Johnny Chen’s software career begins with a disappearing act. In high school, he enrolled in AP Computer Science, failed to get comfortably past installing the editor, and left within three days. At Purdue, the sequel lasted longer but ended much the same way: a 38 percent score on an introductory computer-science midterm, followed by another exit. The future engineer had tried code twice and code had politely shown him the door.
What brought him back was not romance with syntax. It was annoyance. Chen was studying finance and running an algorithmic-trading startup with students from computer science, engineering, bioinformatics and physics. They could build the machinery. He could lead the company, but he could not contribute to its codebase. The dependency bothered him. He wanted the founder’s most practical privilege: the ability to turn an idea into something that worked without waiting for someone else.
He began teaching himself Python and JavaScript. The work was clumsy, then less clumsy. After graduating a year early in 2016, he applied to Hack Reactor. He failed the technical interview. A coach named Keeley Nakamoto gave him assignments and walked him through JavaScript fundamentals. He tried again and failed again, this time forgetting a small but fatal .length in a loop. His note afterward was candid, almost ceremonially so: he had blown it, he wrote, but “the goal lies within the journey itself.”
Nakamoto replied that setbacks happen and offered to keep coaching him. Chen took three days, then returned with a question that now reads like a compact founder manifesto: what were a few more months in exchange for another chance? On the third attempt, he solved the final interview problem with roughly 15 seconds left.
“You got to have grit.”Johnny Chen, on learning to code
The useful kind of zigzag
Graduation from bootcamp did not summon a straight career line. Chen went to Chicago to build a food-cooperative startup with an old college colleague and roommate. The premise was bulk buying for Ohio State students. Chen built the application himself, including pickup scheduling, delivery, QR-code scanning, item voting and Stripe payments. The product moved some merchandise but could not find scale or strong margins. He could finally build. Now he had to learn everything around the build.
Back in the Bay Area, he faced the market’s next lesson: more than 140 interviews before an offer arrived. Ford Motor Company hired him into its advanced R&D lab in Palo Alto, where he became one of two React Native developers on a ridesharing service for King County, Washington. The role sent him between IDEO designers in Palo Alto, Ford engineers in Detroit and the service itself in Seattle. For someone who had once been defeated by an editor installation, shipping on both iOS and Android was a pleasingly specific rebuttal.
At the same time, he started Skillzilla, a recruiting platform for new college and bootcamp graduates. After a year at Ford, he left to work on it full time, moved to Cleveland to join his co-founders, and managed a distributed group of designers and engineers. He was writing code reviews, handling projects and learning fundraising. Skillzilla closed a pre-seed round with FuturePlay, a South Korean AI fund.
Then the pandemic rearranged the market. The company pivoted into Salezilla, applying automation to sales prospecting and personalized cold email. The name was cheerfully literal, but the operating idea had staying power: let people spend less time on the repetitive work they dislike. Chen served as co-founder and chief operating officer. The product, built with GPT-J before ChatGPT’s debut, reached $1 million in annual recurring revenue, according to the biography his next company published through Y Combinator.
Skillzilla helped people find work. Salezilla automated prospecting work. Toothy AI moved the same instinct into dental administration: software takes the repetitive shift.
Follow the frustration
Toothy AI began in 2024 with Chen as CEO, Matt Kerrigan as CTO and Tejas Konduru as CRO. The founders knew one another through friendships and university connections around the South Bay. They also brought complementary scars: engineering, machine learning, operations, finance, marketing and previous companies. By the Winter 2025 Y Combinator batch, they had narrowed their attention to the back office of a dental practice.
The choice was less obvious than it looks in retrospect. Chen has described the team struggling over what its first agents should do. The breakthrough was to stop ranking opportunities by technical neatness and listen for emotional volume. Insurance work did not merely take time. It generated denials, back-and-forth, uncertainty and delayed revenue. The pain had a pulse.
Toothy’s first agents were designed around that chain of work: verify coverage by calling payers and checking portals, enter benefit details into practice-management software, assemble and submit claims, correct denials, draft appeals, and post payments. It is not a glamorous sequence. Glamour rarely spends hours on hold.
The company says an average clinic spends more than 160 hours each month on insurance tasks. Its launch materials say the agents were already working across dozens of practices, and the current Y Combinator directory lists a team of 15. Those are company-reported measures, best read as a description of the wager: if the workflow is common, measurable and miserable, automation does not need a philosophical sales pitch.
Pricing is part of the product
Chen’s account of Toothy’s YC sprint is useful because it includes the unpolished bits. Early pricing was hourly, a sensible echo of the labor being replaced. Customers preferred predictability. On advice from YC partners, the company raised its price and offered unlimited usage. Chen said conversions improved.
There is a neat product lesson inside that adjustment. AI-agent companies like to sell capability, but buyers still budget in ordinary human ways. A clinic owner does not want a small meter running every time software checks eligibility or follows a claim. The automation becomes easier to adopt when its cost feels boring. In business software, boring can be a form of kindness.
It also changes how the software is used. A metered tool invites a tiny debate before every task: is this claim worth another charge? A flat plan removes that hesitation. Usage can spread from the obvious cases to the annoying edge cases, which is precisely where an operations product proves whether it belongs. Predictability, in this sense, is not only a finance feature. It is permission to build a habit.
The same instinct appeared in another change. The team began offering a claim guarantee after building confidence in its billing agent. Whether the underlying work is an insurance call or a resubmission, the customer is not shopping for an impressive model. The customer is shopping for fewer loose ends. Chen’s framing moved the conversation from what the AI can do toward what the practice can stop worrying about.
The customer does not buy the agent. The customer buys the afternoon it gives back.
Stubbornness, edited by feedback
Chen’s public record offers plenty of easy material for a grit parable. Two dropped classes. Two failed bootcamp interviews. More than 140 job interviews. A startup that could not scale. A pandemic pivot. Yet the more interesting trait is not his tolerance for rejection. It is his willingness to let rejection edit the plan.
After the first Hack Reactor miss, he accepted coaching. After the second, he strengthened his fundamentals. The food cooperative taught him that a functioning application was not the same as a functioning business. Skillzilla became Salezilla when the market changed. At Toothy, customer frustration chose the first workflow, and customer preference helped choose the price. He keeps the ambition and bargains freely with the implementation.
That habit also explains his unusual operating range. Chen is a finance graduate who learned full-stack engineering, a software engineer who managed designers, an operator who learned B2B marketing, and a founder who has sold automation in two different industries. The labels stack untidily. The usefulness stacks rather well.
His oldest stated ambition was simply to build the product himself. Toothy’s is larger: Chen has written that the company hopes eventually to build agents for medical practices beyond dentistry. The route there will demand more than endurance. Administrative workflows are fussy, integrations break, and trust is earned one completed task at a time. Fortunately, Chen has spent a career discovering that the glamorous door is usually locked, while the service entrance is open if you are willing to learn how the building works.
He once left computer science because he could not get comfortable with the tools. Now he leads a company whose product is supposed to disappear into somebody else’s tools and quietly finish the job. Careers have a sense of humor. The good ones also have revision history.