Legal departments are rapidly embracing generative AI but adoption of the technology has only just begun. Two prominent leaders driving the transformation of China’s legal and compliance sectors, veteran counsel Cao Haiyang, founder of a legal AI executive network, and Zhou Xin, CEO of AI developer Aivigate Technology, give their views
IN THE PAST 20 YEARS, the role of in-house counsel has undergone two paradigm shifts: from “reviewing clauses and revising contracts” to “building compliance systems and carrying regulatory responsibility”. Today, corporate lawyers stand at the threshold of a third paradigm shift – the dawn of AI.
This is apparent from a striking statistic emerging from this year’s 7th Annual General Counsel Report by international business consultancy FTI Consulting and legal tech software group Relativity. The report reveals that 87% of legal departments are already using generative AI, up from just 44% last year, almost doubling in a single year and representing the fastest rate of technology adoption ever recorded in the legal sector.
However, what truly merits attention is not the 87% figure itself, but the gap between the “changes that have occurred” and the “changes that have not occurred” behind it.
Among noticeable changes that have occurred, individual legal professionals have begun using chatbots – with senior in-house counsel among the most active adopters – and the time required for a first-pass review of a contract has reduced from two hours to 20 minutes, while role boundaries have quietly begun to loosen.
But the fundamental rewiring of knowledge, processes, permissions and accountability, along with the organisational restructuring needed to break down old divisions of labour and traditional team key performance indicators (KPIs) has not occurred.
This leads to one conclusion: using AI does not equate to organisational strengthening. Individual efficiency gains have materialised, but organisational-level upgrading has yet to emerge.
This is precisely what makes the event not a gradual upgrading of tools but a structural leap forward. Today, the challenge facing heads of legal is no longer whether to use AI – but whether they can leverage this wave of AI to reshape the entire legal organisation.
AI-assisted is not AI-native
The most common misconception in the industry is to equate “AI-assisted” or “AI-enabled” with “AI-native”: the organisational logic is entirely different.
In the context of AI-assisted or AI-enabled approaches, AI is positioned as an augmenting tool used to enhance existing tasks and improve individual efficiency. It does not alter the organisation’s operating model; once AI is removed the efficiency gains previously achieved through it are immediately lost. Gains come and go with the tool – a fundamental limitation of the “assisted” model.
On the other hand, AI-native is the complete opposite: AI is no longer an add-on but a new operating system. Processes, role definitions and performance evaluation systems are all redesigned around AI. There is only one real test for determining whether an organisation has entered this stage: if AI is removed the organisation’s operations would become materially impaired, not merely less efficient.
Whether legal departments have entered the AI-native stage should not be measured by the number of models purchased or the scale of subscriptions, but rather by whether they are willing and able to redefine three things: which tasks are delegated to AI; which are assigned to humans; and which are embedded into the system.
By this measure, the legal departments of most Chinese enterprises remain AI-assisted. However, the teams that are first to reach the AI-native stage will gain a new capability that can be described as “organisational compounding”. This will not be lost when a senior in-house counsel leaves. Nor will it reset to zero due to personnel changes.
Adopting is a half-measure
In the past year or two, many companies have been pushing to implement AI internally. However, there is often a significant gap between adopting AI and actually “becoming stronger”.
There are four patterns of failure most commonly experienced by legal departments:
-
- AI is hung at the edge of the organisation, far removed from core business processes. It appears only in year-end power point slides and is otherwise barely touched upon for the rest of the year;
- More critically, the department head fails to adopt the new way of working but demands that employees use AI, directly undermining the legitimacy of the change;
- The phenomenon of “idle tools” arises: a batch of products is purchased, but on review six months later usage rates are negligible; and
- Treating AI as cheap labour with the sole intent of slashing headcount rather than restructuring work methods is a quintessential example of legacy thinking.
When viewed collectively, these four patterns lead to a more pointed conclusion: from the perspective of the legal function, the most dangerous scenario is not the absence of AI, but rather a situation where everyone is using AI while nothing changes regarding knowledge access, approval chains, template rules or accountability mechanisms.
The result is not an upgrade in the capabilities of the legal function, but rather an increase in “shadow AI”, which refers to employees using publicly available models to handle work without the company’s knowledge, oversight or auditing.
For enterprises directly under the central government, listed companies and foreign-invested enterprises, the risks posed by this situation – from data leakage and confidentiality breaches to lack of audit trails and accountability that cannot be traced – far outweigh those of not using AI.
Two tangible walls also stand between personal efficiency gains and organisational upgrades.
The first wall is that performance evaluations are still based on volume and hours worked. Under this type of KPI, working smarter actually puts employees at a disadvantage. As case counts drop, hours shrink and templates are absorbed into the system, it ironically becomes harder to justify output at year-end.
The second wall is the drag on overall efficiency caused by processes, approvals and knowledge silos. AI can triple the efficiency of an individual legal counsel, but if business teams upstream provide materials in the same way, if downstream approval steps remain unchanged, and if every cross-functional handoff point is clogged, the end result will only be localised acceleration while the system as a whole remains stagnant.
Therefore, for general counsel driving legal AI transformation, what truly warrants scrutiny is not how many tools the team is using, but whether the current KPI has begun to reflect “work quality assessments” and “workflow capture”. If not, the so-called transformation remains stuck at the stage of individual efficiency gains.
Defining AI-native
A truly AI-native legal department can be understood as a system composed of a front end, a centre, and a base layer. The front end comprises job tasks themselves, including contract intake, review and negotiation, approval and signing, contract performance, dispute handling, and post-matter review and knowledge capture. The focus is not on the stages themselves, but on the fact that each stage has been deeply reworked by AI.
The centre comprises knowledge, processes, permissions and accountability – including rule packs, templates, precedents, decision logic, approval requirements and audit records. These elements are no longer scattered within the mind of a certain in-house counsel nor in folders on a shared drive; instead, they operate in concert within the system.
The base layer represents integration of intelligent agents within the system: AI handles standardised processing, while humans are responsible for critical judgements. The results are then fed back into the system, forming organisational assets.
The core of this entire structure is not that “everyone can use a chatbot”, but rather the systematisation of the core of the legal department’s work.
First, knowledge must be captured as reusable organisational capability, rather than continuing to rely on the personal expertise of individual senior in-house counsel.
Second, processes must be broken down and redesigned: tasks are first broken down, then decisions are made regarding who will execute them, which actions are delegated to AI and which must be led by humans, no longer assuming that “everything must be reviewed manually”.
Additionally, organisation-wide adoption must be supported by mechanisms for permission, auditing, audit trails, isolation and review. These serve not only as essential safeguards for the enterprise but also mark the compliance dividing line where the legal department transitions from “pilot tool implementation” to “organisational AI adoption”.
Following this line of thinking, the boundary between human and AI roles will also become clearer. Core legal judgments in major transactions, complex dispute resolution and crisis response – as well as strategic-level risk assessments and responsibility determinations – must remain human-led.
In short, AI cannot replace any stage that requires a “sign-off and assumption of responsibility” because it can neither sign documents nor bear legal liability. This boundary is not merely a technical issue, but also a matter of allocation of responsibilities.
In contrast, routine review, drafting, extraction, summarisation, comparison and archiving – as well as rule application, knowledge retrieval, aggregation of similar precedents, and pre-processing, alerting and post-matter knowledge capture within workflows – should be prioritised for systematic handling by AI.
The reason is that in high-frequency, standardisable and low-risk tasks, AI is typically more stable than humans because it does not experience fatigue, does not miss checks, and can be effectively governed.
Only after these two boundaries are clearly defined does the AI transformation roadmap for the person in charge of in-house counsel department truly have a practical starting point.
The first step is not procurement but an inventory: exactly how much of the current workload falls into the “AI should take over” category is still being performed manually. This is often the most direct point for unlocking organisational efficiency.
A new agenda
Ultimately, the bottleneck in AI-native organisations often lies not with the team but with the leader at the helm. A great leap at the organisational level ultimately requires legal leaders to first complete a shift in his or her own role.
The first shift is from task processing to value creation. In the past, the focus of the work of in-house counsel was the number of contracts finalised, the number of disputes resolved and the number of business requests addressed. At this new stage, however, the priority is supporting transactions and strategic implementation, reducing costs and shortening cycle times through legal design and transforming in-house counsel capabilities into operational competitiveness.
Once AI takes over standardised tasks, the legal department should not fill the newly available time with low-value tasks but proactively shift its focus toward business, transactions and operational judgement.
The second shift is from reactive response to proactive prevention and control. Risks must be integrated into business processes to form a closed-loop system of ex ante prompts, interim monitoring and ex-post feedback.
AI enables “ex ante prompting” to scale for the first time. Real-time reminders based on historical contracts, regulatory developments and organisational rules can achieve a level of granularity that past compliance systems could not match.
The third shift is from homogeneous teams to composite teams. The workforce dedicated to standardised tasks will correspondingly decrease, while people who combine law, business, industry knowledge and digital capability will become increasingly important.
This adjustment to team structure involves role reshaping, performance system adjustments, training investments and even decisions to let go and retain personnel, inevitably involving some pain. However, without such adjustments, a legal department will struggle to truly cross this threshold.
Looking ahead, what will truly set legal departments apart is not which ones adopt AI first, but which ones complete organisational restructuring first, transforming the legal department into an AI-native organisation capable of continuously training its systems, capturing workflows and amplifying judgement.
In the AI-native era, a legal department will either become a business accelerator or be relegated to a “cost centre that still operates in the old way” in the eyes of the business.
The divide between these two roles is already beginning to emerge.
