Marcus Lowe likes builders, but the word means something broader to him than it does in most San Francisco conversations. A builder can be a product manager with a weekend experiment. A builder can be a salon owner who understands the small frictions of a booking calendar. A builder can be a firefighter who knows exactly where emergency-response software fails in the field. The common credential is not fluency in JavaScript. It is the experience of seeing a problem clearly enough to want a tool that does not exist.
This distinction sits beneath Anything, the AI app company Lowe co-founded with Dhruv Amin. The product takes a conversation and turns it into web or mobile software, handling the work that usually appears after an impressive demo: the database, authentication, payments, integrations, hosting, deployment, and the bugs waiting behind a login screen. The ambition is not simply to help engineers type faster. Lowe wants the person with the knowledge to become the person who can ship.
His route to that idea looks orderly from a distance. MIT. The Media Lab. Google. A startup. Another startup. Up close, the interesting parts are the discontinuities: moving between product and engineering, choosing a co-founder years before needing one, and deciding that a business producing real revenue had become the wrong vehicle for the mission.
A weekly habit at the frontier
Lowe arrived at MIT in 2010 and spent his undergraduate years around the Media Lab. He has described its appeal in practical terms: it gave him the chance to ship something new, near the frontier of technology, almost every week. The cadence matters. A frontier looks mystical from the audience. From a workbench, it is a series of things that almost function, followed by the stubborn business of making one of them useful.
His coursework included artificial intelligence, algorithms, computation structures, software construction, and computer systems engineering. He also worked as a lab assistant and completed internships at Google, first in software engineering and later in the company’s Associate Product Manager track. After graduating in 2014, he joined Google full time and eventually worked as a product manager on Google Maps.
Maps supplied a different scale of education. Lowe says Google trained him to ask whether a problem mattered to a large population, whether its value could compound, and whether the result would hold under real conditions without constant human intervention. “Production-grade as a design property,” he later wrote. “Not an aspiration.” In a startup, the money and time are radically smaller. The questions survived the move.
After Google, Lowe joined Resource, a recruiting-software company later acquired by Gem. His job crossed an instructive boundary: head of product and senior software engineer. He could think in user flows and then descend into the machinery. That combination would become central to Anything, a product whose surface needs to feel as easy as chat while its internals behave like a development team.
Trust before the term sheet
Amin entered the story at Google. The two became friends, lived together after school, and were best men at each other’s weddings. Seven years after meeting, they founded a company together. Lowe’s advice to would-be founders is to find the co-founder before the need becomes urgent. Compatibility is easy to claim over coffee; it is harder to manufacture after the company has a bank account, payroll, and a problem that makes both founders wonder whether they were wrong.
Their first answer arrived in 2021. Create began as a code-generation system for common app components, paired with a marketplace of developers who could finish the last mile. The founders sold early. They reported bootstrapping the business to about $1 million in revenue before raising venture capital, then growing it to roughly $2 million. Lowe took a lesson from that period that rarely appears in launch mythology: the first paying customer can edit a roadmap more effectively than another whiteboard session.
Then ChatGPT changed the shape of the wager. The developer marketplace was working, but model capability was advancing fast enough to replace more of the manual work. By 2023, the founders faced an awkward choice. They could keep collecting revenue from a business that was only partly aligned with where they believed software was going, or they could stop, reduce the team, and build again.
They chose the second path. This was not the familiar pivot from failure. It was a deliberate retreat from comfort. The mission stayed fixed: give more people the power to create software. The mechanism changed from a marketplace augmented by code generation to an agent expected to do more of the work itself. Trust between the founders became operating infrastructure. Each knew the other was attached to the goal, not merely the version already producing revenue.
The boring parts become the product
In 2024, the rebuilt company released a text-to-app tool under the Create name. Early versions could make components and simple pages. Users promptly asked for the rest: multiple screens, live data, databases, logins, payments, team editing, and a path to publication. These requests sound like feature expansion. Together, they expose the central problem with AI coding. Generating plausible code is a small portion of shipping software that strangers can depend on.
Lowe and Amin kept building the unglamorous layers. By August 2025, they relaunched as Anything. A user could describe a product, generate both web and mobile versions, connect the backend, charge customers, and submit to an app store from one system. The company said the launch caught quickly, reaching a $2 million annualized run rate in two weeks. In September, Anything announced an $11 million Series A led by Footwork at a reported $100 million valuation.
The same month, Lowe introduced Anything Max. The agent opens an app in a browser, clicks through flows, writes tests, branches code and data, makes a change, and tries again. Its existence clarifies the company’s definition of done. A generated checkout page is only a picture until someone can complete a payment. A login is only a form until credentials persist. The agent needs eyes, hands, a computer, and enough patience to encounter its own mistakes.
At scale, even the invisible computer becomes a product decision. Anything’s usage arrives in spikes when a launch or social post travels. The company needed thousands of isolated development environments to start quickly and resume with each user’s code ready. Lowe’s team considered building that infrastructure, estimated it would cost two or three senior engineers several months, and bought the layer instead. The choice is a tidy example of the founder’s real job: know which hard problem belongs to you.
Software for the people with the story
The strongest evidence for Lowe’s thesis is not a benchmark. It is surprise. A firefighter told him about emergency-response software built for fire departments and already in use. Lowe said he would never have known to imagine it. He did not work in fire response, did not experience the weaknesses of the existing tools, and could not have designed the right replacement from San Francisco.
The firefighter could. So could a film director assembling a hard-to-find television catalog, a recruiter building a client-intelligence platform, or a local organizer making a social app for people in Philadelphia. Their advantage is not technical. It is lived specificity. Traditional software creation asks those people to translate their knowledge into requirements, recruit a team, secure a budget, and hope the meaning survives the handoffs. Anything tries to compress that chain into a conversation.
By July 2026, Anything was serving more than one million users. A later customer study described 1.5 million users and more than 800,000 apps created in five months. Those figures indicate reach, but Lowe’s more interesting unit is the unanticipated product: software too local, odd, or experience-dependent to emerge from a conventional roadmap. A broad platform succeeds by giving those narrow ideas somewhere to go.
There are limits inside the dream. Production software remains difficult because customers remain difficult, edge cases remain strange, and the world keeps changing after deployment. Lowe’s answer is not to deny the complexity. It is to move more of it under the floorboards. He talks about agents that can work on several parts of a product simultaneously, run longer, and eventually help applications improve themselves. The simpler the surface becomes, the more disciplined the machinery underneath must be.
The founder as editor
Lowe’s public founder curriculum has four parts: ship things, find a co-founder before the emergency, practice making hard decisions, and locate the first customer quickly. The list is notable for what it excludes. There is no demand for a theatrical origin story. The idea matters less than the repeated behavior of turning judgment into something another person can use.
That behavior links the chapters of his career. The Media Lab made frequent experiments normal. Google supplied a test for scale and reliability. Resource blurred product thinking with implementation. Create taught him to sell before overbuilding. The 2023 shutdown proved that decisiveness can require discarding a result that outsiders would call success. Anything now packages those lessons for a user who may never see the code.
If the platform works as Lowe hopes, the founder of a tiny software business will behave less like a programmer and more like an editor. Choose the problem. Describe the experience. Inspect what comes back. Notice where reality disagrees. Revise. The agent handles more of the production, but it cannot supply the original understanding. That belongs to the firefighter, the stylist, the coach, the recruiter, and every other person who has spent long enough inside a problem to recognize the tool that should exist.
Lowe’s bet is ultimately about who gets to act. For most of software history, expertise without engineering or capital had to wait in line. Anything is an attempt to remove the line, one plain-English request at a time. The code recedes. The judgment becomes visible. And the people who know the problem best finally get to press publish.