Safi Shamsi wrote Graphify alone over a weekend and released it for free. Today it helps power Cursor, Claude Code and a growing ecosystem of AI coding tools, earning its solo founder a place in Y Combinator—and a shot at turning open-source infrastructure into a company.
Developers pull Safi Shamsi’s free software onto their machines about 65,000 times a day. On its busiest day, 79,000. Since April the running total has passed 4 million. He wrote the first version over a single weekend, alone, and gave it away.
The software is called Graphify, and it does one narrow thing that happens to sit under an entire industry. It reads a codebase and draws a map of how every piece connects, then hands that map to whatever AI coding assistant a developer already uses. Claude Code, Cursor, Codex, Gemini CLI and more than fifteen other tools can call it. Each of those assistants is racing to understand big, messy codebases, and each one, left alone, works out the shape of a project from scratch every time someone asks it a question. Graphify does the work once and lets all of them read the result.
Shamsi took Graphify Labs into Y Combinator’s Summer 2026 batch, the accelerator behind Airbnb and Stripe. He went in without a cofounder, the setup the firm has spent years telling applicants to avoid, and the partners pressed him on it. “They pushed me on whether I could build all of it myself,” he says. “That was the real question in the room, whether one person could ship it.” Jared Friedman, a group partner at Y Combinator, had already settled the point in his own mind. “When I interviewed Safi, I was struck by how much he had accomplished as a one-person team,” Friedman says. “There have been a lot of predictions that people who are elite users of AI agents could be 100x or 1000x engineers, and even of when we’ll see the first one-person unicorn company. Safi embodies that ethos.”
He is no longer a one-person team. Graphify Labs has staff now, and the wager YC made is that the founder who shipped the thing by himself can run the group scaling it, which is a different job and a harder one.
The objection every investor eventually reaches is that a giant like Cursor could clone this in a week. Shamsi does not flinch at it. “Even if Cursor builds it, they can’t replicate the research I’ve already done,” he says. “I have conviction in my own research.” The deeper defense is structural. Every assistant needs a map of the code, and none wants to depend on a rival for it, which leaves the neutral ground, free and open, as the one position the incumbents cannot take.
Gunjan Sinha, who sold the search engine WhoWhere to Lycos in 1998 and now helps run Project NANDA, an MIT-spin off effort to build an “internet of AI agents,” has watched that position pay off before. “The companies that last become the layer everyone else routes through,” he says. “Owning the neutral ground beats owning any single product on top of it.”
Shamsi did not arrive by the usual route. He wrote his first line of code at eleven, in a small town near Delhi, played competitive chess as a teenager, and came to fashion through his family’s clothing business, sketching embroidery onto the garments they made. The business side, he says, is a family bloodline. The detour taught him the line he now repeats: “The gap between knowing a technology and deploying it in the world is where most learning actually happens.” The craft was the graph. The distribution was building it in the open from the first commit, so anyone could try it the day it existed.
Graphify stopped looking like a viral hit and started looking like infrastructure when other companies began building on it. Rootly, an incident-response company, shipped a plugin in April that points Graphify at a team’s own outage history and turns it into a graph its AI can query mid-incident. More than 130 outside developers have sent code. The project has pushed more than 150 releases since April, at times more than one a day. Software that other firms wire into their products, and that strangers maintain for free, is hard to write off as a moment.
Some of the attention invites a fair challenge. Graphify carries more than 100,000 stars on GitHub, the platform’s version of a like, and stars have become a debased currency. According to Friedman, Graphify is the first YC company to achieve this milestone during the program. Researchers at Carnegie Mellon, North Carolina State and the security firm Socket counted some six million suspected fake ones across more than 18,000 repositories. Shamsi does not lean on them. “Stars are the least interesting number I have,” he says, pointing instead to the figures that are harder to fake: more than 10,000 forks, a number-three global trending spot on GitHub in July, over a million mentions of the keyword on X in a month, and the downloads. Gabriel Jarrosson, a former solo founder who now runs Lobster Capital, a Silicon Valley fund with more than a hundred Y Combinator bets, flagged that last one early. “Fifty-two thousand downloads in a single day on an open-source project is a real signal,” he says. “Distribution is the hardest part, and you’re clearly past it.”
The free tool is the wedge, not the business. Shamsi is building the second act, a Rust engine called Penpax aimed at enterprises, with more than 4,500 names on the waitlist and, he says, Fortune 500 teams among them. The bet under both is that AI now writes enough production software that the tooling around it becomes a market of its own. Not everyone buys it. Anthropic’s Dario Amodei predicted in early 2025 that AI would write 90 percent of code within months, a forecast critics including Redwood Research’s Ryan Greenblatt called overstated even inside Anthropic, and a 2025 METR study found AI tools sometimes slowed experienced engineers down. Shamsi does not need the 90 percent. At today’s share, he says, the teams emailing him already feel the strain.
What he has is a graph, a download counter that keeps climbing and, now, a payroll. Whether a founder who shipped alone can turn free software into a durable business is the question YC pressed him on, and the one his enterprise waitlist will answer. The infrastructure is already load-bearing. The company is the part he still has to build.
