26th December 2025 - (Hong Kong) A humanoid robot and a robot dog -- both reportedly wearing foodpanda courier uniforms -- were filmed making deliveries in San Po Kong, and the footage ricocheted across Threads with the familiar mixture of wonder and dread. The spectacle was easy to enjoy as a novelty. However, Hong Kong should resist the temptation to file the incident under "cute tech demo" and move on. This looked less like a gimmick than the opening scene of a labour-market transition that will happen faster than most couriers, policymakers and even platform executives publicly admit.
Within an hour of the post spreading, Hong Kong AI and robotics entrepreneur San Wong explained that the sighting was part of an early test conducted during a short holiday break. The robot dog, he said, carried an insulated delivery bag and used a locally developed navigation system built in Hong Kong. The system can handle basic autonomous navigation but is not yet trained for complex scenarios such as crossing roads. Human minders therefore accompany the robots to reduce accident risk and comply with current rules, which do not permit fully unsupervised roaming in public space. The humanoid robot's navigation remains even more experimental, Wong added, and full automation in messy, real-world streets is still some distance away -- though he hopes one day Hong Kong will be filled with robots running on home-grown navigation.
Those caveats are accurate. They are also, from a labour perspective, oddly reassuring in the way early warnings often are: "Not yet", "still some way to go", "needs a minder". These phrases comfort people into believing displacement will be slow. Yet in technology adoption, the distance between "not yet" and "now common" is rarely measured in decades. It is measured in deployments, regulations and unit economics.
Food and goods delivery platforms employed nearly 13,000 workers locally according to a government survey conducted from December 2023 to March 2024: 12,900 people reported platform delivery work, with about a third working for more than one platform. Nearly 65% said delivery was their main source of income. The workforce is not only young -- 52.5% were aged 15-39 -- but also includes a substantial older segment: 24.2% were aged 50 and over. Earnings are spread thinly: 37.1% reported at least HK$15,000 a month; 26.1% between HK$5,000 and HK$14,999; and 34.9% below HK$5,000. Hours vary: almost all could arrange their own working time; about a third worked under 25 hours weekly, while about a quarter exceeded 44. A majority received two to three orders per hour. Couriers used an assortment of modes -- cyclists, walkers, motorcyclists, van drivers -- and operated under the now-standard architecture of rewards and penalties tied to acceptance rates, punctuality, customer complaints and "online rate".
This is exactly the sort of labour pool automation targets first - large, fragmented, price-sensitive, and managed by algorithmic systems that already quantify performance as if the worker were a component in a machine. When work has been reduced to metrics, it becomes easier -- morally and operationally -- to replace the person who generates them.
The next ingredient is the platform incentive. Delivery is a brutal business i.e. customers dislike fees, restaurants dislike commission, and platforms dislike paying humans to do physically demanding last-mile work in bad weather and heavy traffic. Labour is the largest variable cost. If a robot can do even a portion of deliveries at lower marginal cost -- particularly during off-peak times, in dense estates, or on repeatable routes -- the platform has a clear economic motive to shift volume away from humans.
There is a common retort - robots are expensive. So are humans, once you count recruitment churn, support costs, insurance, disputes and the inefficiencies created by variable availability. A robot is capital expenditure; a courier force is an ongoing operational expense with volatility. Finance has a long history of preferring capex when it can be depreciated and scaled.
The third ingredient is global precedent. Delivery robots are already used in controlled environments -- campuses, residential estates, business parks -- in places such as the United States and Singapore. In the UK, Uber Eats has announced a partnership with Starship Technologies to deploy autonomous delivery robots in multiple markets, with Starship operating thousands of units and boasting millions of deliveries. Starship describes its devices as "level four" autonomy for their domain: able to navigate pavements and crossings without human supervision using sensors and AI trained on massive operational data.
Hong Kong is not a campus and San Po Kong is not a private estate. However, the direction of travel is unmistakable - autonomy improves most quickly where there is high repetition and rich data. Delivery offers both. Every journey supplies training material; every curb, zebra crossing and lift lobby becomes a datapoint. As hardware falls in cost and software improves, the question is not whether robots will match humans everywhere. It is whether they will become "good enough" in enough places to change the labour market.
That is where Hong Kong's geography becomes decisive. The city is dense, with concentrated demand, predictable building clusters, and a heavy reliance on lift-enabled vertical travel. At first glance, that sounds hostile to robots - tight pavements, relentless obstacles, impatient pedestrians, torrential rain. Yet density is exactly what makes automation profitable. Shorter distances reduce energy and time costs. Repeat orders in the same estates make routes learnable. Concierge desks, security guards and building management offices can become part of a controlled "handover protocol" that replaces the informal negotiation couriers perform today.
In practice, robot delivery in Hong Kong will not arrive as a single, dramatic switch. It will arrive as a patchwork.
Phase one is what San Po Kong likely previewed - supervised trials in public space, and unsupervised operation in semi-controlled areas. The first winners will be large housing estates, business parks, university campuses and new developments with wide walkways and consistent signage -- places where the "edge cases" that break autonomy can be engineered away. Platforms will quietly steer suitable orders to robot fleets and leave humans to handle the complex remainder.
Phase two is regulatory accommodation. Wong noted current rules require human minders and do not permit full autonomy in public space. That is true today; it will not remain true forever. Regulation typically follows three steps: prohibit by default, permit trials, then legalise with conditions once the technology has political cover. The political cover comes from safety statistics and from the promise of "innovation leadership". A locally developed navigation stack -- "made in Hong Kong" -- adds an industrial-policy gloss that is catnip to officials who want the city to be seen as more than a property-and-finance hub.
Phase three is substitution through "augmentation". Platforms will describe robots as helping humans rather than replacing them. This will be partially true in the early period. A robot may handle the long, flat segment of a trip while a person handles the handoff; or a courier may become a roaming supervisor, managing multiple robots at once. But "augmentation" is often a corridor, not a destination. Once a single worker can oversee several machines, the maths becomes unforgiving: the labour demand per delivery falls.
This is the part the gig economy is least prepared for. Hong Kong's platform couriers are not a workforce with clear progression ladders inside the companies that rely on them. They are treated -- contractually and often rhetorically -- as flexible supply. If robots reduce the need for that supply, the market will not politely redeploy thousands of cyclists and walkers into "robot technician" roles. Some will transition, yes. But the number of new technical jobs created by a fleet of delivery robots is usually small compared with the number of human couriers displaced. Starship itself has been held up as an example of scale with a relatively lean headcount.
Nor is this simply a story about job losses. It is about bargaining power. The more credible the robotic alternative becomes, the less leverage couriers have to argue for better pay, safer conditions, or fairer penalty systems. A threat does not need to be executed to be effective. If platforms can point to pilot robots waiting in the wings, negotiations -- formal or informal -- tilt quickly.
So what should Hong Kong do, beyond gawp at a robot dog in a courier uniform?
First, regulators should stop treating delivery automation as a curiosity and start treating it as a transport-and-labour policy issue. If trials expand, the city needs clear rules for sidewalk priority, speed limits, right-of-way, insurance, data collection, and liability. "A minder was present" is not a policy framework; it is a temporary patch. Second, labour policy should anticipate uneven displacement. The government survey shows many couriers rely on delivery as their main income and that a meaningful share are older workers -- precisely the group least served by breezy "reskill into AI" slogans. Any credible plan must include transition support that matches the reality of the workforce: targeted training, wage supplements during retraining, and enforcement against abusive reward/penalty mechanisms that intensify as platforms squeeze remaining human labour.
Third, Hong Kong should decide what it wants to be. Wong's aspiration -- streets filled with robots using local navigation -- could be a genuine industrial opportunity. However, it cannot be pursued on the assumption that the social consequences will manage themselves.
The San Po Kong sighting was striking because it was literal: machines walking the same pavement as the people whose work they may soon absorb. It is tempting to see it as theatre. It would be wiser to see it as a timetable. The replacement of Hong Kong's roughly 13,000 platform couriers will not happen all at once, and it may not happen completely. Yet the direction is clear -- and the pace, driven by economics and enabled by policy, is likely to be quicker than the city expects.