Dim Sum Cart to Tablet Ordering: How the Bay Area Lost the Lady With the Cart

The Room That Ran Itself

On a Sunday morning in 1994, the dining room at Yank Sing on Stevenson Street in San Francisco held two hundred people and approximately zero servers taking orders. The room ran on momentum. Women with steel carts moved in slow ellipses between tables, calling out dish names in Cantonese, lifting bamboo lids, stamping cards with red ink. A table of eight could eat for ninety minutes without making a single formal request. The food came to you. You said yes or no. That was the entire transaction.

This is what dim sum cart service was, at its operational core: a restaurant that distributed its menu spatially rather than textually. There was no printed list, no waiter standing with a notepad, no moment where a family had to agree on what to order before the food appeared. The food appeared, and then you agreed. The sequence mattered. It made the meal feel abundant even at a $12-per-person price point, because abundance was baked into the delivery model itself.

By 2024, most of that was gone. The carts at Yank Sing had been gone for years. Koi Palace in Daly City had moved to a checklist. Pacific Restaurant in the Richmond had a QR code propped against the soy sauce. Across the Bay, Legendary Palace in Oakland's Chinatown was one of the few large-format rooms still running partial cart service on weekends, and even there the cart only covered half the room. The rest of the floor ordered off a laminated sheet.

The transition did not happen all at once, and it did not happen for one reason. It happened over roughly a decade and a half, accelerated hard by the pandemic, and it is now functionally complete. The era of the dim sum cart, in the Bay Area at least, is over as a dominant format. What replaced it is more efficient, more legible to non-Cantonese-speaking diners, and significantly less interesting to sit inside of.

The Labor Math Nobody Wanted to Say Out Loud

Ask a dim sum restaurant owner why the carts went away and the first answer you will get is the pandemic. The second answer, if they trust you enough to give it, is labor. The real answer is that those two things arrived together and made a problem that had been building for twenty years impossible to ignore any longer.

A full cart-service operation at a 200-seat dim sum house requires somewhere between twelve and eighteen cart attendants working a Sunday morning shift, depending on the size of the room and the volume of the kitchen. These are not unskilled positions. A good cart attendant knows the entire menu, can read a table's pace, knows when to push and when to hover, and can manage the temperature of three dishes simultaneously while navigating a room where every aisle is forty-two inches wide and someone's grandmother is always sitting in the exact center of it. The job takes months to learn and years to do well.

Through the 1980s and 1990s, the labor pool for this work was stable. Hong Kong immigration to the Bay Area was high, Cantonese-speaking restaurant workers were plentiful, and the internal economics of the community restaurant meant that wages could stay low because the job carried other forms of value: proximity to the community, a flexible schedule, tips that were shared in ways that suited multigenerational households. By 2005, that pool was thinning. By 2015, the owners of places like Ton Kiang on Geary and Hong Kong Lounge on Balboa were running Sunday service short-handed and absorbing the difference in pace.

The tablet and the checklist solve the labor problem cleanly. A 200-seat room that needed sixteen cart attendants can run on six or seven floor staff if the ordering moves to paper or screen. The kitchen still needs full staffing, but the floor becomes manageable. The margin math changes enough to matter. For a restaurant operating on 8 to 12 percent net, "manageable" is not a comfort metric. It is the difference between staying open and not.

The Dish explored a similar dynamic in pho, where the economics of the strip-mall Vietnamese restaurant model created its own set of labor and real estate constraints, at The Dish's feature on strip-mall pho culture. The pattern holds across immigrant food formats: the model that built the cuisine's American footprint is often the first model to break under rising costs.

What Real Estate Did to the Cart

There is a physical fact about cart service that the conversation about labor tends to skip: the carts need room. A standard dim sum cart is roughly eighteen inches wide and five feet long. A room running six to eight carts simultaneously needs aisles wide enough for two carts to pass, which means a table layout that sacrifices covers. A 3,000-square-foot dining room configured for cart service holds fewer tables than the same room configured for conventional service. In a city where restaurant rents in 2019 were running $80 to $120 per square foot annually, that spatial inefficiency had a dollar value, and it was not small.

The large-format dim sum palace was, economically, a 1970s and 1980s product. The buildings it required, the lease structures it depended on, and the neighborhood economics that made its price point viable were all functions of a San Francisco that no longer exists. The Empress of China on Grant Avenue, which ran cart service in a sixth-floor ballroom for four decades, closed in 2014. The building it occupied is now a boutique hotel. Harbor Village in the Embarcadero Center closed in 2000. Asia Garden in the Financial District closed shortly after. In each case, the building outlasted the restaurant and found a more lucrative tenant.

What replaced them was a smaller format. The dim sum restaurants that opened in the 2010s, particularly in the suburbs, Millbrae, San Mateo, Cupertino, built rooms in the 100-to-150-seat range rather than the 300-to-500-seat range. Those rooms are physically incompatible with full cart service. The carts have nowhere to go. The checklist is not a concession; it is the only thing that fits in the space.

Fook Yuen Seafood Restaurant in Millbrae and Dragon Beaux in the Richmond both opened in this smaller format and both went to checklist ordering from day one. Neither owner framed this as a loss. They framed it as a design choice: faster turns, more legible ordering for non-Cantonese speakers, less waste from dishes that circled the room and came back cold. The framing is honest. The format works. It is still not the same thing.

The Checklist Generation

The children of the cart attendants and the cart-era restaurant owners are, in a lot of cases, now the ones running these restaurants. They watched their parents work doubles on Sunday morning. They watched a woman push a 60-pound steel cart from 9 a.m. to 2 p.m. for tip income that worked out to eleven or twelve dollars an hour in the good years. They are not sentimental about the cart.

This is not a complaint about them. It is a structural observation about what happens when a labor model depends on a first-generation willingness to absorb conditions that the second generation — rightly — will not accept. The dim sum cart ran, in significant part, on the labor of women who had few other options and who built community and income inside a job that the broader economy would not have valued at its actual difficulty. That labor is gone because the conditions that produced it are gone. The checklist is what you get when you stop extracting labor from people who could not say no.

The tablet ordering systems that showed up in the mid-2010s, and that accelerated sharply after 2020, added a third logic on top of labor and real estate: data. A checklist tells the kitchen what to make. A tablet tells the kitchen what to make and records what the table ordered, when they ordered it, how long they sat, and what they didn't finish. For a restaurant group managing multiple locations, that data has real operational value. For a single family-owned house trying to survive a rent increase, it is mostly overhead. But the systems are marketed to both.

Some restaurants have tried to keep partial cart service as a hybrid. Yue Restaurant in the Outer Richmond runs carts for the first hour of service on Saturday and Sunday, then switches to checklist when the carts empty out. It is an honest compromise. The room still gets the theater of the cart for the peak period, the kitchen gets the data of the checklist for the back half, and the owners get to say they still do dim sum the old way, which they almost do.

What the Algorithm Sees in the Shift

The scoring pattern across Bay Area dim sum restaurants is not what you would expect if the tablet were simply a neutral substitution for the cart. Flavor scores hold. The kitchens are the same kitchens, the recipes are the same recipes, and a plate of har gow from Koi Palace in 2024 is still excellent. The algorithm has no complaint with the food.

Context scores are where the separation appears. The cart-service rooms that remain, Legendary Palace in Oakland, the partial-cart operations on the Peninsula, a handful of surviving Saturday morning rooms in the Sunset, score significantly higher on context than their tablet-ordering counterparts in comparable cuisine categories. The gap is not small. It tracks against everything that cart service does that tablet ordering cannot replicate: the ambient noise of metal lids, the social permission to take something just because it appeared, the way a table of eight self-organizes around a shared choosing process rather than an individual list.

This is the thing the efficiency argument misses. Cart service was not inefficient. It was differently efficient. It produced a dining experience where the pace, the abundance, and the social choreography of the table were all managed by the room rather than by the diners. Tables turned because the food kept coming, not because the server kept coming back. The pacing was external. When you remove the cart, you push all of that pacing back onto the diners, and most tables don't know what to do with it. They order conservatively, they order slowly, and they leave having eaten less well than they would have if the cart had decided for them.

Value scores are the other surprise. The tablet restaurants are not cheaper. In many cases they are more expensive: Dragon Beaux runs $60 to $80 per person at Sunday brunch, a price point that would have been unthinkable in a cart-service context. The premium is real and it is not entirely unjustified — the food is technically accomplished and the room is beautiful — but it represents a fundamental repositioning of dim sum as a category. The format that fed a family of six for $40 is now a $240 outing. Something changed and it was not only the ordering system.

The Places Holding the Line

Not everyone moved to the tablet. The holdouts are worth naming.

Legendary Palace. China Village. New Gold Medal Restaurant. These are the operations that kept partial or full cart service through the pandemic and into 2024, mostly in Oakland's Chinatown and in the Richmond and Sunset districts of San Francisco, mostly in buildings with older leases and owners who are either old enough not to care about the data or young enough to understand what they'd be giving up.

The rooms that kept the carts share a few things. They tend to be large, 150 seats and up. They tend to be in neighborhoods where the Cantonese-speaking customer base is dense enough to fill them without relying on crossover traffic that needs English-language menus. They tend to have staff who have been there long enough that the cart labor model is absorbed into institutional knowledge rather than continuously retrained. And they tend to have owners who understand that their competitive advantage is not the food alone but the experience of eating the food in a room that still feels like itself.

The analogy that keeps surfacing, when you talk to the people running these rooms, is the premium-ification pattern that remade the pastry market: the thing that was cheap and abundant becomes a specialty product, the specialty product gets priced accordingly, and the original customer base that built the category loses access to it. The croissant became a $7 object. Dim sum is becoming a $70 meal. The food at both price points can be excellent. That is not the point. The point is who gets to eat it.

There are also the weekend operations that are too small to be noticed by the trend pieces but too consistent to be dismissed. A room seating forty people in the Outer Sunset. A restaurant in San Jose's Japantown corridor that does cart service on Sunday morning only, by reservation only, in a space that holds maybe sixty covers. These are not going to save the format. But they are evidence that the format is not dead. It is just rare in the way that things become rare when they stop being economically rational for the majority and start being a choice that the few make deliberately.

What the Cart Was Really For

The cart solved a problem that is easy to forget existed. In the 1960s and 1970s, when Hong Kong immigration into San Francisco and Oakland was accelerating and the community restaurant was one of the primary social institutions for people who had arrived recently and didn't yet have much else, the cart made large-format communal dining accessible to people who didn't have a lot of money and didn't have a lot of time to negotiate a menu. You sat down. Food appeared. You ate as much as you wanted of what you wanted. You paid based on what you took. The transaction was honest and the threshold was low.

It also made the restaurant a viable place to bring grandparents who didn't speak the language of wherever they had arrived, children who would not sit still through a menu-reading process, and extended families whose members had different preferences and different budgets. The cart collapsed all of that complexity into a single social format that worked for everyone simultaneously. This is a harder design problem than it sounds. Most restaurant formats don't solve it. The dim sum cart solved it, and it solved it for about fifty years in the Bay Area before the economics came for it.

The food trend comparisons are instructive. The Birria Boom: How a Jalisco Wedding Stew Became America's Most-Scored Taco tells a version of the same story from the other direction: a format that was communal and occasion-specific gets extracted from its context, stripped down to its most portable element, and spread across a market that consumes it without the social infrastructure that originally made it make sense. The birria taco is not a wedding. The tablet-ordered dim sum is not a Sunday morning in a room that smells like sesame oil and has been running since 7 a.m. The food survives the translation. The thing around the food does not always make it.

The cart was the room. It was the pace, the abundance, the way four generations of a family could sit together and have everyone get exactly what they wanted without any of them having to be in charge of the decision. When you take the cart away, you have a meal. When you had the cart, you had a room. The distinction is not sentimental. It shows up in the data.