InLoop Robotics rents out robotic arms by the month and points them at the messiest job in logistics: picking, packing, and kitting real orders - then learning from every mistake.
There is an unglamorous truth at the center of modern shopping. Before a package lands on a doorstep, a human being stood at a table in a warehouse and picked one item off a shelf, folded a box, taped it, and slid it onto a belt. Multiply that by a few billion orders a year. Now consider that the average warehouse worker doing this quits roughly once every year, and during the holiday crush, an entire floor can turn over. InLoop Robotics looked at that churn and asked a blunt question: what if the worker was a robotic arm you could rent by the month?
That is the pitch, and it is deliberately not the pitch you have heard before. The founders - Zakariea Sharfeddine, Stepan Feduniak, and Pasha Rizali - are not selling a million-dollar automation cell that takes a year to install and locks you to one product catalog. They are renting out robotic labor. You pick a task. They deploy a trained robotic arm within weeks. You pay a flat monthly fee, and if it breaks, that is their problem, not yours. Their own line for it is short: "Why buy a robot when you can hire one?"
The distinction matters more than it sounds. A robot you buy is a bet you place once, at the top, on a machine whose capabilities are fixed the day it ships. A robot you hire is a relationship that can be turned up during peak season, turned down after the holidays, and cancelled if it does not earn its keep. It moves warehouse automation out of the category of a capital decision - approved by a board, depreciated over years - and into the category of an operating expense, decided by the person who actually runs the floor and lives with the turnover.
InLoop's product is a bimanual robot system it calls Loop V1 - two arms, a camera, and a stack of AI models trained to do the fiddly work of fulfillment. That means picking items off a shelf and into an order, assembling and folding boxes, kitting several items into one shipment, running visual quality inspection, wrapping and labeling, and unpacking incoming goods. The company reports its arms run past 300 picks an hour and, importantly, generalize across hundreds of different products from a single policy rather than being hand-tuned for one SKU at a time.
That last number is the one warehouse operators feel first. Traditional automation is a capital project - somewhere between $200,000 and well over a million dollars per line, plus a specialist integration effort that can run past $100,000 for a single station. InLoop's answer is to charge nothing upfront and bill monthly for the hardware, the software updates, and around-the-clock remote support. It is robotics priced like a software seat, and that reframing is most of the story.
Generalizing across hundreds of products is the quieter technical claim, and it is the one that has historically broken older systems. A conventional pick cell is tuned for a known set of items - the same box, the same bottle, the same blister pack - and the moment a warehouse adds a new product line, someone has to re-teach the machine or call the integrator back. InLoop is betting on a single learned policy that treats a new SKU as a variation rather than a rebuild. If that holds at scale, it removes the hidden tax that made automation so brittle for anyone whose catalog changes with the seasons.
Most robotics companies are waiting. They want the model to be good enough before it touches a customer's floor, because a robot that fumbles a fragile item in front of a paying client is a bad day. InLoop made the opposite call, and it is the most interesting thing about them. They ship imperfect policies into real warehouses now. When the robot is confident, it works. When it hits something it cannot handle, a Safety Module pauses it and pings a remote human operator who takes control over teleoperation and finishes the task.
Here is the part that turns a liability into a flywheel. Every one of those human interventions is captured as training data. The correction the operator makes teaches the model. Over time, the robot needs the human less often, on more tasks, across more products. The company's own framing is unadorned: every failure becomes training data, every intervention makes the system better. It is the same loop that trained self-driving programs, pointed at a table instead of a highway.
The team reads like three people who got impatient with the distance between research and reality. Sharfeddine, the CEO, spent his machine-learning career inside Bosch and BMW before turning to embodied AI. Feduniak, the CTO, did robot-learning research at 18 and comes out of an olympiad-mathematics background, working on the vision-language-action models that let one policy handle many tasks. Rizali helped lead Europe's robotics community through groups like RoboTUM and ESRA before the founders relocated to San Francisco. They could have written another paper. Instead they put robots on real floors and started charging for it.
That instinct shows up in the traction. InLoop says it has paid pilots live in customer warehouses today - not a demo reel shot in a clean lab cell, but arms packing real boxes for money. For a company this young, backed by a $130,000 seed round, membership in Y Combinator's 2026 batch, and a spot in Nvidia's Inception program, being in production at all is the point they keep making. A pilot is a small thing, but it is a very different small thing than a slide. It means a real operator agreed to put a real robot next to real people and real inventory, and that the robot was allowed to fail there, safely, on purpose.
It is worth sitting with how contrarian that is inside robotics. The instinct of the field is to hide the robot until it is impressive. InLoop's instinct is to expose it while it is still awkward, because the awkwardness is the data. The company's name is not decoration - the loop is the whole thesis. A robot that never fails learns nothing new; a robot that fails, gets caught, and remembers becomes the only kind that improves without an engineer standing over it.
The subscription covers all of it: the hardware sitting on the floor, continuous model updates that arrive without a truck roll, and 24/7 teleassistance so a human is always reachable when the robot is not sure. The company's phrasing - "Zero downtime. Infinite learning." - is notable mostly for what it does not claim. There is no promise that robotics is solved, no talk of general intelligence for the loading dock. It promises a machine that keeps running and keeps getting a little better. In a field prone to overclaiming, restraint reads as a feature.
For the operator on the receiving end, the day-to-day is meant to feel less like commissioning a factory line and more like onboarding a new hire who happens to never call in sick. There is no capital request to shepherd through finance, no year-long install to project-manage, no dependency on a single integrator's calendar. The task is chosen, the arm arrives, and within weeks it is doing the work that a rotating cast of temporary staff used to do badly during the busiest weeks of the year. The robot does not know it is the holiday season, and it does not care.
Warehouse automation is a crowded room. Vecna, Locus, Covariant, Ambi, RightHand, and Dexterity all want a piece of the pick-and-pack problem, and several are years and many millions ahead. InLoop's wedge is not a claim to a smarter arm. It is the combination of two things the incumbents mostly do not pair: a rental model that removes the upfront check entirely, and a teleoperation-plus-learning loop that lets it deploy before the model is finished and improve while it earns. The first lowers the barrier to saying yes. The second lowers the cost of being early.
The customer, for now, is the mid-to-large fulfillment operation - the warehouses, distribution centers, and factories that live and die by throughput and cannot keep enough hands on the floor. These are the businesses that feel the labor problem as a monthly emergency rather than an abstraction, and they are exactly the ones for whom a six-figure automation contract has always been out of reach. By meeting them with a subscription instead of a sales quote, InLoop is aiming at a far larger slice of the market than the handful of giants who can afford to build custom cells.
Whether that holds as pilots turn into fleets is the open question - remote operators are a real cost, and confidence-aware handoff has to actually get cheaper per order for the economics to compound. But the shape of the bet is coherent. Warehouses need labor they cannot reliably hire. InLoop offers labor that does not quit, priced under the wage it replaces, that gets faster with every box it packs. For an operator staring at a 49% turnover rate, that is a proposition worth a pilot.