The following is a guest post from Temo Benidze, founder and CEO of Orbis AI and a 30-year hospitality vet. Opinions are the author’s own.
In May, Starbucks quietly switched off the artificial intelligence inventory system it had rolled out only nine months earlier. The lesson most people will take is that the technology wasn't ready. I think that's the wrong lesson. The technology was probably fine.
The problem is that it was one more system for already-stretched people to feed, check and manage. In restaurants, that is how most technology dies.
I've spent 30 years in hospitality, behind the bar, on the floor and in the back office at two in the morning reconciling invoices nobody else had time to read. I became the country of Georgia's first certified sommelier, ran resorts and restaurant groups and trained the next generation of operators. In all that time, I never once wished for another dashboard, I wished for time.
That gap is the thing the industry keeps missing. The National Restaurant Association says about a quarter of operators now use some form of AI. Adoption is real, but adoption is not the same as usefulness. Most of what gets sold as restaurant AI is a smarter screen: another login, another tab, another place the owner is supposed to go and look. The owner is not at a desk. The owner is expediting, calming a four-top and counting the drawer.
Software in this industry is often built, mostly, by people who have never worked a Friday closing shift. It optimizes for the things software people value, like features and dashboards and configurability, and quietly assumes the operator has the time and the calm to sit and absorb them. They don't. The most valuable thing in a restaurant at 7 p.m. on a Saturday is the operator's attention, and almost every tool on the market spends it freely.
Thirty years taught me the opposite instinct: owners don't want more software, they want less. The right role for AI is not to give the operator more to look at. It is to take things off the plate entirely, to read the invoices, catch the overcharges, draft the review replies, flag the compliance dates and then stay quiet. The win is measured in tasks removed, not screens added.
Take the least glamorous job in the building: invoices. A mid-sized kitchen sees a stack of them every week, and somewhere in that stack may be a supplier who quietly raised a price, or billed twice, or shorted a case. Catching that means sitting down with a calculator on a Tuesday morning, which is exactly when the deliveries are arriving and the prep list is long. So it doesn't get caught — week after week — and the margin bleeds a point or two no one can later explain. That is the kind of work software should simply do and then speak up only when something is actually wrong.
If that's true, the interface matters more than the intelligence. The channel that actually reaches a restaurant operator is the one already in their hand all day: their phone. Not because it's novel, but because it's where they already are. Meet people where they are, in the language they already speak, and you don't need a training program. If you ask them to learn another platform, you've already lost.
And there is one more discipline almost nobody talks about: restraint. The hardest part of building useful AI for operators is not getting it to act. It's getting it to stay silent. Most alerts should never have been sent. An assistant that pings you about everything is just a noisier version of the problem it was supposed to solve. The skill is in knowing what genuinely needs a human, and protecting the operator from everything that doesn't.
None of this is anti-technology — I build software for a living now — it's a plea for technology that respects the reality of the room. The test of restaurant AI shouldn't be how advanced it is, or how long its feature list runs. The test is simpler and harder: can the owner forget it's there, and still trust that the restaurant is handled?
Starbucks can afford to run a nine-month experiment and walk away. The independent operator down the street cannot. They don't need the most ambitious AI. They need the one that quietly gives them back the one thing no vendor can sell them: time. Build for that, and the industry might finally get technology it actually keeps.
