There is a particular sort of person who notices the thing a system throws away. Henry Quillin appears to be one of them. An unfinished music project, abandoned on a producer's drive. A conversation with an AI coding agent, erased after the code is committed. A competitor signal, hiding in a job posting while a government contractor tries to settle on a bid. These are not glamorous objects. They are leftovers, context, the material around the official result. Quillin has spent much of his still-early career turning that material back into something useful.
He is now a co-founder of Revere, a Y Combinator Fall 2025 company building pricing and competitive-analysis agents for government contractors. Revere assembles evidence behind a pursuit, identifies likely bidders, models how they may compete, and keeps the price-to-win view alive as the facts change. The neatness of the final number is not the point. The trail behind it is.
That preference did not arrive fully formed in a San Francisco startup. Quillin's public history looks less like a single grand plan than a succession of workshops. In fourth grade, while friends played video games, he tried making them. Before leaving high school, a shadowing stint at Jackson & Ryan Architects had become paid work. He took other internships and software contracts, earned Eagle Scout, and kept several personal projects in motion. The boy at the keyboard was learning an adult truth early: curiosity becomes much more educational once somebody depends on the result.
An apprenticeship with deadlines
Quillin spoke about that education twice on the Azure & DevOps Podcast, first as a high-school senior in 2022 and again as a University of Texas at Austin student in 2024. The two appearances make a small time-lapse. The first catches a young developer preparing for the profession, interested in software, entrepreneurship, and blockchain. The second finds him inside UT Austin's Canfield Computer Science and Business Honors program, talking about data structures, discrete mathematics, operating systems, internships, and the unexpected value of college.
The contrast matters. Quillin had once joked that he was supposedly smart enough to skip university and “still make bank,” immediately adding that his mother did not agree. He chose Austin anyway, studying computer science beside business in a program designed to make each discipline irritate the comfortable assumptions of the other. In 2024, he and fellow student Akshat Shah received the Parker Skiba Memorial Endowed Presidential Scholarship for the 2024-2025 year.
By then, he was also a software engineering intern at Bank of America and the CTO of Scraps Audio. His schedule had acquired the crowded quality that older people call ambitious and students call Tuesday. The work made his argument about internships tangible. A course can teach operating systems. A working team teaches what happens when another person's deadline meets your abstraction.
The business hiding in the scraps
Scraps Audio began with a neglected asset. Producers create far more music than listeners ever hear. Project files, samples, templates, and nearly finished tracks accumulate because the usual music business has no graceful place for them. Scraps treated the reject pile as inventory. Artists could list production resources and unreleased work; buyers could study a project, remix within its license, or acquire publishing rights where offered.
Quillin served as CTO alongside CEO Alejo Orvañanos and operations and finance lead Dhruv Gomber. The team described itself plainly as three college students with backgrounds in music production, entrepreneurship, finance, web development, and DJing. Their pitch was more charming: “Turn Trash into Scraps.” A later marketplace sale listing described hundreds of sellers, hundreds of products, a large email list, and organic traction without paid advertising. Whatever its ultimate scale, the underlying product judgment was crisp. A discarded file can still contain a lesson, a component, or a livelihood.
Recover the useful thing beside the official output
Before the code, review the conversation
The next version of this instinct appeared in Quillin's own coding routine. He was using Claude Code, running Codex beside it, and copying each response from one model into the other for a second opinion. The procedure caught weak plans and bugs. It also broke his concentration with the clerical dignity of moving text between windows.
He automated it. Sage ran in a separate terminal, followed the coding agent's conversation, and used a persistent review thread to critique responses and plans. The developer received a verdict, alternatives, and a message that could be pasted back to the main agent. Crucially, the reviewer did not take over. Its feedback could be accepted or ignored. Quillin wanted “a tool that was automatic, persistent, and separate from my main agent.”
“Code gets committed, but the reasoning and intent is lost.”Henry Quillin, Tig demo, October 2025
Tig supplied the broader memory. Its demo showed a web app that collected a team's AI chats, linked interactions to commits and pull requests, and offered a command resembling git blame for intent. Instead of merely identifying who changed a line, “tig blame” could return the AI conversation that created it. A bland commit message might say normalization fix. The conversation could explain that dots became hyphens because of the way a particular tool created directories. The difference between those explanations is the difference between a label and a reason.
Sage became an open-source project, written in TypeScript with a terminal interface and integrations for coding agents. Quillin published a short explanation with a headline that carries his product style neatly: “Got tired of copy-pasting Claude's responses into other models, so I built an automatic cross-checker for AI agents.” Irritation had become software. The repository drew more than 100 public stars by September 2026, a modest but concrete sign that other developers recognized the itch.
The ratio above is the project's approximation, not a law of engineering. Its argument is sound without decimal places: if most of the consequential discussion happens before the diff exists, a review tool that sees only the diff arrives late to the party and compliments the furniture.
A pivot toward consequential arithmetic
Quillin entered Y Combinator with Drew Taylor, Matthew Fan, and Shreyas Chennamaraja. He announced the news in October 2025 with four names and a promise to say more. The initial idea involved putting prompts into version control. He later wrote, with useful directness, that the team had pivoted away.
The destination was Revere. Its current field is government contracting, where a price-to-win analysis tries to estimate how a bid will fare against a moving competitive field. It is consequential arithmetic. A contractor has labor assumptions, technical strengths, indirect costs, contract requirements, and a view of likely rivals. The customer has preferences visible through prior awards. Competitors hire people, win adjacent work, and reveal intent through public traces. A solicitation's questions and answers can suddenly add relocation or lodging costs. The winning number is not waiting serenely in a cell. It shifts when the evidence shifts.
Revere's product organizes that motion. It can revise a bidder model when a vehicle holder posts award-contingent jobs, tighten a pricing guardrail when recent awards suggest an incumbent has lower indirect costs, or adjust a team's position when an adjacent task order changes a rival's past-performance score. Each example has the shape of a detective story made employable: clue, consequence, revised assumption.
This is also where the pivot from Tig becomes less abrupt. Both products distrust an unexplained output. A code diff without its conversation can be technically correct and organizationally mysterious. A modeled probability of winning without its evidence can be numerically tidy and strategically useless. Revere's value depends on making the trail available for disagreement. People can challenge an assumption because they can see the fact that moved it.
Several projects going
The most revealing description of Quillin may be an old one: he always had several software projects going. It sounds like biography written by a browser with too many tabs. Yet the projects have not been random. Games taught him to make the entertainment rather than merely consume it. Internships turned abstract skill into obligations to a team. Scraps gave neglected creative work a marketplace. Sage made an AI agent's thought process reviewable. Revere gives a changing field of competitive evidence somewhere to accumulate.
There are personal details around the machinery. He has listed weightlifting, cooking, reading, and podcasts among his interests. His LinkedIn About section is simply the hook-'em-horns hand sign, a compressed alumni essay. Engineers who reported to him at Scraps described a patient, approachable technical leader who handled onboarding carefully and made complicated processes understandable. That testimony adds a useful counterweight to the velocity. Moving quickly is common. Helping another person keep up is a separate skill.
Quillin's public career is too young for a final moral, which is fortunate because final morals are usually where profiles become embalming. The more interesting observation is provisional. He keeps moving toward work where context changes the decision, and where losing that context becomes expensive. The music file beside the single. The reasoning beside the code. The evidence beside the bid.
Many builders chase the visible output. Quillin keeps checking the floor around it. That is where the scraps are, and sometimes the business too.