The Engineer Putting Robots on Payroll

Walk into most warehouses at peak season and you will find a strange kind of math. Half the floor staff from a year ago are gone. During the holiday rush, more people leave than the entire headcount. Managers spend their days hiring for jobs that people quit almost as fast as they can be trained. Zakariea Sharfeddine looked at that churn and did not see a staffing problem to solve with better recruiting. He saw a job description that a robot could take over, if only the robot were reliable enough to show up every day.

That idea became InLoop Robotics, the company he co-founded and now runs as chief executive. It went through Y Combinator's Spring 2026 batch, the one labeled P26, and it sells something that sounds almost too simple when you first hear it: robot employees you can hire by the month. No million-dollar purchase. No contractor rebuilding your facility. A robotic arm arrives, it starts packing and kitting orders, and you pay a flat fee the way you would pay any other worker.

The pitch

A robot you rent, not a machine you buy

The distinction matters more than it looks. Traditional warehouse automation asks a customer to spend somewhere between two hundred thousand and a million dollars up front, then wait months while integrators wire everything into place. The moment the product catalog changes - a new box size, a different item to pick - much of that custom rig becomes dead weight. It is a big bet on things staying the same, placed in an industry where nothing stays the same.

InLoop flips the arrangement. The company owns the risk, deploys the hardware, and keeps it working. A customer describes the repetitive task that is eating their labor budget, and InLoop points a robot at it. The billing is monthly, the upfront cost is zero, and if demand spikes, capacity can flex with it. In the founders' framing, this is less like buying a forklift and more like adding staff who never call in sick and never quit.

The product is called Loop V1, a two-armed robot built for the messy middle of fulfillment - assembling boxes, kitting orders, inspecting items, prepping content, unpacking shipments. It is offered as Robot-as-a-Service, which bundles the hardware, the ongoing AI updates, and round-the-clock remote support into one price. The promise the company keeps returning to is deployment in hours instead of months, with no redesign of the building required.

There is a confidence layer sitting on top of all of it. Loop V1 is built to know when it is unsure. A safety module watches the robot's own certainty and pauses the arm the instant something falls outside what it trusts, handing the moment to a person before a mistake can happen. That framing - a robot that raises its hand rather than plowing ahead - is what lets InLoop put arms next to real inventory without weeks of hazard planning. The company points out that none of its deployed robots have broken so far, a small claim that says a lot about how cautiously the system is tuned.

The idea in the name

Keeping a human in the loop

Most robotics demos show the moment the robot gets it right. InLoop built its company around the moment it gets it wrong. The name is the thesis. When a robot on the line hits an edge case it cannot handle - an odd item, a torn package, a grip it does not trust - it does not freeze the whole operation. A remote operator takes over the controls, guides it through, and the line keeps moving. Then that same intervention becomes training data, and the robot is a little better the next time it sees the situation.

This is a quietly radical design choice. Plenty of teams treat teleoperation as an embarrassment to be engineered away as fast as possible. Sharfeddine and his co-founders treat it as the product. Deploying imperfect robotic policies safely, keeping uptime at one hundred percent while the models improve on the job, is the hard infrastructure problem they set out to solve. It lets them start earning in real warehouses now, rather than waiting for a model good enough to run alone.

When the robot hits an edge case, a remote operator takes over control and the line never stops.The core idea behind InLoop

The strategy is already on the floor. InLoop has paid pilots running in customer warehouses, with stations handling more than three hundred picks an hour and a single policy generalizing across hundreds of different items. Each of those stations, the company says, stands in for the kind of custom integration project that used to cost six figures and weeks of a systems team's attention.

The path in

From Karlsruhe to a packing line

Sharfeddine did not arrive at robots through a garage tinkering story. He came through the front door of European engineering. He studied informatics at the Karlsruhe Institute of Technology, one of Germany's serious technical schools, finishing his degree between 2020 and 2024. Along the way he did machine learning work at Bosch and BMW, two names that carry the full weight of German industrial precision, and he later took on reinforcement learning and humanoid robotics as an engineer at RoboTUM in Munich.

It is a resume built on discipline, and it shows up in how the company talks about reliability rather than spectacle. But somewhere in that stretch of ML jobs, Sharfeddine seems to have decided that the interesting work was not tuning another model in a lab. It was getting robots to do useful, repetitive labor in the real world, where the floor is dusty and the boxes never look quite like the training set. He describes the goal plainly: building the future of embodied AI.

His co-founders come from the same current. Stepan Feduniak, the chief technology officer, was doing robot learning research at KIT and TUM at eighteen, working on vision-language-action models and reinforcement learning after a run in olympiad mathematics. Pasha Rizali rounds out the founding team. It is a young group, heavy on research pedigree, betting that the gap between a promising policy and a paying deployment is exactly where a company should live.

You can see the same instinct in the small choices around the company. The founders spend their time in warehouses rather than at conferences, asking operators to name the single most repetitive task on their floor and then pointing a robot at exactly that. It is an unglamorous way to build a robotics company, closer to a plumber's trade than a research lab, and it seems to be deliberate. The team would rather solve one narrow bottleneck completely than promise a general-purpose robot that impresses in a video and disappoints in a shift.

Why San Francisco

German engineering, American appetite

InLoop was founded in Germany and is now based in San Francisco, and that move is not incidental to how Sharfeddine thinks. On LinkedIn, after sitting down with the founders of Starcloud - a team trying to build data centers in space - he wrote that one of the best things about the United States is that it dares you to dream big, and added that Europe needs more of that energy. It is a telling line from someone who carries Europe's engineering training but chose America's ambition to build in.

One of the best things about the US is that it dares you to dream big. We need more of this energy in Europe.Zakariea Sharfeddine, on LinkedIn

The company has picked up the kind of backing that follows that ambition. Beyond Y Combinator, InLoop is supported by Nvidia Inception, the chipmaker's program for AI startups - useful company to keep when your product depends on running robotic policies quickly and cheaply. The team is still small, a handful of people, which is part of the story rather than a footnote. The claim is that a tiny group plus a fleet of remotely supervised robots can stand in for a warehouse floor's worth of labor.

What comes next

The bet on getting cheaper

The line the company likes to repeat is that its warehouses get better, faster, and cheaper with every order. That is the whole wager compressed into a sentence. Human labor gets more expensive over time and does not improve on its own. A robotic system that learns from every intervention should move the other way, sliding down a cost curve as the edge cases get rarer and the remote operators are needed less often. If that holds, the monthly fee that looks competitive today looks like a bargain in a year.

There are real questions in front of that vision. Generalizing a single policy across hundreds of SKUs is genuinely hard, and warehouses are unforgiving places to test. The human-in-the-loop model has to scale in a way where one operator can watch over many robots without becoming the bottleneck the whole system was meant to remove. And selling a subscription against entrenched automation vendors means convincing operators to trust a young team with a live production line.

But the shape of the bet is clear, and it is a coherent one. Sharfeddine is not waiting for robots to become perfect before putting them to work. He is deploying the imperfect version now, wrapping it in enough human judgment to keep it safe, and letting the failures pay for the improvements. For a founder who spent his early career inside the labs where the perfect model is always one more experiment away, that is a pointed decision about where value actually gets created - not in the demo, but on the floor, at three hundred picks an hour, order after order after order.