On paper the sommelier looks exposed: recommendation is exactly what algorithms do, and an app scanning a wine list can suggest a pairing in seconds without salary or ego. Much of the job's information layer — what goes with what, what a region tastes like, what a fair price is — has already leaked out of professional heads into everyone's pocket.
But the job was never primarily information retrieval. It is sensory verification — is this specific bottle sound tonight — performed by a tasting human; it is trust built face to face in a dining room; and it is financial stewardship of a six-figure cellar. The pattern emerging is clear: software eats the desk work around the craft while the floor craft itself, like the chef's, stays stubbornly embodied.
Roughly a quarter of the work — pairing lookups, inventory management, list analytics, routine wine education — is realistically automatable now or soon. The core is not: no system yet tastes a bottle for cork taint at the moment of service, reads a table's occasion and budget from posture and hesitation, or carries personal accountability for a cellar's capital. The role most exposed is the entry-level recommendation function; the role least exposed is the trusted floor professional who verifies and vouches.
按任务计分，而非头衔。越低越安全。
AI难以取代的职业 →Cork taint, heat damage, premature oxidation — faults strike bottle by bottle, not label by label, and only a taste at service catches them. Electronic noses remain laboratory curiosities; the working fault detector in every restaurant is a human tongue.
A recommendation lands because a guest trusts the person making it — someone visibly accountable, reading the table's mood, budget and occasion in real time. That relationship is the product; an app suggesting the same bottle is not the same service.
Decanting a fragile 1985 off its sediment by candlelight, managing a twelve-course pairing's timing against the kitchen's, opening champagne silently at a funeral lunch — embodied skills performed live, with no automatable substitute in sight.
Building a cellar means judging producers years ahead of demand, verifying provenance in a market where fraud is real — collector Rudy Kurniawan's fakes fooled auction houses until 2012 — and staking the house's money on your own palate.
The thirty-second story that makes a table order the unfamiliar Savennières is theater, timing and permission — a persuasion craft tied to a present human. Generated text can describe a wine; it cannot stand at the table and be believed.
Apps and language models already produce competent pairing suggestions from a menu photo, and for the price-conscious majority of diners a free good-enough answer beats a professional one. The routine recommendation layer of the job is substantially gone already.
Stock counts, reorder triggers, bin mapping and sales analytics — hours of nightly clipboard work — are now standard software features. The midnight manual count survives mostly as discipline and audit, not necessity.
Margin optimization, sell-through analysis and dynamic pricing tools increasingly shape list decisions that once ran on instinct. The judgment of what belongs on the list stays human; the arithmetic beneath it no longer is.
The knowledge transfer that once required a sommelier's staff training — regions, grapes, service basics — is now free on video platforms and apps, eroding the teaching monopoly that justified part of the trade's status.
Preservation systems like Coravin and keg wine have shifted fine wine from bottle commitments to single pours, turning lists into tasting programs. The sommelier's craft moves from selling one big decision to orchestrating many small ones across a night.
Sake credentials, tea pairings and serious zero-proof programs are becoming standard sommelier territory — the world's top restaurants now build non-alcoholic pairings with the same rigor as wine ones, and the professional's title increasingly reads 'beverage' rather than 'wine'.
Top somms now run video channels, publish books and sell online courses, monetizing the credential far from any dining room. The floor remains the credential's proving ground, but for a growing share of the profession it is no longer the paycheck.
After the 2018 exam leak and the 2020 harassment investigation, the Court of Master Sommeliers overhauled governance and testing conduct, and the wider trade pushed hard on inclusion — scholarships and mentorship programs targeting women and sommeliers of color reshaped who enters the pipeline.
The sommelier's skill set scaled across a restaurant group or hotel company: cocktails, coffee, sake and wine under one P&L, more spreadsheet than tastevin, and the trade's most common promotion path off the floor.
Teaching the world's drinkers directly — courses, videos, newsletters, books — an economy that barely existed in 2010 and now supports former floor sommeliers as full-time educators with global audiences.
Recommendation platforms, online retailers and restaurant-tech firms hire sommeliers to train the algorithms and curate the catalogs — the trade's knowledge sold to the very tools automating its entry level.
Building non-alcoholic pairings and fermented drinks programs for restaurants where a growing share of guests do not drink — a young specialty applying the sommelier's structural palate to an entirely open category.
The split runs between the job's information layer and its verification layer. Everything that is knowledge about wine in general — pairings, regions, prices — automates, because it is text. Everything that is judgment about this bottle, this table, this cellar tonight resists, because it requires a tasting, trusted, accountable human in the room.
Employment will likely concentrate where that verification layer pays: fine dining, luxury hotels, private clubs and high-end retail, while casual restaurants lean on apps and staff training videos. Global wine consumption has been falling since the late 2010s even as spending per bottle rises — a premiumization that favors exactly the settings where sommeliers work.
The trade has one more defense that shows up in no automation analysis: it is theater people willingly pay for. Nobody orders the tasting menu to interact with a tablet. As long as celebration means a table, a bottle and an occasion, someone will be paid to make the opening of that bottle feel like part of the evening — and that someone will need a trained palate to back the performance.
不限同一领域，六项评分最接近的职业。
打造并复活腕上那台微型机器的工匠——一门熬过了石英危机、如今却人手比订单还紧俏的手艺。
抗AI 80 🧵The craftsman who cuts cloth to one particular human body — a trade the sewing machine industrialized, ready-to-wear shrank, and no machine has finished replacing.
抗AI 78 🎮游戏的建筑师——决定一款游戏向玩家提出何种要求的人，从古老棋盘游戏Senet到俄罗斯方块，直至今日拥有三十亿玩家的产业。
抗AI 65 🕹️从1972年斯坦福实验室的《太空战争》锦标赛到座无虚席的体育馆决赛——现代体育中最年轻、最短、观众最多的竞技生涯。
抗AI 74 🔥The trade that made every other trade's tools — from Hittite iron and guild forges to the artist-blacksmiths who brought the anvil back in the twenty-first century.
抗AI 82 🎞️决定画面之间发生什么的艺术家——从夏尔-埃米尔·雷诺在巴黎手绘投影，到用免费软件制作的拉脱维亚奥斯卡获奖影片。
抗AI 56在每晚的高压下统领整个厨房班组的专业人士,从埃斯科菲耶的巴黎,到大阪的寿司吧台,再到利马的酸橘汁腌鱼小馆。
抗AI 78 🪚用木材建造的人——一门比智人这个物种本身还要古老的手艺,从奈良的古老寺院到挪威的木结构高塔,至今仍在量两次、锯一次。
抗AI 88 ⌚打造并复活腕上那台微型机器的工匠——一门熬过了石英危机、如今却人手比订单还紧俏的手艺。
抗AI 80 🧵The craftsman who cuts cloth to one particular human body — a trade the sewing machine industrialized, ready-to-wear shrank, and no machine has finished replacing.
抗AI 78 🔥The trade that made every other trade's tools — from Hittite iron and guild forges to the artist-blacksmiths who brought the anvil back in the twenty-first century.
抗AI 82