Outship Doesn't Grade Your Code. It Grades How You Drive the AI.

For about twenty years, the price of admission to a software job was a whiteboard and a linked list. You reversed it, you balanced a tree, you talked through Big-O, and someone decided whether you could build. Then the whiteboard learned to code by itself. Outship, a startup from Y Combinator's Winter 2025 batch, is one of the first companies to treat that shift as a hiring problem worth solving from scratch.

The premise is blunt. If an AI agent can produce a working pull request from a one-line prompt, then producing a working pull request is no longer proof of much. Outship's founders put it plainly on their own site: being able to ship is the new baseline. What matters after that is harder to see and, until recently, impossible to record - how a person breaks a task apart, how they steer the agent, which suggestions they throw away, and what they do when the build turns red at the worst moment.

What Outship Actually Does

---

Outship is a technical screening platform for engineering hires. A company imports a task - either from a GitHub repo or from its own codebase - and sends the candidate an invite link for a take-home or a live interview. Instead of a puzzle in a text box, the candidate lands in a real environment: an isolated cloud VM running a browser-based VS Code, with coding agents ready to go. They get a terminal, they can run web apps, forward ports, and work the way they actually work on a Tuesday afternoon.

While they build, Outship records the session. Every prompt sent to the agent, every edit, every command, every decision. When the task is done, the platform analyzes both the final code and the transcript of how it was made, scoring the session against a rubric the hiring team can customize. The pitch to recruiters is that they finally get to read the process, not just the artifact.

The Interview, Step by Step

---

# candidate session - recorded by Outship

candidate> read the failing test, then patch the parser

agent: proposing 42-line diff in parser.ts ...

candidate> too broad. only touch the tokenizer

agent: revised 9-line diff

✓ 218 tests passing | recovery logged | rubric: judgment +2

Why It's Different

---

The incumbents in technical screening - the platforms that gave the industry its timed algorithm rounds - were built for a world where a human typed every character. They measure whether the final answer is correct and how fast it arrived. Outship measures something the older tools were never designed to capture: the shape of the work when an AI is in the room. In a LeetCode round, using an agent is cheating. In an Outship round, it's the whole point.

That difference plugs into the tools recruiters already use. Outship connects to applicant tracking systems including Ashby, Greenhouse, and Workday, so recorded assessments and their rubric scores land inside an existing pipeline rather than in a separate dashboard nobody checks. AthenaHQ is among the early customers using it.

Who's Building It

---

Outship was founded in 2025 by Saner Cakir, the CEO, and Kayla Lee, the COO. Both came out of Berkeley AI Research - Cakir worked on robot learning, Lee on natural language processing. The idea did not arrive as a hunch. They spent time talking with more than 50 YC teams and kept hearing the same thing: nearly everyone was building aggressively with coding agents, and nobody's interview process reflected it. Teams weren't hiring people to type code line by line anymore. They wanted builders who could point the newest tools at a problem and get a week's work done in a day.

The company is small - three people, based in San Francisco, backed by Y Combinator with partner Gustaf Alstromer. That size is part of the story. Outship is a research-shaped team building a product about how work gets done, which is roughly the thing they spent years studying before they started selling it.

Where It Fits

---

Outship sits in the technical-assessment market, next to names like HackerRank, CodeSignal, CoderPad, and Karat. Its bet is that the category is due for a reset. If the industry agrees that agent fluency is now a core engineering skill, the tools that can actually observe it have a reason to exist. If most companies decide a clean final PR is good enough, Outship's edge is narrower. The wager is that the process is the product - for candidates, and for the company grading them.

For a hiring team, the practical use is simple to describe. Replace the take-home that a candidate could have finished with an agent in secret with one where the agent is expected, visible, and recorded. Read how someone thinks under a real workload. Compare candidates on judgment rather than typing speed. And do it inside the ATS you already run.