He built an open-source tool AI coding assistants couldn’t ignore — TFN

Safi Shamsi wrote Graphify alone over a weekend and launched it free of charge. Right this moment it helps energy Cursor, Claude Code and a rising ecosystem of AI coding instruments, incomes its solo founder a spot in Y Combinator—and a shot at turning open-source infrastructure into an organization.

Builders pull Safi Shamsi’s free software program onto their machines about 65,000 instances a day. On its busiest day, 79,000. Since April the operating whole has handed 4 million. He wrote the primary model over a single weekend, alone, and gave it away.

The software program known as Graphify, and it does one slender factor that occurs to sit down beneath a complete business. It reads a codebase and attracts a map of how each piece connects, then palms that map to no matter AI coding assistant a developer already makes use of. Claude Code, Cursor, Codex, Gemini CLI and greater than fifteen different instruments can name it. Every of these assistants is racing to know massive, messy codebases, and each, left alone, works out the form of a venture from scratch each time somebody asks it a query. Graphify does the work as soon as and lets all of them learn the end result.

Shamsi took Graphify Labs into Y Combinator’s Summer season 2026 batch, the accelerator behind Airbnb and Stripe. He went in with out a cofounder, the setup the agency has spent years telling candidates to keep away from, and the companions pressed him on it. “They pushed me on whether or not I may construct all of it myself,” he says. “That was the actual query within the room, whether or not one particular person may ship it.” Jared Friedman, a gaggle accomplice at Y Combinator, had already settled the purpose in his personal thoughts. “Once I interviewed Safi, I used to be struck by how a lot he had achieved as a one-person workforce,” Friedman says. “There have been loads of predictions that people who find themselves elite customers of AI brokers could possibly be 100x or 1000x engineers, and even of once we’ll see the primary one-person unicorn firm. Safi embodies that ethos.”

He’s now not a one-person workforce. Graphify Labs has workers now, and the wager YC made is that the founder who shipped the factor by himself can run the group scaling it, which is a distinct job and a tougher one.

The objection each investor ultimately reaches is {that a} large like Cursor may clone this in every week. Shamsi doesn’t flinch at it. “Even when Cursor builds it, they’ll’t replicate the analysis I’ve already finished,” he says. “I’ve conviction in my very own analysis.” The deeper protection is structural. Each assistant wants a map of the code, and none needs to rely on a rival for it, which leaves the impartial floor, free and open, because the one place the incumbents can not take.

Gunjan Sinha, who offered the search engine WhoWhere to Lycos in 1998 and now helps run Challenge NANDA, an MIT-spin off effort to construct an “web of AI brokers,” has watched that place repay earlier than. “The businesses that final develop into the layer everybody else routes via,” he says. “Proudly owning the impartial floor beats proudly owning any single product on high of it.”

Shamsi didn’t arrive by the standard route. He wrote his first line of code at eleven, in a small city close to Delhi, performed aggressive chess as an adolescent, and got here to style via his household’s clothes enterprise, sketching embroidery onto the clothes they made. The enterprise aspect, he says, is a household bloodline. The detour taught him the road he now repeats: “The hole between understanding a expertise and deploying it on the earth is the place most studying truly occurs.” The craft was the graph. The distribution was constructing it within the open from the primary commit, so anybody may attempt it the day it existed.

Graphify stopped trying like a viral hit and began trying like infrastructure when different firms started constructing on it. Rootly, an incident-response firm, shipped a plugin in April that factors Graphify at a workforce’s personal outage historical past and turns it right into a graph its AI can question mid-incident. Greater than 130 exterior builders have despatched code. The venture has pushed greater than 150 releases since April, at instances multiple a day. Software program that different companies wire into their merchandise, and that strangers preserve free of charge, is difficult to jot down off as a second.

A few of the consideration invitations a good problem. Graphify carries greater than 100,000 stars on GitHub, the platform’s model of a like, and stars have develop into a debased foreign money. According to Friedman, Graphify is the primary YC firm to realize this milestone throughout this system. Researchers at Carnegie Mellon, North Carolina State and the safety agency Socket counted some six million suspected pretend ones throughout greater than 18,000 repositories. Shamsi doesn’t lean on them. “Stars are the least fascinating quantity I’ve,” he says, pointing as an alternative to the figures which are tougher to pretend: greater than 10,000 forks, a number-three world trending spot on GitHub in July, over one million mentions of the key phrase on X in a month, and the downloads. Gabriel Jarrosson, a former solo founder who now runs Lobster Capital, a Silicon Valley fund with greater than 100 Y Combinator bets, flagged that final one early. “Fifty-two thousand downloads in a single day on an open-source venture is an actual sign,” he says. “Distribution is the toughest half, and also you’re clearly previous it.”

The free software is the wedge, not the enterprise. Shamsi is constructing the second act, a Rust engine known as Penpax geared toward enterprises, with greater than 4,500 names on the waitlist and, he says, Fortune 500 groups amongst them. The guess beneath each is that AI now writes sufficient manufacturing software program that the tooling round it turns into a market of its personal. Not everybody buys it. Anthropic’s Dario Amodei predicted in early 2025 that AI would write 90 % of code inside months, a forecast critics together with Redwood Analysis’s Ryan Greenblatt known as overstated even inside Anthropic, and a 2025 METR examine discovered AI instruments typically slowed skilled engineers down. Shamsi doesn’t want the 90 %. At in the present day’s share, he says, the groups emailing him already really feel the pressure.

What he has is a graph, a obtain counter that retains climbing and, now, a payroll. Whether or not a founder who shipped alone can flip free software program right into a sturdy enterprise is the query YC pressed him on, and the one his enterprise waitlist will reply. The infrastructure is already load-bearing. The corporate is the half he nonetheless has to construct.