AI Agent Hong Kong · AI Agent Development Company
AI Agent Development in Hong Kong
Super Cat Technology builds AI agents that act on live systems: voice agents that answer and book, customer service agents that hand over with context, and workflow agents that collect payment. The 24/7 AI call centre we built answers 200,000 calls a month across 60 US HVAC brands.
Every agent is secured before it gets tool access: its own credential, the least access the task needs, a person who can step in, and a trace of every run. That is what the PCPD and the HKMA now ask of agentic AI in Hong Kong.
Expert-Vetted top 1% · 100% JSS (source: Upwork profile, October 2026) · production AI at a 5,000-employee US enterprise and a Big 4 firm
What can an AI agent do for a Hong Kong business?
Take a whole task off a team, not just answer questions. An AI agent reads a request, calls your systems, acts and hands over when unsure. Super Cat Technology built one that answers 200,000 calls a month across 60 US HVAC brands, booking 67% of calls against 77% for human agents.
Read the 24/7 AI call centre case studyHow do you choose an AI agent development company in Hong Kong?
Ask to see an agent in production and who runs it. Super Cat Technology builds agents on your systems from Hong Kong and London, time-and-materials, and hardens tool access before launch. Cat Yung leads the work: Expert-Vetted on Upwork, top 1% platform-wide, 100% Job Success Score (source: Upwork profile, October 2026).
Who is Cat Yung?Can an AI voice agent answer customer calls in Hong Kong?
Yes, if it runs on real telephony and books against live systems. The voice agent we built answers 200,000 calls a month for a 5,000-employee US enterprise, around the clock. For a Hong Kong line, Cantonese, English and code-switching callers need their own test set of real call transcripts before launch.
How the voice agent handles 200k calls a monthWhat does an AI customer service agent need before it goes live?
Four things: approved answers it cannot improvise beyond, a handover to a person with the context written down, regression tests on real transcripts, and a trace of every conversation. Those are the controls on the call centre we built, which books 67% of calls against 77% for a trained human team.
See the six design decisions behind itWhat does an AI agent solution for a Hong Kong enterprise include?
A scoped workflow, the agent, and the controls around it. We map one task end to end, connect the agent to your systems with the least access it needs, build an evaluation set from real cases, then ship with tracing and a human fallback. Our AI payment-collection agent collected over US$1M in its first three months.
Read the AI payment collection case studyWhat do the HKMA and PCPD expect from a secure AI agent in Hong Kong?
Accountability you can show. The PCPD’s agentic AI guidance of 25 August 2026 keeps the deploying organisation responsible for personal data its agents handle. The HKMA’s ART Framework of 15 October 2026 flags a discussion paper exploring a registry of material AI agents. Secure AI agent development means least-privilege access, human approval for risky actions, and full logs.
Secure AI agent developmentThe Proof
Which AI agents has Super Cat Technology shipped?
Two agents in production: one answers the phone, the other collects what is owed.
24/7 AI Agent Call Center: 200k Calls a Month Across 60 US HVAC Brands
200k calls/month · 67% booking rate vs 77% humanFor an HVAC brand the front desk is the business: a missed call is a missed booking, and after-hours and peak-hour calls went to voicemail that most callers never leave. Across 60 brands under one 5,000-employee US enterprise, staffing every shift for that volume was hard to justify, and leaving it unanswered cost real business.
An AI front desk that runs the call centre 24/7: real telephony with barge-in handling so callers can interrupt naturally, booking, rescheduling and cancelling against live availability, answers only from what each brand has approved, and tool-calling into live systems so the agent can look things up and write back rather than just talk. It started as an Upwork engagement at MVP stage, 247 hours over three months from June to August 2025 and rated 5.0 by the client, and went into production from there.
Read the full case studyAI Payment Collection and Reminders
US$1M collected in the first 3 monthsSmall businesses are owed money by customers they want to keep. Chasing payment is awkward, easy to postpone and usually done late, by the same person who should be doing billable work, so receivables age and cash flow suffers.
An automated collections flow: reminders sent on a schedule the business sets, over the channels customers actually read, each carrying a secure payment link; replies are read and routed, disputes go to a person, and the sequence stops as soon as an invoice is paid.
Read the full case studySecure AI Agent Hong Kong
How do the HKMA and PCPD rules change an AI agent build?
Four controls we design in from the first session, each tied to the guidance a Hong Kong bank, insurer or data user will be asked about.
The organisation stays accountable
The PCPD’s guidance on Protecting Personal Data Privacy in the Use of Agentic AI (25 August 2026) treats the deploying organisation as the data user under the PDPO. An agent is not a legal person, so its actions are yours. We document who owns each agent, what personal data it touches and which supplier does what.
Least privilege and segregation
The same guidance sets out nine recommendations, including data minimisation and security safeguards such as environment segregation and network controls. In practice: each agent gets its own credential, scoped to the tools one task needs, in its own environment.
A person can step in
The PCPD recommends continuous risk assessment with a human in the loop, and the HKMA’s ART Framework (15 October 2026) names effective intervention mechanisms. Our agents hand over with the context written down, and risky or irreversible actions wait for approval.
An entry for the agent registry
The HKMA reports that more than 60% of use cases in the second GenAI Sandbox cohort explored agentic AI, and plans a discussion paper on agentic AI governance exploring a registry of material AI agents. Every agent we ship comes with a trace of each run and a written record of what it may do, so a bank can fill in that entry.
Sources: PCPD, Protecting Personal Data Privacy in the Use of Agentic AI, 25 August 2026; HKMA, ART Framework press release, 15 October 2026. General guidance on how these rules shape an agent build, not legal or regulatory advice.
FAQ
AI Agents in Hong Kong, Answered
How is an AI agent different from a chatbot?
A chatbot answers; an agent acts. The call centre agent we built checks a live calendar, then books, reschedules or cancels in the scheduling system instead of taking a message. That is why an agent needs more design than a chatbot: every tool it can call is something it can get wrong, so tool access is scoped, logged and tested before launch.
Can the agent run on our own infrastructure?
Yes, where the data cannot leave. We have already shipped AI fully on a client’s own infrastructure: an on-premise retrieval system for a Big 4 firm, tuned on its own H100 cluster. An agent that reads confidential documents can sit on the same kind of stack, and the retrieval half is described on our Hong Kong RAG page.
How does an AI agent engagement run?
A five-day discovery sprint maps the workflow, the data and the constraints, prototypes the riskiest part and scores each use case by value, effort and risk. Embedded delivery follows on time-and-materials, with an evaluation suite from week one. Afterwards you either keep us on a monthly retainer to run and monitor the agent, or take a handover of code, evaluation suite and runbook.
Agents that read confidential documents need retrieval first: the Hong Kong RAG solution.
Bring us the task your team keeps putting off.
The after-hours call, the overdue invoice, the request that waits for a free pair of hands. We scope one, build the agent and harden it before launch, led by Cat Yung. The security side is covered in secure AI agent development.
First-Sprint Data Scientist
A vetted engineer embedded in your team, a shipped deliverable by end of sprint 1.
Pick a time below and you will speak with me directly. Bring the use case, the data and the constraints, and I will give you a straight view on feasibility, effort and the first thing to ship.
Book a callCat Yung
Hong Kong & London time zones
Book a call · 30 min with Cat Yung