
Sommy is an AI wine app, and our blog is written with AI too. We would rather tell you exactly how than let you guess. Here is the whole arrangement, including the parts that are less flattering.
A custom-built AI agent, created by Curt Tudor, does the drafting. Internally, we call it winepress-agent. It researches the topic, reads the sources, even reads the Sommy source code when a post describes the app, and then it writes the post. The winepress-agent runs on a schedule, a few times a week, and it works from a set of writing rules we maintain: plain sentences, real examples, no filler, no thesaurus words nobody says out loud.
Every post on the blog carries Curt Tudor's name. It means he reviewed it, corrected it, and stands behind it. It does not mean he wrote the draft completely from scratch, and on most posts he did not.
Yes, absolutely. The winepress-agent cannot publish anything. It writes into our CMS as a draft, and a person at Sommy has to open it, review it, make any adjustments we think are necessary, and then press publish. That is a hard rule in how the system is built, not a promise about good intentions.
It is not (yet) a perfect gate.
For example, in September 2026 four posts reached the live site before anyone had read them, because of a setting in our CMS that behaved differently than we expected. We found it, and we fixed it. We mention it here because a policy page that claims a flawless process is not worth reading.
No, we do not. And this is the line we care about most.
Writing about how a specific bottle tastes means somebody tasted it. Software did not taste it. So the blog does not publish tasting notes, scores, or vintage verdicts dressed up as personal experience. You will not find a post that says we poured something and this is what we found, because we did not.
What the winepress-agent writes instead is the practical, checkable half of wine: which grape suits which dish and why, what the words on a label mean, how two regions differ. That material does not need a palate. It needs to be correct.
We also never tell you a wine will taste a particular way to you. Taste is personal, and oh, by the way, that is the entire reason Sommy exists.
They are AI-generated, every one of them. They are illustrations, not photographs of real bottles, real restaurants, or real people. Nothing on the blog is a photograph presented as documentary evidence of anything.
Every figure is supposed to be checked against the source it cites, and links are supposed to point at something that actually says what we claim.
We have caught our own failures. The specific one worth naming: an invented statistic attached to a real, reputable report that did not contain that number. It read perfectly. That is the failure mode that matters with AI writing, and it is invisible to every "does this sound like a robot" test going around. Style tells you nothing. Opening the citation tells you everything.
So we re-check sources on older posts as we revisit them, and if you find a number on the blog that does not hold up, email support@sommy.ai and we will correct it or pull it.
The drafting is. The rules, the editing, the topic decisions, the corrections, and the decision to publish each post are ours. Some articles are completely human craft (primarily, the opinion pieces from Curt).
Because most wine questions people actually search for are questions of fact and structure, not questions of taste, and a well-directed machine answers those accurately at a scale we could not. The questions of taste are what the Sommy app is for, and we keep the two separate on purpose.
Trust the checkable parts, and check us on them. We publish our sources so you can. Do not read anything on the blog as a verdict on how a bottle tastes, because we have not made that claim.
Email support@sommy.ai with the post and the claim, and we will check the source and correct or remove it.
Sommy is the AI wine app that learns what you like and remembers every bottle you have tried. We are pre-launch. Join the waitlist and we will tell you when it is ready.