//definition

The AI Collaboration Score, for coding agents.

kodwai's AI Collaboration Score is a 0 to 100 score for one coding session with an AI coding agent (Claude Code, Cursor, or Codex) that measures how well the developer directed the agent, across three axes: Direction, Outcome, and Lift.

It is the score kodwai gives every challenge submission. It rewards how you steer, verify, and decompose the work, not what you memorized, and it ranks you on a public leaderboard.

By Hakan Karaagac, co-founder of kodwai · Last updated 2026-09-22

The three axes

Direction, Outcome, and Lift, built from 11 signals. These are the labels the app shows on every scored session.

Direction

How you steer, verify, and decompose.

Spec Precision: Stating clear requirements and constraints before writing code.
Verification Rigor: Checking the AI's output, catching mistakes, pushing back.
Decomposition: Breaking the problem into ordered steps instead of one mega-prompt.
Recovery: Redirecting effectively when the AI goes down the wrong path.
Intent Fidelity: The final solution matches what you actually asked for.
Engagement: Staying engaged and iterating, instead of pasting the spec and walking away.

Outcome

What shipped, replayed and stress-tested to prove it holds.

Tests: Share of the challenge's tests your solution passes.
Code Quality: Clean, readable code without obvious smells.
Complexity: Reasonable structure and nesting.

Lift

How far you beat a solo AI, not just that you passed.

Edge-Case Coverage: Handling subtle requirements a careless one-shot would miss.
Lift over AI: How far you out-perform a solo AI on this challenge.

In the default profile Direction is worth 50 points, Outcome 35 and Lift 15. Other challenges use other splits, and a challenge with its own rubric scores Direction, a Challenge Rubric axis and Lift. On the default profile Direction carries the most weight: it is the part a careless prompt cannot fake.

How a session is scored

kodwai is a platform where developers solve real coding challenges on their own machine with their own AI coding agent (Claude Code, Cursor, or Codex) and get scored on how well they direct the agent, across three axes: Direction, Outcome, and Lift.

You start a challenge with npx @kodwai/cli challenge <slug>. The CLI asks which agent you will use, downloads PROBLEM.md, starter files and tests, inits a git repo, and starts the timer.
You solve it on your own machine, in your own editor, with your own agent. No browser sandbox, no artificial constraints.
npx @kodwai/cli submit packages your code, git history, test runs, agent transcript, and the time you took. The CLI only collects from the challenge workspace.
The session is scored on Direction, Outcome, and Lift. Every signal cites its own evidence from your transcript, commits, and test runs.
You get a score from 0 to 100 with a confidence interval instead of false precision, per-signal evidence, and a place on the public leaderboard.

Why passing tests is not enough

A one-shot “solve this” prompt can clear the tests, so tests alone say little about how someone works with an agent. The score is dominated by how you direct the agent: whether you stated the spec, checked the agent's output, broke the work into steps, and recovered when it went wrong. Lift then asks how far you beat a solo AI, not just that you passed.

Who it is for

Developers

Practice directing Claude Code, Cursor, or Codex on ticket-sized work and see which signals held you back. Your public profile at app.kodwai.com/developers/your-username is built to send to anyone, including a hiring manager instead of a take-home.

It is also practice for interview rounds where AI use is allowed, such as the AI-enabled coding interview Meta started rolling out in October 2025.

Hiring teams

Teams use kodwai's hiring track: custom interview projects, one link per candidate, the full session live (prompts, commits, test runs, tools, time, and API cost), and AI scoring against your own per-role rubric next to your team's manual scores. Candidates work in Claude Code on their own machine.

See kodwai for hiring

Try it on a real challenge

The starter challenge is Bookshelf REST API (easy, backend, 60 minutes). Create a free account, then run:

$npx @kodwai/cli challenge bookshelf-rest-api

easybackend

How kodwai compares

Honest comparisons that say where each product is the better choice.

Frequently asked questions

Is a high score the same as passing the tests?

No. Tests are one signal inside Outcome. Passing tests is necessary but not sufficient: the score is dominated by Direction, the part a careless prompt cannot fake. A solution that clears tests with no steering, no verification, and no decomposition scores poorly on the axis that matters most.

Which coding agents can I use?

When you start a challenge, the CLI asks whether you will use Claude Code, Cursor, or Codex. You work in your own editor on your own machine.

Are the axes weighted the same on every challenge?

No. In the default profile Direction is worth 50 points, Outcome 35 and Lift 15. Other challenges use other splits: a debugging profile weights Direction higher, and a challenge with its own rubric scores Direction, a Challenge Rubric axis and Lift.

Can I see why I got my score?

Yes. Every signal cites its own evidence from your transcript, commits, and test runs, and your score comes with a confidence interval instead of false precision.

What does the CLI collect?

Only the challenge workspace: the code files, the challenge session's git history, and agent traces from the challenge time window. Nothing from your other projects.

Is it free?

Yes. Solving challenges, your score, your profile, and the leaderboard are free for developers.

Is this the score hiring teams see?

Not directly. kodwai's hiring track scores interview sessions against each team's own per-role rubric, 0 to 10 per dimension, with a written justification. That is a different number from the public 0 to 100 score described here.

Is the name unique to kodwai?

No. Other products use similar wording: Corrix, for example, offers an AI Collaboration Profile about how effectively people work with AI in general. On kodwai, the AI Collaboration Score always means the score for a session with an AI coding agent.

How this was made

Written from kodwai's scoring code and the live site copy, with help from Claude, an AI model, on 2026-09-22. The signal names and descriptions are the ones the kodwai API serves with every challenge. If something here is wrong or out of date, email hakan@kodwai.com.

Get your own score.

Pick a challenge, solve it with your agent on your own machine, and see how you direct it. Free for developers.

Create a free account→