On December 19, 2025, in Central, Hong Kong, I hosted a closed-door session on the theme of "GC × AI Adoption." It was the latest stop — after Silicon Valley, Beijing, and Shanghai — in a three-month run of conversations building out the Silicon Valley Legal Tech Frontier Community in another major city.
The format was small-group, in-depth discussion, with venue support generously provided by Herbert Smith Freehills Kramer.
That day, online and offline together, we gathered nearly 30 senior in-house counsel and legal-tech founders from Hong Kong and Shenzhen, with backgrounds spanning big tech, multinational financial institutions, healthcare organizations, state-owned enterprises, and more.
The discussion was clear-eyed and pluralistic — less a showcase than a "reconciliation," putting AI back into a real business context. What kept coming up was a more grounded class of problem:
AI's limits, increasingly, come less from the capability of the model and more from the organization and the system itself.
Healthcare and Finance: Real Demand in Highly-Regulated Verticals
The perspectives from healthcare and financial-institution counsel in the room revealed a form of demand that has long been overlooked, yet is very real.
Within the healthcare system, many legal staff don't come from a traditional legal background. Quite a few started out as physicians or in public-health roles, and only later took on legal and compliance responsibilities. That means that, faced with fast-moving regulation and compliance requirements, simply keeping information current and digesting the rules is itself a high-cost undertaking.
At the same time, regulatory pressure keeps rising. Whether it's contract management, internal quality control around clinical workflows and medical-records standards, or risks tied to doctor-patient disputes, the checks and demands coming from the health authorities keep escalating. In practice, this leaves hospitals with clear gaps in legal support and tooling.
From a demand standpoint, hospital legal teams aren't chasing complex or "cutting-edge" legal-tech products. Contract management, risk alerts, and compliance information that stays current as regulations and adjudication trends shift — those alone would already relieve a meaningful share of the workload.
The real difficulty is that this kind of demand usually sits on top of highly sensitive, non-standard scenarios:
- On one hand, medical-data compliance requirements are extremely high — patient privacy, internal documents, and medical-quality assessments — which makes it hard for external products to be integrated directly into native systems;
- On the other hand, even in relatively generic contract scenarios, existing standardized products struggle to fit the way a hospital actually operates, in terms of template types and business-scenario coverage.
The result: the demand genuinely exists, yet few products can actually get up and running on a foundation of compliance, security, and system fit.
A similar bind showed up in the remarks from financial-institution counsel — denser regulatory requirements, more frequent regulatory change, and a heightened sensitivity to data security and the boundaries of liability all make the conditions for deploying legal tech in finance equally demanding.
In that sense, healthcare and finance are both archetypal scenarios of highly-regulated, vertical legal demand. Here, using AI isn't simply rejected — but it must satisfy clear boundaries, well-defined responsibility, and an explainable process. "Usable" often matters more than "advanced."
Big Tech: AI as Infrastructure, Not a Tool
Compared with hospitals and financial institutions, the counsel from big tech companies described a different posture.
In these organizations, AI isn't an isolated tool choice so much as a kind of infrastructure capability. The focus of discussion is usually not "can we use it," but "how do we use it systematically."
Some shared threads:
- Companies tend to build their own tools, and to first wring maximum value out of general-purpose models and in-house tooling, rather than prematurely bringing in highly customized vertical products
- Embedding AI deeply into existing internal systems, rather than bolting it on from the outside
- Drawing a very clear line between which tasks can be handed to AI and which judgments must remain with humans
In this context, AI's value lies not in what content it generates, but in whether it genuinely lowers the internal cost of communication and friction within the organization. That's also why big-tech legal teams tend to be more positive about AI: for them, using AI is a long-term engineering optimization, not a short-term technological leap.
Why Is "Legal AI" Fundamentally a Systems-Engineering Problem?
In the discussion, several founders with years of legal-tech experience shared a common view: the hard part of legal AI usually isn't the algorithm — it's the systems engineering itself.
First is the data problem. If you want AI to produce reasonably reliable judgments, it has to be built on accurate, traceable data. Once the data source itself is incomplete or unstable, AI easily produces bias, or even hallucination.
Second is the return-on-investment problem. Many legal-AI projects aren't infeasible to build; they're just hard to make "good enough" on a limited budget. On one side, there's a natural tension between legal practitioners' near-100% accuracy expectations and what engineering can deliver. On the other, when procuring, domestic government bodies and enterprises are — relative to standardized products — more willing to pay for highly customized services.
This high cost also shows up in the complexity of system integration. Take contract management: the hard part often isn't the contract itself, but connecting to the procurement, finance, warehousing, and OA systems the enterprise already runs. Once you enter the system-integration phase, the engineering effort scales up fast.
Finally, there's organizational coordination. Even when the budget is in place, deploying legal AI still requires repeated back-and-forth among legal, IT, information security, and management. It tends to be a quiet, slow, but unskippable process.
That's exactly why, in many scenarios, legal AI isn't a case of "there's no product" — it's that real-world conditions aren't yet sufficient to let it run smoothly.
A Few Observations on Hong Kong's Legal-Tech Market
Being in Hong Kong this time also gave me a more concrete feel for the legal-tech ecosystem here.
By sheer size, Hong Kong's legal-tech market isn't large. Whether measured by number of startups or overall commercial scale, it's noticeably smaller than the mainland's first-tier cities.
But it has some very distinctive features.
For one, support from the Hong Kong government and universities is consistent — events like the annual "Hong Kong Legal Week" provide relatively stable soil for legal tech.
For another, many Hong Kong legal-tech products are, in design philosophy, closer to the US/European system, emphasizing document structure and evidentiary relationships in a common-law context. That makes them easier to understand and get attention internationally. International legal-tech products like Harvey are already in real use through Hong Kong's international law-firm networks.
At the same time, Hong Kong has some highly localized needs that outsiders aren't familiar with — for example, meeting-transcription and translation tools for mixed Cantonese, Mandarin, and English settings. Such products may not apply to other markets, but locally they carry very clear value.
Another emerging trend: lawyers from large firms — especially those with a capital-markets background — are increasingly moving into legal-tech entrepreneurship. They tend not to chase coverage of every workflow, but instead go deep on a single point — diligence in M&A, prospectus disclosure verification in IPOs — forming a "narrow but deep" product path. This trend is consistent with what I've observed across several large firms in New York and Silicon Valley.
Year-End Reflection and a Look Ahead
From the three offline events in Silicon Valley, to Beijing, Shanghai, and Hong Kong, one thing has become increasingly clear to me: there is no single, unified path for legal tech to land.
Different markets, different organizations, and different legal systems each shape entirely different choices. And the genuinely valuable discussions tend to happen inside specific scenarios, not in macro pronouncements.
Going forward, alongside continued offline events, we'll gradually try some scenario-based online interviews and livestreams, taking the questions that are suitable for public discussion into a larger context.
That's also why I've opened an interview invitation for global legal-tech enthusiasts next year, and — in a personal, vibe-coding way — built myself a Calendar Assistant to host these ongoing conversations.
If you'd like to go deeper on law, AI, career paths, or related topics, you're welcome to visit helenlab.com and use the booking entry on the page to schedule a one-click online conversation with me. (If overseas deployment prevents access, you can also reach me at helen@helenlab.com.)
The second half of AI is still unfolding. This is just a snapshot in time.
Happy New Year — see you next year!