We can optimize everything except dinner
What happens when AI agents are more in charge of your meal?
One of the best ways to spend an evening in New York right now is a Stars-to-Le Veau d’Or-to-Stars boomerang.
This is especially true on a Monday, when the crowd populating America’s best new wine bar is at its hottest and least inundated by TikTok-influenced East Villagers. J Lee has my favorite encapsulation of what makes this tiny room from the Claud and Penny team so special. I popped in six times over two weeks despite being on a drinking hiatus, making the stools at the U-shaped bar a temporary home, one where I could blissfully ignore how much sugar was in my N/A, rosemary-scented glasses of Villbrygg Fjell 01.
Wine bars like Stars work best during the romantic hours of a before-dinner pregame or an after-dinner refusal to go home. They are spaces of anticipation and denial. We left Stars around 7:30 to take the 4 uptown for a celebration dinner at Le Veau. When we made it back three hours later, the room had mostly cleared out. To our right was one regular with subtle, impeccable wine knowledge. To our left was a couple happy to drink anything at all. Then the staff snapped into gear, having heard a group of wine collectors were popping in ready to spend.
This is a big reason Stars works. It is accessible enough — both in some bottle prices and in the way the staff meets you at any level of wine interest — that it does not feel like a club for snobs. But it still knows how to serve these high-ticket collectors. Thirty minutes before the listed closing time, eight people slid through the doors, drunk and boisterous. I could not name what bottles got pulled, but I could feel the anxiety of whatever Amex was about to hit the bar. I got roped into a cheers with one of these strangers, a sommelier at a multi-Michelin-starred restaurant in the city. He asked what I did for work.
“Oh, I’m at an AI startup.“
“Bro! AI. I deal with that shit every day.”
“Oh, yeah? Are y’all using it for inventory?”
“No, bro. No one will talk to me.”
More and more, my guy finds himself walking up to a table before the tasting menu begins, ready to start a conversation. Instead, someone quietly points at a line on the wine list. Before he’s sent off to fetch, he looks down and sees a ChatGPT conversation open on a phone, recommending that exact bottle.
AI and agentic efficiencies are coming for most parts of our lives. I live this every day, working with a small team to push at the edge of what’s possible with frontier models. I’m far more AI optimist than alarmist. But these kinds of stories make me anxiously protective of one favorite thing to do in the world: go to a great dinner with people who matter to me, meet the restaurant on its terms, and let the night become more memorable than anything I could have orchestrated online.
For every bit of ease that brings a model-recommended, optimized experience closer to your fingertips, I find myself retreating from the impulse to inject these tools into the parts of life that are better with some mystery, some discovery and some much-needed romance.
I understand the urge to ChatGPT your way into a better meal.
Dinner in New York has gotten too expensive to approach casually. That’s compounded by reservation scarcity, social media hype, Substack guides, Infatuation lists, and the ambient panic of wanting to get all the must-order dishes at the newest, best spot. It no longer feels reliable to drop in, order whatever looks good, and leave the variables up to chance.
Technology has made it seamless to query the consensus best version of a night. This bottle, with this menu, at this restaurant, at this time. You can construct the right moves, at the cost of allowing for surprise and delight. If food tasted better 10 years ago, acceleration here is a big reason why.
I’m not above engaging in aspects of this game. I’ve got my Resy hacks, and that Le Veau d’Or table was only possible due to OpenTable’s private booking link feature — a reward that arrived in my inbox one day either due to crossing a minimum Substack subscriber threshold or spending years yelling about Frenchette having perfect pancakes. Most of my Torrisi splurges happen via Dorsia. But that’s basically my new limit: Technology can get me in the door for an occasion if nothing else will suffice, but then it disappears. Once inside, I only want my phone out to take a couple pictures of my friends or the food. And I want one or two recommendations from the people working the floor, not AI-compiled gameplans at which I’ve witnessed more diners glancing.
Few establishments do restaurant romance like Le Veau d’Or. It does not matter if you are there on a date or a celebration with friends. The room hums. The staff is flirty and fun. Their spiels are actually good. The weirder avenues you trace the food down, the more you’re rewarded.
We started with a pâté en croûte compliments of the kitchen, baked in a lovely pastry and served with a thrilling mustard. We ended with little cocktail shots balanced on an vintage sardine tin filled with pebble ice, a sweet sendoff gesture for a friend toasting her imminent New York departure. These are perks that come from regular dining, industry background and just having good chat. But technology is working on closing that gap.
LVD is on Blackbird, a company using tech to help restaurants and diners engage more deeply in the loyalty that fuels staying power and bustling dining rooms.
Regulars are the backbone of good restaurants, but recognizing and rewarding loyalty at scale is hard. Not every relationship can live in a host’s head, an Instagram DM, or a text thread, even if that’s my preferred loyalty mechanism. Blackbird gives restaurants infrastructure for a better process while rewarding diners with points systems, cash back, concierge booking, and access to exclusive events. When I spotted the Blackbird tap-in on the host stand at Raf’s, I felt a simultaneous respect for the product and a physical rejection of the technology’s place in one of Manhattan’s best rooms.
A friend and Blackbird fan just called me out on this, saying I’m an abnormal diner. He’s right. I briefly worked in restaurants. A lot of my friends still do. I have this small food blog. I’m also an extrovert who loves restaurant people and makes it a point to remember and connect with them. I understand that if you live next to Raf’s and want to put it in regular rotation, an app like Blackbird undeniably makes that more feasible. It might help the restaurant find you and reward you with a prime seating or a complimentary crudo. This is a mutually beneficial, lightweight version of the earned, in-person intimacy that has existed for decades.
In the middle of that Raf’s dinner, I looked over at the table next to me and saw that three grown adults had each ordered the whole dorade as their entrée. Maybe it was a health-informed decision. Maybe an internet search told them it was the go-to order. Maybe all three of them really wanted the same fish and refused to perform family-style curiosity for an onlooker’s benefit. Still ...
I do not go to restaurants to protect my own perfect plate from the table. I go to share, pass, steal, negotiate, over-order, get talked into indulgences, and let someone else’s appetite change mine. If you arrive closed off, no recommendation algorithm can make the night break open.
When the dining experience is made more efficient through technology, what else creeps into the dining room alongside it? There’s some non-zero risk of further disconnection from the already tenuous romance that makes the best meals elevate beyond consumption or status. When you take the heart of hospitality and turn it into yet another exchange on your own terms, one that’s only a click or agent away, you’re more likely to end up in your own little world, getting exactly what you want, sharing three identical plates of dorade, predictable and satisfied and bored.
A reliable sign you may regret spending more than $100 a person at a New York restaurant is realizing that you and your friend are the hottest people in the room. That was the energy recently inside an impressive yet stilted new Manhattan tasting menu spot. Less dinner party, more restaurant-gap relationships capitalizing on eager reservation notify triggers.
A friend and I were sitting side by side in front of these gorgeous tables, polishing off a dessert that was too clever by half. A member of the staff came over and plopped a big Square reader in front of us with the high-priced bill staring directly at both of our faces. A small touch carrying resoundingly bad vibes.
The financial exchange at a restaurant is part of the choreography. Who reaches. Who pauses. Who insists. Who slyly handles it away from the table. A custom billfold enclosing a nice pen is not just aesthetic affectation. It’s part of the dance. And the glaring blue screen is an unwelcome blunt object.
Like the apps, there are clear operational reasons restaurants drop these things on your table. They’re efficient, they’re easy, they reduce friction for the staff. But at that price point, in that room, after that much vibe-conscious labor, the device felt like an affront to a desired fantasy. In a few seconds, any hope of transportation was turned back into a workflow.
I am an AI enthusiast, consistently blown away by the tasks I do not want to do that agents can now take on for me. There are parts of my job I never thought I would be capable of doing that I can now execute at an extremely high level. Exploring this technology is challenging and addictive in ways I find endlessly fun and fascinating. And it’s likely essential to having a thriving career in the near future. But I also think having a fulfilling, joyful, romantic life requires knowing when not to reach for it.
My current rule is that as little AI as possible lives on my phone. When I am away from my computer, I want to be away from this world. I want to go for a long run. I want to read a book. I want to be present when my friends are talking. I want to sit in a gorgeous dining room and let the experience come to me rather than projecting an LLM-recommended, probably very good, not especially indelible night onto the room in front of me.
Anthropic and Resy recently announced an official integration. You can ask Claude for restaurant recommendations based on your taste and location, and it loads a Resy widget with availability straight into the chat. The three suggested prompts are illuminating:
“What are the hottest new restaurants in Austin with availability this week?”
“Book a table for 8 at a steakhouse in Chicago for next Thursday”
“Find a reservation for 2 at a romantic restaurant in the West Village this Saturday around 8pm”
You still leave Claude to complete the booking through Resy. And without a decent knowledge of AI coding tools, you are not yet able to set up agents that automatically monitor your hit list, grab the best table the second it opens, or learn your preferences over time. But you should assume all of that will be table stakes soon, and not just to Resy.
It will probably hit The Infatuation, which has a deep library of where to eat and what to order for every occasion. It may come through a personalized Blackbird AI concierge. The next great travel publication on this platform would be wise to gate a content-library-filled MCP behind a paid-subscriber-only authorization flow.
A lot of this will be varying degrees of useful. You should let agents remove the dumb little computer chores that keep you from the life you actually want. Moving all of my reservation booking from apps to agentic iMessages is my favorite tech innovation of 2026. But the more software gets better at solving the meal before it starts, the more I want to preserve the parts of dinner that require a little mystery.
Crossing into hour eight of Anthropic’s Code With Claude developer conference in San Francisco today, I tested out the Resy integration one more time:
Not quite. I closed my laptop and went outside to the rooftop reception. While servers passed out colorful mochi donuts, a Bay Area engineer expounded on the difference in grindset aura between SF and New York. He (maybe?) joked about wanting to have 10 kids to help curb the global fertility crisis and only having three years to build before getting stuck in the permanent underclass. He rejected the donuts. Then he turned his attention back to his phone.