private interviews · claude code

kodwai for hiring: See how candidates really build.

Give candidates a realistic coding task as a private interview and see the whole session: every prompt, how they recover when the agent gets it wrong, how they check its work, and the final result. Watch it live, score it against your own rubric and compare candidates side by side.

Set up interviewsa developer?start a challenge
  • claude code on their own machine
  • scored 0 to 10 on your rubric
  • full transcript of every step
cand_03sr backendtime18:24 / 60:00cost$0.42 / $5.00tools9
~/kodwai/rate-limiter · claude codeuploaded
Example interview. The candidate runs the interview command, writes a failing test first, fixes a concurrency bug the agent introduced, commits, and exits at 18:24. The session uploads and is scored against the team's rubric.
example interview · sample candidate

the problem

The take-homeis broken.

A take-home is homework. You wait days for a final diff that hides how it was built, and the candidate with the most free time wins. You never see how they work with an AI agent, which is most of the job now.

  1. 01

    It is not the real job.

    You grade a final diff days later, without seeing any of the decisions behind it.

  2. 02

    It rewards free time.

    Whoever can spend a weekend on it wins. That is not the signal you are paying for.

  3. 03

    Easy to outsource.

    You cannot tell who actually did the work, or how.

  4. 04

    You never watch them work.

    How they guide an agent and recover when it gets things wrong matters most in the job today. A take-home hides exactly that.

take-homenot recorded

A zip file and a README, three days later.

how they got there: unknown
kodwai interviewrecorded

Every prompt, test run and commit, in the order it happened.

how they got there: every step

how it runs

One link.One command.Their machine.

Candidates do not fight a browser sandbox or sign up for a portal. They work the way they do every day.

  1. 01

    Write the interview

    Write a problem that looks like your own codebase. Set a time limit, a difficulty and a rubric: the skills you care about, each weighted 1 to 10, with a short note on what good looks like. Reuse it for every candidate.

  2. 02

    Send one invite

    Add the candidate by name and email. The email carries a session id, a token and one command.

    $npx @kodwai/cli@latest start <session-id> --token <session-token>
  3. 03

    They solve locally

    The CLI fetches the problem and the time limit, sets up a sandboxed workspace, starts the timer and launches Claude Code. Agent usage runs on a separate API key that you pay for, capped at the budget you set.

  4. 04

    Watch, or read it later

    Follow the session as it happens or open it when it ends. When time runs out or they type /exit, the session uploads automatically.

  5. 05

    Score and compare

    The AI scores the session against your rubric. Your team adds its own scores, pins comments to specific moments, and compares the candidate with others for the same role.

session replay

See every stepthey took.

A take-home gives you the final files. kodwai shows you everything that led there, in order: what they asked the agent for, what it did, what they checked and where they pushed back.

It is a written transcript you can read and comment on. There is no video.

example transcript · candidate 03 · rate limiter10 events shown
  1. 00:41promptfailing test for window eviction first, then implement.
  2. 01:12agentadds test_window_eviction
  3. 01:21test1 failed · NameError: SlidingWindow
  4. 04:55agentimplements SlidingWindow in limiter.py
  5. 05:03test7 passednote 1
  6. 06:10prompttwo workers hit the same key. make it safe.
  7. 08:52testRuntimeError: deque mutated during iteration
  8. 09:30promptsnapshot the window before pruning. keep the lock narrow.note 2
  9. 12:47test9 passed · ruff clean
  10. 13:05commitcommit: sliding-window limiter, per-key lock

the scorecard

Score them onyour rubric.

You write the rubric for the role. Every finished session is scored against it from 0 to 10 per dimension, with a written reason for each score. Your team then adds its own scores, and you see both side by side.

set per interview

time limit
per interview, in minutes
budget
API spend cap in USD
tools
allow or block per interview
roles
admin · interviewer · viewer
scorecard · candidate 03sliding-window rate limiter · sr backend
ai score
8.4/ 10
team score
8.0/ 10
Problem decomposition
Split the ticket into eviction, concurrency and cleanup before any code.
×38.58.0
Direction and verification
Wrote a failing test before every change and ran the suite after each agent edit.
×39.08.5
Code quality
Small, readable limiter. Kept the lock narrow on the second pass.
×28.08.0
Recovery under failure
Found the deque error from the traceback. Did not cover clock skew.
×27.57.0

ai summary

A methodical session. Planned the task before touching code, guided the agent in small steps that were easy to check, and cleanly fixed the concurrency bug the agent introduced with its first lock.

strengths

  • Wrote a failing test before each change
  • Caught the iteration bug the agent introduced
  • Kept the commit small and clearly labelled

weaknesses

  • Left clock skew as an uncovered edge case
  • Did not write down the final trade-off
example scorecard · sample candidate

Compare candidates for a role.

Example candidates with sample scores.
Example candidate compare for one role, scores out of 10
candidateDecomposeDirectionQualityRecoveryoverall
cand_017.07.59.0best6.57.5
cand_026.06.57.08.5best6.9
cand_038.5best9.0best8.07.58.4

what is included

Nothing toconnect.

You send one link and get back one scored session. There is no ATS integration to set up, because kodwai does not have one yet. Share a session with a link and invite your team as admins, interviewers or viewers.

Candidates work in Claude Code. Cursor and Codex are on the developer track, not in interviews yet.

the hiring track8 included · 6 not included

included

  • Interview projects: problem, time limit, difficulty
  • A rubric per role, weights 1 to 10
  • Budget cap and allowed or blocked tools
  • One invite and one command per candidate
  • Live transcript, file changes, tool calls, time and API cost
  • AI score and manual scores, comments pinned to events
  • Candidate compare for a role
  • Team roles: admin, interviewer, viewer

not included

  • ATS integrationsnot yet
  • Video recordingsyou get a written transcript
  • A candidate portalcandidates just run one command
  • Schedulingsend the link when you like
  • A template libraryyou write the problem
  • A pricing pageemail us

questions

Before yousend a link.

candidateHow does a candidate take a kodwai interview?

You add them by name and email, and they get an invite with one command: npx @kodwai/cli@latest start with their session id and token. They solve the problem on their own machine in Claude Code. There is no browser sandbox and nothing to install beyond the CLI.

observeWhat can I see while they work?

The full session, live: every prompt, the file changes, the test runs, the tools used, the time elapsed and the API cost. Follow along as it happens, or open the session afterwards and read it in the order it happened.

scoringHow are interviews scored?

You define the rubric for the role: the dimensions, a weight from 1 to 10 for each, and what each one means. Every finished session is scored against it from 0 to 10 per dimension, with a written justification, strengths and weaknesses. Your team adds manual scores and comments next to the AI score.

problemCan I use my own problem?

Yes. Write the ticket so it looks like your real codebase, then set the time limit, the budget and the tools the candidate may use. Reuse it for every candidate so the comparison stays fair.

teamDo you connect to my ATS?

Not yet, and we will not pretend otherwise. There is nothing to wire up: you send a link and a scored session comes back. Share that session with your team by link, and invite teammates as admins, interviewers or viewers.

pricingIs the hiring track free?

No. Solving public challenges is free for developers, but the hiring track is paid. There is no public pricing yet. Email hakan@kodwai.com and we will get your team set up.

get started

Hire for howthey workwith AI.

Send one link, watch the real work, score it on your own rubric and decide as a team.

the command your candidate runs

$npx @kodwai/cli@latest start <session-id> --token <session-token>