Dinner Goes Out at 5:30
Working the floor in fine dining means you ship every single night. Not in the startup sense. Actual ship: 160 covers plus the bar and walk-ins, most of them marking something (an anniversary, a deal closing, a first date), all of them expecting perfection and to feel wowed. The house either gets it right on the night, or it doesn't. There's no "we'll launch next week." There's no "still working on it." At 5:30, service starts, whatever shape the prep is in.
My path to AI building: a restaurant floor in Florida, eight years across New York dining rooms (Eleven Madison Park, Maialino, Legacy Records, Charlie Bird), four and a half years at Cision going from support to account executive, now here.
Three different careers on paper. Same work in different uniforms in practice: context management, systems work, orchestration.
Most of what I know about shipping I learned on a floor.
Mise en Place Is Context Engineering
Walk into any serious restaurant at four o'clock. Not the kitchen yet. The dining room. Glassware being polished, tablecloths ironed flat on the table, napkins folded, centerpieces lined up with the edge of a bread knife (nobody carries a ruler at 7pm). Reservations marked with allergies and occasions (birthday, anniversary, first date, regular). By preshift, each server already knows who's sitting at Table 14, what they drank last time, what the kitchen is 86'ing tonight, and what the two new specials are down to the last garnish.
That's mise en place. French for "put in place." It's the invisible work that makes service look effortless.
It's also context engineering, a decade before anyone called it that.
A well-run dining room is a team of servers who share context about the menu, context about the wine list, context about every special running tonight, context about which tables are regulars and which are first-timers nervous about the price. The prep work is what makes the execution fast. Swap "server" for "AI assistant" and the structure holds. The tool is useful in direct proportion to the context it has before the ticket prints.
When I write about context as the only AI skill that compounds, I'm not drawing on AI research. I'm drawing on eight years on a dining room floor.
The "Oui" System
In a serious kitchen or dining room, when someone tells you something, you say "Oui."
That's it. One syllable. Not "yes chef I heard you and I'll handle that right now." Not "got it, I'm on my way." Just "Oui."
It's an acknowledgment that packs three things into one syllable: I heard you, I understand, I'm on it. Under pressure, repeating the whole request back wastes time you don't have. The word closes the loop so fast nobody has to confirm twice.
The AI version of this is underrated. Good agents work the same way. They acknowledge what you asked, lock onto it, and execute. They don't narrate their intention back at you for five paragraphs. They don't confirm that they're about to confirm. They take the order and go.
Teams that get real value out of AI tools are the ones who build that rhythm. Short asks. Fast acknowledgment. Execution. No preamble, no victory lap. Oui, and keep moving.
Lunch Somm
For a stretch I wanted to be a sommelier. Took every chance I could to learn wine and sell it. At Maialino they called me the lunch somm because when I was on the floor as dining room manager, they didn't bother scheduling one.
Selling wine in a fine-dining restaurant is pure consultative selling. You read the table. The occasion, the budget they haven't said out loud, the palate signal from what they already ordered, the regular you recognize from a month ago and can't quite place. Then you match a bottle to what's in front of you. Not the most expensive option. Not the one the distributor pushed last week. The bottle that would actually make their meal better.
Every consulting conversation I have now uses the same shape. Listen until you understand what they actually want, which is never exactly what they asked for. Read the room. Recommend the thing that would genuinely improve the outcome, not the thing that's easiest to sell.
Wine service also taught me the hardest sales discipline: knowing when the answer is "you don't need anything more expensive than this." Nothing builds trust faster than steering somebody down instead of up.
Ship Every Night, Recover Every Night
Fine-dining restaurants don't run on perfection. They run on recovery.
Every service, something goes sideways. A VIP walks in without a reservation. The kitchen 86's the signature dish at 5:50. A four-top gets double-sat and nobody notices for ten minutes. A regular's usual wine is the one you just poured for the table next to them and there's one bottle left in the cellar.
The test of a floor manager isn't whether those crises happen. It's how fast you're back in rhythm.
That reflex transfers. The public builds I've shipped over the last year (a multimodal route planner, a voice AI portfolio, a multi-agent task system, a handful of smaller things) all had moments where the thing broke in production, the agent got stuck, or the idea hit a wall. The reason they shipped anyway is the same reason you finish service when the reservation book blows up an hour in. You don't quit. You commit the last good state, you reset, you pick up the next guest.
The startup language for this is "iterate." The dining room language is shorter: "next table."
Scope Is Mercy
The chefs I worked for cut scope ruthlessly. A new dish going on the menu starts as five components in testing. By the time the first ticket prints, it's three. Sometimes two. The things that stayed are what the line could execute consistently when the house was on fire, not the things that tasted best in a quiet moment at three in the afternoon.
That instinct is the thing most AI builders don't have when they start. They describe a product that would take six months to ship and want it tomorrow. They haven't yet had to answer the question a chef answers twenty times a week: what's the simplest version of this that still delights the guest?
Hospitality teaches you that scope cutting isn't about doing less. It's about doing the thing that works, repeatedly, at a real pace, for the people who showed up tonight.
Seller Who Builds, Builder Who Sells
Between the dining rooms and the AI building, I spent four and a half years at Cision. Started in support, moved to account management, and ended as an account executive on the client development side (so, existing customers expanding, not net-new logos).
Enterprise B2B looks nothing like a dinner service. Except it does. An account manager is an expediter translating between the kitchen (product, engineering) and the guests (customers). You carry context for both sides. You know which accounts are running hot. You know what the kitchen can execute and what it can't.
Sales taught me a different version of the same service skill. How to listen until you understand the actual problem, not the stated one. How to translate capability into outcomes a buyer cares about. Closing a conversation with a concrete next step instead of a vague "let me send you some materials."
Seller who builds, builder who sells. The overlap's smaller than most people think, and more useful than any single skill.
The Throughline
Hospitality is systems work dressed up as service. Product building is service work dressed up as systems. The server who cares whether the seared scallop is hot when it lands in front of you is doing the same job as the founder who cares whether their tool helps the person using it. The vocabulary differs. The attention is the same.
All three chapters (dining rooms, Cision, building) are variations on the same three disciplines: context management, systems work, orchestration. You carry what the guest or the account or the user needs. You build the rhythms that make execution fast. You keep the handoffs tight when four things are happening at once.
Most of what I write about on this site (the tools I'm building, the workflow I use, the teaching I'm doing) traces back to something a server or a chef or a sommelier modeled for me first. If you're reading and you came from a non-traditional background, that's the point. Some of the best preparation for shipping with AI is having shipped something else, under pressure, to a real person who's paying attention.
More of that pattern shows up here, one build and one lesson at a time. If you want to follow along, the rest lives at /writing.