Six years from now, if you’re lucky enough to walk into a Seattle startup that has figured out how to make enterprise software actually feel intelligent, there’s a good chance you’ll hear about a company called Latch. It’s not a hardware company, despite the name. It’s not another productivity app trying to help you manage your calendar. It’s something more ambitious and, in its own quiet way, more radical: a company trying to teach artificial intelligence how your business really operates, not how it looks on an org chart or in a strategic plan. Latch was founded by Stefan Kalb, a familiar name in Seattle startup circles, and Jared Kofron, a University of Washington physics alum and former principal software engineer at Pioneer Square Labs. The company began life under the name Super Labs and rebranded earlier this year. Kalb has been here before. In 2009, he started Molly’s, a fresh food supply company servicing Seattle-area cafes and hospitals KNOWLEDGE BASE, which led to Shelf Engine, a machine learning company that dramatically reduced food waste for retailers like Target, Kroger and Walmart. Shelf Engine raised $60 million and was eventually acquired by Crisp in 2025. Now Kalb is back with Latch, an $8 million seed round from Seattle-based FUSE, and a vision that sounds almost utopian: liberating humanity from the work that owns us. That phrase deserves a second read. Kalb isn’t talking about eliminating all jobs. He’s talking about eliminating the parts of jobs that no one truly wants, the repetitive, hidden, invisible labor that exists only because software never learned how the company actually runs.
The core idea behind Latch is simple to explain, deceptively so. Employees record themselves doing a task, narrating their work the way they would if they were training a new hire. Latch then watches that recording, figures out what they did and why they did it that way, and transforms it into a knowledge graph of the company’s processes. That knowledge graph becomes a kind of operating manual for AI agents. Instead of your AI tools guessing how to route an invoice, handle a customer complaint, or approve a purchase order, they finally know how your business genuinely runs. In Kalb’s words, “Latch captures how work actually gets done. An employee records themselves doing a task and narrates it like they’re training a new hire. Latch turns that into a knowledge graph of the company’s processes and serves it to your AI agents. Your agents finally know how the business really runs.” This is not another tool for making videos. The video is just the input. What Latch learns from that video is the product. It’s a subtle distinction that Kalb says people often miss. Some see a screen recording and think, “Oh, it’s Loom.” But Loom is a video sitting in a folder waiting for a human to watch it. Latch actually watches the recording, understands what was done, extracts the decisions embedded in the work, and converts that into something an AI agent can act on. It’s less about capturing a task and more about capturing the invisible logic of an organization, the unwritten rules, the moments where an employee makes a call that they don’t even realize is a decision.
That hidden complexity is what Kalb is obsessed with. He describes the problem in terms of liberation. Every job has hours in it that exist only because software never learned how the company runs. Someone has to reconcile discrepancies, manually approve exceptions, re-enter the same data into three different systems, or chase down information that should have been obvious. Kalb wants that work to disappear, and he doesn’t think anyone will miss it. But before that can happen, companies need to understand what’s actually happening inside them. And that’s where Kalb discovered a strange, almost uncomfortable truth: people are bad at explaining their own jobs, and they know it. Ask someone to document their process and you’ll get a five-step list. Watch them do it and it’s forty steps with a dozen decisions they never mention, because to them it isn’t a decision. It’s just Tuesday. That little observation contains the whole thesis for Latch. Humans are not reliable narrators of their own work. We compress, simplify, and forget. We don’t realize how much tacit knowledge we carry around until we try to hand it off. Latch is designed to capture that tacit knowledge before it walks out the door, before a retirement, before a promotion, before a layoff. It’s a way for companies to document the real process, not the idealized process on the whiteboard. And once that knowledge is captured, it can power AI agents that actually reflect the business, not some generic vertical software assumption.
The personality behind Latch is worth understanding too. Kalb is a Bainbridge Island resident with a famously low-key, human approach to entrepreneurship. The Startup Spotlight questionnaire reveals a man who drinks a decaf Americano with no room, gave up caffeine years ago, and cheerfully asks you not to judge him for it. He’s listening to Mr Little Jeans, hoping she starts making music again. He’s reading Laura Ingalls Wilder to his kids, and he finds it gives him perspective on modern life. His entrepreneurial idol is Ray Dalio, the Bridgewater founder, because Dalio is someone of integrity who built something great and shared knowledge. This is a person who thinks about process, yes, but also about people, about the texture of everyday life. He’s not retreating into an abstract world of pure technology. He’s grounded in the physical world, the world of food supply chains, grocery stores, hospitals, and cafés. That background matters. Shelf Engine was not a flashy consumer app. It was a hard, unglamorous problem, reducing food waste, using machine learning, and it worked. Kalb has spent his career in the messy middle of real-world operations. He knows what it’s like to deal with loading bays and refrigerators and shelf lives, and that sensibility seems to carry directly into Latch. He’s not trying to sell you a magic wand. He’s trying to build a layer of infrastructure that sits underneath your enterprise AI and makes it useful, by teaching it what everyone else assumes you already know.
One of the most interesting things about Latch is how AI has changed Kalb’s approach to building a company itself. He says the obvious stuff is real, faster coding, smarter tools, more automation, but the more interesting change is organizational shape. The ratio of product to engineering has flipped. The ratio of sales development representatives to closers has flipped too. Building software used to be the bottleneck, so you staffed for that. You hired engineers, you built features, you tried to ship fast. Now, with AI, building is easier and cheaper than ever. The bottleneck is deciding what to build and who to sell it to. So you staff for that instead. You hire people who can think deeply about customer problems, who can have real conversations, who can understand the difference between a feature request and a fundamental pain point. That shift has major implications for how startups are built, how capital is spent, and what kinds of founders succeed. The cheapest part of the startup is no longer the idea; it’s the execution. The expensive part is the judgment. That’s a profound change, and Kalb seems to have embraced it fully. At the same time, he’s candid about how much harder AI has made the go-to-market side of things. You might think AI would make outreach easier. Instead, every inbox is full of AI-written outbound, and cheap channels are gone. Latch decided to stop competing on volume. They made the deliberate, painful choice to spend real money on fewer, deeper conversations corrected SHOWING UP in person. Kalb admits it costs far more per account than they planned for, and it’s the only thing that works. There’s something reassuringly old-school about that, in an AI era full of automated everythingcars. The way to win trust is still to show up, look someone in the eye, and listen.
The long-term vision for Latch is best captured in its own definition of success. Kalb says, “We’ll know we’ve made it when someone gets furious that we’re down. Not because they lost a file, but because they can’t do their job without us. That’s the moment we stop being a tool and become key infrastructure.” That’s a powerful image. Infrastructure is invisible until it stops working. You don’t think about the power grid or the water system until the tap runs dry. Kalb wants Latch to be the same kind of quiet, trusted foundation for enterprise AI. He wants the day to come when a company genuinely cannot function without the knowledge that Latch has capturedants. When a customer says that if you go down, we go down, that’s not a bug, that’s the whole point. And then there’s the startup’s advice for other entrepreneurs: “Look away from the obvious. If someone is telling you about the front of their store, ask them about the loading bay. The best problems are the ones nobody is talking about, usually because they can’t put words to what’s happening. That’s also where you’ll have the least competition.” That advice sums up Latch neatly. The front of the store is the polished dashboard, the user interface, the demo. The loading bay is the messy, undocumented, human labor that keeps everything running. Latch is built for the loading bay. It’s a reminder that the most valuable problems are often the ones we have learned to ignore, the ones hiding in plain sight. And if Latch succeeds, the reward isn’t just a better AI system. It’s something closer to freedom: fewer hours spent on work that only exists because software never took the time to learn how work really happens. For Kalb, that’s not just a business opportunity. It’s a mission.
