Hong Kong enterprises in the AI era:
structural shift and workforce rebuilding
and what does a transformation record built on first-hand practice actually show?
Not a problem of tools.
A problem of structure.
For the past two years the question facing Hong Kong firms has no longer been “should we use AI”. Four structural changes are arriving at once: demand is shifting, the cost base is shifting, technical capability is shifting, and the policy framework is shifting.
Any one of these on its own is manageable. Arriving together, they test management’s judgement.
This session draws on the government’s latest manpower projection and local SME surveys to show the real scale of these changes, and sets out what firms can do at the operating level.
AIX Society spent a year studying the local market, then implemented a complete AI-agent operating system inside our own company. That experience became “The Year I Handed My Company to AI”.
Thirty minutes, four parts
Market, operations, technology, policy: four structural changes
Government manpower projections and SME survey data, to show the position local firms are actually in.
Evidence from one company
The operating system recorded in “The Year I Handed My Company to AI”, and what it is built from.
The talent gap and the training path
Why training is the core of this transition, and how in-house and public training divide the work.
AIX Society: role and forms of collaboration
The Society’s positioning, its authorisation background, and the formats we can arrange for chamber members.
Michael C.S. So Founder and President, AIX Society
Twenty years in business management, corporate governance and technology innovation. Former Executive Director of a Hong Kong main-board listed company, INED of two GEM-listed firms, and COO of a US OTC-listed company. Founded Originals Group in 2012; its e-commerce and e-learning businesses still run, two of them acquired by listed companies.
Since 2022 he has worked on generative AI and agentic applications, with his own company as the first site of implementation — that year’s record is the source of this book.
AIX Society — Institute of Applied AI Research
Founder and President · Principal Consultant
Certified Instructor, “AI+ Business Training”, DingTalk AI Academy (Alibaba Group) · Certified AI Expert (Blockchain Council)
“Internet Technology 3.0” column, HK01
“A Little More AI” column, Sin Chew Daily
Presenter, Metro Radio (2026)
BBA (Hons), Lingnan University · Graduate Certificate in Technology Management, QUT · Executive Certificate in Private Equity and IPO, Peking University HSBC Business School
Four structural changes
all four turning at once, and together they define the real trading environment.
Adoption is high; depth is still shallow
Hong Kong SMEs have broad exposure to AI, but usage clusters at the easiest entry points. Most firms have finished “trying”; few have started “integrating”.
AI tools within the next 12 months
pay for AI tools
is broader in 2025 than in 2024
or workflow improvement
Marketing and content production 56%, customer service 42%, sales support 35%, data analysis 30% — all single-point, assistive uses with immediately visible results.
The tools are in the building; the process has not changed. While AI stays at the level of assisting individual tasks, the operating structure does not move, and neither does the cost base.
Cost pressure has not gone; it has changed shape
For local SMEs, staff cost has always been the largest and least compressible item. What AI changes is not whether you hire, but the amount of work each role carries.
The government projects that AI adoption can lift job productivity by 20–30%; the same report notes that within three years, 20–30% of jobs will see their daily duties change substantially.
In other words, the question is not “how many people do we cut” but “what will this role do next year”. Answer that badly and productivity gains turn into plain headcount reduction — usually followed by a loss of quality and experience.
Government estimate of the productivity gain AI adoption can bring to jobs
Roles whose daily duties will change substantially within three years
Source: Labour and Welfare Bureau, Manpower Projection Mid-term Update, 22 September 2026.
The technical barrier fell; the judgement barrier rose
Two years ago, a firm talking about AI first had to answer “can we even do this”. Today models and agents can complete a large amount of daily work, and the technology itself is no longer the obstacle.
Surveys bear this out: 47% of SMEs that have used AI found 2025 easier than 2024 — maturing technology, more tools to choose from, and better staff acceptance.
The real barrier has moved to three questions: which process is worth handing over, how do we control it once it is handed over, and who answers when it goes wrong. No tool can answer these three.
Can it be done
Model capability, tool choice, ease of use — all have fallen sharply.
Should it be handed over · how to control it · who is accountable
This is governance and process design. It needs management judgement and cannot be outsourced to a tool.
AI risk governance sits with the Chief Secretary for Administration; the Digital Policy Office will appoint an AI Commissioner to drive sector guidance, certification and governance frameworks, and issue AI-agent safety guidelines.
The direction is set; firms need to find their own lane
The Hong Kong Five-Year Plan (2026–2030) and the 2026 Policy Address have made “AI+” an industry direction, pushing on four fronts at once: compute, funding, talent and governance. For a firm, the point is not how grand the policy is, but which lane it belongs to.
training courses and activities (to Q1 2028)
against Hong Kong’s current total compute
to help SMEs adopt AI
Funding (Digital Transformation Support Pilot Programme, BUD) · technical support (HKPC’s “Digital DIY” platform) · training (“AI for All” and delivery-partner courses) · governance (sector guidelines and certification).
Policy is driven by public bodies; what the Society does is implementation at the civic and enterprise level — turning policy text into processes and training a firm can actually run.
Four changes overlap; four questions firms keep asking
Where do we start?
Too many tools, too many paths. Firms usually buy the tool first and think about the use later, then end up back where they started.
How do we measure the return?
38% of SMEs that have not adopted AI say the return on investment is unclear; 29% say they have not found a suitable use case.
What about staff attitudes?
Resistance usually comes from not knowing whether AI will replace them. The more abruptly the instruction is issued, the stronger the resistance.
How do we define compliance and liability?
Whether data can leave the local environment, and who bears the cost of an AI error, are questions not yet written into internal policy.
Evidence from one company
What worked and what went wrong that year is all in “The Year I Handed My Company to AI”.
“The Year I Handed My Company to AI”
A year ago I opened a new office on a different governance model: AI sits in the management layer, with people underneath. Management no longer chases each person’s progress; it sets outcomes.
The book is the full record of that year and a transformation manual you can follow — from the data foundation to the seven-step loop to the division of work between people and machines. It describes an operating system, not a list of tools.
Belief · System · Execution · Region
the AI-native operating system
distilled from the whole book
Hong Kong · Shenzhen · Malaysia
Published October 2026 | Traditional Chinese print (Hong Kong) · e-book · Simplified Chinese edition (mainland) · Cantonese audiobook | By Michael C.S. So, with AI assistant Maix
Why a pile of tools changes nothing: the problem is governance, not tools.
A first-hand account, not an observer’s view. Every figure can be checked.
Tested in Hong Kong, Shenzhen and Malaysia — three compliance regimes.
The seven-step loop: an AI-native operating system
Data foundation
Let AI understand your company
Sense
Spot opportunity before competitors do
Understand
Turn data into insight
Decide
Let AI sit in on the judgement
Create
From insight to courses and products
Execute
Delivery and experience
Learn
Feedback loop, with agents running through every layer
One full turn of the seven steps
and the company moves one square forward
The point of the loop is not the steps themselves but that each one leaves an auditable record: decisions with reasons, execution with traces, learning with feedback.
Four parts, from belief to execution
Belief
- Why “using AI” will not save your company
- The AI-native company in one diagram
- One person, one legion: my digital staff
The seven-step loop: the operating system
- Data foundation · sense · understand
- Decide · create · execute
- Learn: feedback loop and the cross-cutting layer
In practice: tools, roles and the production line
- The tool stack and the division of labour with digital staff
- One working day and the content line
- AI HRM and an automated CRM
Region
- Hong Kong: compliance, cost and the entry point
- Mainland China: the DingTalk and WeCom ecosystem
- Malaysia: going abroad with cross-border teams
Parts III and IV are the most practical — they answer “what can I do when I get to the office tomorrow”.
Ten actions for the owner
The whole book compresses into ten. Each is a slogan plus an action for today — and together they are the order my own company actually went in.
These ten are not parallel suggestions; they are an assembly sequence. The book sets a starting point by company size; in the talk we pick the three that fit the room and work through them.
Do number four: have AI hand you a briefing every morning.
It moves the start of your day from yesterday to today. After one day you will know why the other nine are worth doing.
1 Work out where your hours actually go
2 Name the three things you should not be doing
3 Build the foundation before you buy the tools
4 Have AI hand you a briefing every morning
5 Turn “gut feel” into “laid out in front of you”
6 Keep only two approval red lines
7 Turn deliverables into checklists
8 Keep every mistake inside the system
9 Let AI be the coach, not the supervisor
10 Free yourself up
The whole book in three sentences: boundaries before tools. Foundation before AI. One thing before the whole company.
The talent gap and the training path
not a supplement after it is done.
Surplus and shortage, at the same time
This is the set of numbers that needs the most explanation. The government’s latest projection shows Hong Kong facing neither a simple “shortage of hands” nor a simple “disappearance of jobs”, but both at once.
What the numbers mean: the market is not short of “people”; it is short of “people who can take on the new processes”. A firm that only cuts headcount without rebuilding roles will hold both the shortage and the surplus at once — unable to hire the right people, and unable to keep the ones it has.
Four years of preparation, three years of restructuring
When a firm hires, it is buying knowledge built over four years of university. But the part hit hardest by this round of restructuring is precisely knowledge that can be taught through textbooks, courseware and written procedure — because that is exactly what AI absorbs and reproduces best.
Rule-based · repetitive · teachable in writing
Data entry · spreadsheet tidying · routine email replies · first drafts of documents · standardised customer service · entry-level translation.
These used to be where newcomers built experience.
Requires practice · judgement · accountability · presence
Process design · quality control · risk judgement · client relationships · cross-team coordination · work that needs someone on site.
These capabilities come from repeated practice and coaching, not from courseware.
The practical implication: the door into entry-level jobs is narrowing, while the capability demanded of the middle layer is rising. A firm that keeps hiring and training the old way pays twice — wrong hires, and people it cannot keep.
Training is not a supporting item; it is the precondition
45% of SMEs have not adopted AI, and of those 57% cite a lack of knowledge or skills — the largest single obstacle, ahead of unclear return (38%) and no suitable use case (29%). Training is not a marginal topic; it addresses the largest obstacle directly.
When 20–30% of roles see their daily duties change substantially, what firms need is not a one-off course but capability building that runs alongside job redesign: define the output of the new role first, work backwards to the capabilities required, and only then design the course.
Hong Kong has only about 4,000 practitioners with both domain knowledge and AI skill — 0.1% of the employed population — while demand is expected to multiply. This cannot be solved by recruiting; it can only be grown inside firms.
Quality control, data boundaries and accountability all require management and front line to argue in the same vocabulary. Without it the rules cannot be written; and written, they cannot be enforced.
Why we do this
AIX Society operates as a social enterprise. Policy has made AI literacy a public goal and is pushing it in three tiers — foundational literacy, professional conversion and applied uplift — and it is the applied tier, where SMEs and working professionals sit, that most needs work at the civic level.
We spent a year studying the local market and implemented a full AI-agent operating system inside our own company. That process made one thing clear: what firms lack has never been information, but judgement — which process is worth changing, how far, and who signs off.
So the Society concentrates on four things: training, research, case building and practice. They support one another; they are not four separate product lines.
The Society is not run for profit. Surpluses are not distributed to shareholders but returned in full to research, case building and public-interest training. For commercial clients our fees and delivery standards are the same as any training provider.
DingTalk AI Academy, Alibaba Group — authorised strategic partner for Hong Kong and Malaysia; QwenWork authorised training partner.
Hong Kong · Mainland China · Singapore · Malaysia. Teaching languages: Cantonese · Putonghua · English.
Four tiers, streamed by seniority
AI literacy foundation
half day
Functional AI specialist
on request
AI transformation leadership
course series
Executive AI briefing
1–2 hours
Common method: value first, reusable prompt templates, hands-on build module by module, a closing rehearsal — always on synthetic sample data.
Bring society in, not only serve those who can pay
If training only serves firms that can afford it, the talent gap widens. So the Society also keeps public learning channels open — not a main source of income, but the thing that decides whether “AI for All” is real.
Student HK$180/year · Professional HK$880/year
Course access, a weekly AI briefing, selected free courses, career support and a peer network.
Public talks and industry exchange
Including small-table executive sessions with limited seats, focused on real transformation decisions.
Interactive learning stations
Open free of charge, for colleagues meeting AI for the first time.
Two books and public writing
Books, columns and the policy tracker are all openly readable — continuous material beyond the classroom.
The “AI for All” tracker
Government funding terms, courses and events, updated every working day, free for SMEs.
Speaker profile
Michael C.S. So
Founder and President · Principal Consultant, AIX Society
DingTalk AI Academy, Alibaba Group — authorised strategic partner for Hong Kong and Malaysia
QwenWork authorised training partner
What we can arrange for your chamber
Keynote (30 minutes + 15 minutes Q&A)
The format of this proposal, with the argument and the case mix adjusted to your sector.
Policy and funding briefing
On “AI for All” and the Digital Transformation Support Pilot Programme, helping members work out which tier they are in and what they can apply for.
Member-only AI literacy workshop
A condensed version of the L1 course, half a day, suitable as a membership benefit.
Industry roundtable and case discussion
Small-table format, using members’ own processes as the material, to decide which step to hand over first.
Operating process review for member firms
One review session for members ready to start, to define the first pilot.
- A year studying the local market, implemented in full inside our own company
- Familiar with government manpower projections, SME surveys and policy text
- Proven in three markets; cases and figures can be checked
- Authorised partner status; first-hand information on tools and ecosystem
- Teaches in Cantonese, Putonghua and English
Thirty minutes, how it runs
Opening: one set of numbers
Hong Kong is short of 130,000 people and at the same time carries 30,000 surplus clerical staff. Start from that contradiction.
Four structural changes
Demand, operating cost, technical capability, policy framework — each with data.
Evidence from one company
The seven-step loop and the ten actions, each stage illustrated from our own operating records.
The talent gap and the training path
Why training comes first; how in-house and public training divide the work.
Closing: three suggestions for members
The whole talk, condensed into three judgements members can use in a meeting next week.
Q&A
Questions from the floor; time permitting, a preliminary diagnosis for one or two member firms.
Technical and administrative needs
Projection and sound
16:9 projection or large screen, HDMI or USB-C input; a wireless microphone if available.
Time
30-minute talk + 15-minute Q&A. Can compress to 20 minutes, though we would keep the Q&A.
Audience
If you can share the sector mix and size of your members in advance, we will adjust the argument and the case mix.
Language
Cantonese by default; Putonghua or English on the day if mainland or Southeast Asian members are present.
Supporting material
A one-page self-assessment and a post-event reading list, for members to take back to the office.
Follow-up
After the event, an optional process review for interested members, as a basis for deciding whether to go further.
Help your members grasp the judgement, not just the information
AIX Society — Institute of Applied AI Research · Operates as a social enterprise · Surpluses are not distributed to shareholders
Every figure cited in this proposal carries its source and publication date.