The pitch sounds almost too tidy for the messy world of logistics: unbox a robotic arm, show it your picking workflow once, and by tomorrow it's packing orders alongside your crew. No redesigning the floor plan. No consultants billing by the hour for half a year.
InLoop Robotics—a small team in the single digits, Y Combinator pedigree, and a confidence that borders on audacious—insists it's already doing exactly that in customer warehouses. The San Francisco-area startup unveiled its Loop V1 system this past May, and if the company's claims hold, it could represent a genuine departure from how warehouse automation has worked for decades. Or at least, how it's supposed to work.
The core idea isn't full autonomy. It's something closer to intelligent humility. When InLoop's bimanual picking arms encounter an item they haven't seen before—a weirdly shaped box, an unfamiliar product SKU, something that doesn't match the training data—they don't guess. They stop. And they phone a friend.
That friend is a remote human operator, somewhere on the other end of a teleoperations link, who takes manual control, resolves the problem, and moves on. The robot watches, logs the intervention, and files it away as fresh training data. Next time it sees that oddball item, maybe it won't need help. The system learns by exception, not exhaustive pre-programming.
"We're replacing $100,000-plus integration projects per station," InLoop wrote in its YC launch materials—a claim the company stands by, though without independent verification—citing early deployments clocking more than 300 picks an hour across hundreds of SKUs. No client names yet. No independent benchmarks published, either—though perhaps that's to be expected from a team that only announced funding in April.
Learning on the Job
Traditional warehouse robotics has always demanded a kind of upfront certainty that real-world fulfillment centers rarely offer. You map the facility. You catalog every product. You program pick points and grasp strategies and failure modes. Months of integration work, often six figures per robotic station, and even then the system struggles the moment you introduce a new product line or change your packaging supplier.
InLoop's bet is that the old model is backward. Why pour resources into predicting every edge case when you can just handle them as they arise? The company calls it "confidence-aware AI"—a term that sounds like marketing speak until you see what it actually means in practice. The robot maintains a running estimate of its own certainty. When confidence drops below a threshold, it escalates to a human. That human resolves the issue via teleoperation, with a safety module vetting every command before the arm moves. The logged intervention becomes part of the training set, and the system's confidence grows incrementally, task by task, shift by shift.
The business model follows the same stripped-down logic: robotics-as-a-service, flat monthly fee, no upfront capital expenditure. InLoop describes it as "24/7 teleassistance fallback," though the company hasn't published pricing. They do claim the math works out cheaper than human labor, which in today's tight fulfillment labor market is a low bar for many operators.
What InLoop has disclosed: workflows ranging from kitting multi-item orders to box assembly, visual inspection, and product wrapping. The framing is "AI-native fulfillment," which sounds buzzwordy but captures something real—the notion that the system is designed to learn continuously, not arrive fully formed.
Show, Don't Tell
At MODEX, the sprawling supply chain trade show held earlier this year, InLoop set up a demonstration that cut to the heart of its value proposition. According to a May blog post on Founderland and social media updates from CTO Stepan Feduniak, the team trained the robot overnight using roughly an hour of demonstration data. The setup relied solely on wrist-mounted camera vision—no overhead sensors, no pre-mapped environment, at least according to the company's account. Feduniak later mentioned "over 400 conversations" at the booth, which suggests the pitch resonated, or at least intrigued.
Still, a trade show demo and live deployments in undisclosed customer facilities aren't quite the same thing as peer-reviewed performance data. InLoop says its robots are operating in warehouses today, pulling real orders, hitting that 300-picks-per-hour benchmark under actual conditions. The 24-hour deployment timeline appears in company materials and founder statements, though it's unclear whether that's best-case or typical, or what "deployment" actually entails in practice. Does the clock start when the robot arrives, or when the integration team shows up? Details matter in logistics.
Independent verification of these performance claims has yet to surface, which isn't necessarily damning for a company this young—but it does leave some questions hanging.
Who's Behind It
CEO Zakariea Sharfeddine comes out of AI and manufacturing R&D at Bosch and BMW—industrial heavyweights where pilot projects move slowly and reliability is non-negotiable. Feduniak, the CTO, has a robotics pedigree from KIT and TU Munich, the kind of academic-technical background that tends to show up in European deep-tech startups. Chief Product Officer and COO Pasha Rizali co-founded student robotics networks including RoboTUM and spent time working with Boston Dynamics' Spot platform, which means exposure to one of the few robots that's managed to jump from lab curiosity to deployed tool.
Founding engineer Martin Mohammed previously started a company of his own and cycled through roles at Reply and IBM—another signal that this isn't a team of fresh graduates building their first prototype.
Funding-wise, InLoop raised $500,000 from Y Combinator as part of the Spring 2026 batch, announced in April via founder posts on LinkedIn—structured as the standard $125,000 initial investment for 7% equity plus additional SAFE notes. Dealroom records a March seed entry of $125,000 from YC, likely the standard 7% SAFE note that's part of the accelerator's deal structure. The company is also listed as backed by NVIDIA's Inception program, which offers technical support but not necessarily capital.
It's a lean round by Valley standards, but then again, the pitch is that you don't need tens of millions to deploy these systems. If that's true, capital efficiency becomes a feature, not a bug.
The Field Is Crowded
InLoop isn't stepping into open territory. Warehouse automation is having a moment—again—and this time the players are bigger, better funded, and increasingly focused on end-to-end solutions rather than point products.
Locus Robotics launched its Locus Array system in April, pairing a mobile base with a picking arm designed to handle entire aisle operations. The company announced early access with DHL Supply Chain and framed the product as "Robots-to-Goods," aiming for 90% manual labor reduction. That's not a competitor trying to sell you one workstation. That's a competitor trying to sell you the whole floor.
Amazon, which operates more warehouse robotics than perhaps any company on Earth, continues pouring resources into fulfillment AI. February updates highlighted foundation models for fleet orchestration and pack operations—the kind of research that doesn't show up in press releases unless it's already yielding results. Dexterity announced advances in its world model for truck loading in March. Berkshire Grey launched a trailer unloading robot the same month.
The question InLoop is effectively asking: can a lean team with human-in-the-loop learning outmaneuver traditional system integrators and outpace well-funded competitors building fully autonomous platforms? The 24-hour deployment claim is a direct challenge to the six-to-twelve-month integration cycles that have historically kept warehouse robotics confined to Fortune 500 operators with capital budgets to match.
Maybe the teleop fallback is a competitive advantage—speed to deployment, continuous learning, lower upfront cost. Or maybe it's a crutch that won't scale, a stopgap until the AI gets good enough to work unsupervised. InLoop is betting on the former. The market will test the latter.
What Happens Next
For now, the company claims to have live systems in customer facilities, YC backing, and a product narrative built on speed rather than promises of future autonomy. That's something. In an industry littered with vaporware and perpetually delayed pilot programs, having robots actually picking orders—even if a human occasionally has to take the wheel—counts for more than a slick demo reel.
The early pilots will reveal whether teleop-assisted continuous learning can deliver reliable throughput at a price point mid-market fulfillment centers will actually pay. Whether warehouse operators trust a system that openly admits when it's confused. Whether the learning curve flattens fast enough that human interventions become rare exceptions rather than constant interruptions.
InLoop hasn't published client names, hasn't released audited performance data, and hasn't explained exactly how the economics pencil out compared to hiring another shift of packers. Those disclosures may come. Or they may not, if the company decides stealth beats transparency in a competitive market.
What's clear is that the founders have built something that's at least plausible enough to get deployed, funded, and demonstrated at a major trade show. In robotics, that's often half the battle. The other half is making it work when no one's watching.
