HackerRank's Open-Source ATS Scored One Resume 74, 88, and 90. Does Y Combinator Use AI to Screen Your F26 Application?

Today's top story on Hacker News is a developer who ran HackerRank's newly viral open-source applicant tracking system on his own resume and watched it return a different score almost every time: 90, then 74, then 88, then 83. He looped the same resume through 100 times. The scores ranged from 66 to 99. If a company set its cutoff at 85, that one resume would have been rejected 65% of the time, on pure luck.

The repo (interviewstreet/hiring-agent) spiked to nearly 3,000 GitHub stars in June 2026, and engineers are passing it around with the same uneasy question: if this is how my resume gets filtered, is the whole thing a coin flip? It's a fair worry for a job application. So if you're about to spend a weekend on your Y Combinator F26 application, you should ask the obvious follow-up.

Does Y Combinator Use AI or an ATS to Screen Applications?

No. YC does not run your application through an automated scoring model the way that HackerRank tool scores a resume. Every YC application is read by people. The core reviewers are YC group partners and visiting partners, founders who have built, scaled, and sold companies themselves, plus a network of trusted YC alumni who help with the first pass and surface promising applications up the chain.

Garry Tan and other partners have said this repeatedly: software manages the workflow, but humans make the decisions. YC's stated reasoning is exactly the failure mode the HackerRank experiment exposed. The thing that actually predicts a great founding team, the nuance of who these people are and why this idea, is too subtle for an algorithm to grade consistently. So they don't try.

That means the distribution problem that wrecks AI resume scoring does not apply to your YC application in the same mechanical way. You are not failing a keyword threshold at temperature 0.1. A person is reading your answers.

Why the HackerRank Mess Is Still a Warning for F26 Applicants

Here's the useful part, because "a human reads it" is not a free pass. Look at what the open-source ATS could and couldn't score reliably.

The one category it nailed was technical skills: the developer scored 8/10 on 98 of 100 runs. Why? Because skills are a checklist. You know React or you don't. There's nothing to judge. The categories that swung wildly were the judgment calls, projects most of all, where the same work was alternately praised for "real-world deployment" and dinged for lacking "architectural complexity."

Your YC application has the same two kinds of content. The checklist parts (what you've built, your traction numbers, your launch status) are unambiguous. The judgment parts (is this idea big, are these the right founders, is the insight real) are where a reviewer forms an opinion in seconds. The lesson isn't "game the checklist." It's the opposite: a checklist is the floor, not the case. If your application leans on buzzwords and keyword density the way candidates pad resumes to beat an ATS, you're optimizing for a filter that YC doesn't use, and starving the part a human actually weighs.

What a Human Reviewer Is Actually Scoring

The leaked HackerRank rubric is almost comically thin where it matters. Its entire "experience" prompt is two lines with no anchors, so a junior engineer with one internship and a principal engineer with a decade of distributed systems both get 25/25. Useless. Tellingly, the same template hands out bonus points for "founder roles, co-founder positions, or early-stage engineer roles."

A YC partner reading your application does the thing that rubric can't: they read for specifics. Not "we're building an AI platform," but what you shipped, who is using it this week, what you learned that your competitors haven't, and how fast the line is going up. The single most repeated piece of YC advice is to be concrete and concise. Vague applications read as vague thinking. The HackerRank tool gives everyone 25/25 because it can't tell the difference; a human reviewer can, and the difference is the whole game.

So the answer to "how do I beat the screen" is unsatisfying and freeing at once: there's no screen to beat. There's a person to convince in about a minute of reading.

Where AI Does Touch the YC Funnel

The one place AI shows up is on your side of the table, not YC's. Plenty of F26 applicants are drafting answers with ChatGPT or Claude, and YC partners have been openly skeptical of applications that read like generic model output. A human reviewer who reads thousands of applications notices the flat, hedged, everyone-says-this cadence instantly. Using AI to tighten your writing is fine. Letting it write answers that could belong to any of the tens of thousands of applicants is the actual risk, because the reviewer is precisely the pattern-matcher that catches it.

How to Pressure-Test Your F26 Application Before a Human Sees It

If YC's filter is a human, the best preparation is a human, ideally one who has sat where the partners sit. Run your draft past someone who can tell you, in plain terms, whether your traction is actually impressive or just dressed up, whether your idea reads as big or incremental, and where a partner would lose interest by line three. That's the feedback an automated scorer literally cannot give you, and it's the reason YC Roaster exists: it connects F26 applicants with YC alumni who'll read your application the way a partner will and tell you what's working before you hit submit.

The takeaway from today's HackerRank story isn't that AI screening is broken, though it clearly is. It's that YC deliberately doesn't use it, because the things worth funding can't be scored to two decimal places. So stop writing for an imaginary algorithm. Write for the experienced founder who's going to read your application start to finish, and make the case so specific that no temperature setting could blur it.

Quick FAQ

Does YC use an ATS to filter applications? No. YC group partners, visiting partners, and alumni reviewers read applications directly; software only manages the workflow.

Will using AI to write my YC application get me auto-rejected? There's no auto-rejection bot, but human reviewers are good at spotting generic AI-written answers, which hurts you. Use AI to edit, not to invent your story.

What matters most to a human reviewer? Concrete specifics: what you've built, real usage and growth, and a clear, non-generic insight, delivered concisely.