Three founders automated the front end. The gap was behind it.
A founder in Argentina, one in London and one running smart vending in San Antonio on what AI took, and what still needs a person.
A sale with no order behind it.
A message sent that should not have been.
A smart cooler that nobody restocks.
Three founders, three countries, one shared warning: the automated step is the easy part. What comes after it still needs an owner.
We asked which parts of a job AI has actually taken, and which still need a person. Each founder put the answer in the same place: just after the step that automates cleanly.
The cooler took the checkout. It didn't take the why.
Manuel Mojica, Founder/Owner of MM Healthy Vending (San Antonio, United States)
“I’ve seen inventory counting and checkout move to AI completely in smart coolers: cameras, sensors, remote monitoring, and automatic card charges handle the transaction better than a person standing there. What still needs a person is product judgment: why salads stopped moving, why a school needs different items than a gym, or why a machine placement is failing.”
“The expensive mistake is automating the front end and forgetting the service layer. A “smart” cooler that no one restocks or fixes quickly just becomes a very expensive empty fridge.”
Checkout is solved. Keeping the machine worth walking up to is not, and that job still has a human name on it.
The sale ran itself. The order didn't follow.
Andrei Zhizdra, Founder of Santa AI (Argentina)
“We automated the sale but not the check that the sale actually produced an order.”
“When something goes wrong, though, parents want a person to answer, so that part stays human.”
At Santa AI, every video and live call is AI. Humans keep support, refunds and the investigation when something fails, which is where the gap showed up.
A green checkmark isn’t the outcome.
Deepak Shukla, Founder of Pearl Lemon Group (London, United Kingdom)
“An agent can tell you it successfully sent a message. It cannot, on that evidence alone, tell you that the message should have been sent.”
“One expensive mistake is mistaking a functioning workflow for a successful outcome.”
A delivery receipt is not judgment. A person still decides whether the next action was sensible, and someone stays accountable when a situation falls outside the rules.
Where they split
Where AI took over: checkout and inventory counting in smart coolers, every Santa video and live call, and repeatable work like collecting information and categorizing requests.
Where a person stayed: product judgment, support, refunds, and deciding whether an action makes sense.
Where they differ: the vending founder uses both a person and an agent where trust, contracts or a customer relationship is at stake. The London founder keeps a person accountable where a situation falls outside the rules. The Santa AI founder puts one where something has gone wrong.
Where they agree: the automated step is the easy part to see, and the hard part to own is what comes after it.
Where it leaves hiring: the open seat may be less about doing the task and more about owning the check that follows it.
My take
Automation gets judged at the front door, but the work isn’t done until someone checks the back. The sale goes through, the message sends, the checkout clears. The easy thing to miss is the check that the work actually did its job. Owning that check may be the job AI creates, and it’s worth naming before a role is posted or an agent is bought.
Your turn
Which step in your business reports success while nobody checks the outcome, and who should own that check: a person, an agent, or both?
Next: agents ran a 561-site study and took over event admin, but three founders draw the line where a relationship or a client win is on the line, and one warns owners they automate the wrong half.
