shortages and export restrictions. Some of those barriers may have been sidestepped through grey‑market channels. There is an entire underground industry helping sanctioned countries acquire restricted tech. Even so, resource constraints are very real.
Those limits have influenced outcomes. Instead of chasing scale at all costs, many Chinese teams doubled down on efficiency with leaner architectures, better memory handling, smarter data use. The result is state‑of‑the‑art performance with a much smaller compute footprint.
Legal organisations face their own constraints: partner resistance, tight budgets, legacy platforms and limited in‑house engineering talent. These will not vanish. So the solution is not to wish them away but to build smartly within them.
The good news is that we have more tools than ever to do this. Open source models, cloud‑based inference APIs and AI infrastructure‑as‑a‑service platforms now handle the boring, hard bits like vector search, chunking, parsing, hosting and scaling. That frees up time and budget to focus on the parts that really matter: domain logic, user experience and solving legal‑specific pain points.
When limits are known upfront, focus improves, adoption improves and results improve. Constraints are not just acceptable; they are useful as long as we design for them.
DeepSeek’s breakthrough did not come from more GPUs. It came from rethinking assumptions about what was required. That same mindset shift is overdue in legal tech.
If you are responsible for legal technology in your organisation, start here:
- Identify where technology meets existing workflows and build there
- Use “good enough” tools that solve real problems today and improve over time
- Prioritise integration over reinvention by meeting users in places they already are
- Balance risk properly by accounting for the cost of inaction as well as action
- Connect by sharing ideas, frameworks and taxonomies such as NosLegal to avoid solving the same problems in parallel
Technology is already reshaping legal work. The real challenge now is staying close enough to the change to influence it, not just reacting after it’s already moved on.