Kendor vs traditional take-home assignments
Facts last reviewed: August 3, 2026
Kendor keeps what take-home assignments get right — realistic, job-shaped project work — and fixes what they cannot deliver: a view of how the work was done. Candidates work in a provisioned VS Code-based workspace in the browser, the screen goes out automatically when your ATS moves them to the right stage, and they run the project’s tests from the Tests tab as they go. You get take-home realism with a timeline of how the work actually happened and a review your team runs in the candidate’s actual workspace.
A classic take-home is an email, a zip file, and trust. In 2026 — when a polished solution can be generated in minutes — that trust gap is the whole problem. The fix is not surveillance; it is evidence.
| Kendor | Traditional take-home | |
|---|---|---|
| What candidates doThe format of the screen itself | Real project challenges (Node/TS, Laravel, Python, Go, Java, Rust, C++), scoped and timeboxed, with PostgreSQL or Valkey inside the sandbox when needed — from a template or imported from any Git host, private repos included | An open project on the candidate’s machine, usually unbounded |
| EnvironmentWhere the candidate writes code | Full VS Code-based IDE in the browser: terminal, live preview, multi-service apps — a link, no account, zero setup; Docker, Compose or Kubernetes sandboxes on Enterprise | Their own machine and IDE — setup cost is theirs |
| Hiring flowGetting applicants into a screen | Ashby, Greenhouse and Lever, included in Pro: moving a candidate to a stage you choose sends their screen or a link to book a live coding interview, and Kendor reports progress back to the ATS. Or invite candidates directly | Manual email logistics: instructions out, zip files back |
| Judging the workWhat reviewers decide from | The code and its diff, terminal and AI-agent transcripts, and a keystroke playback, plus on Pro an optional AI summary on your own Anthropic or OpenAI key that points to the lines worth reading — a person decides, not a number | Trust — nothing shows whose code it is or that it runs |
| Integrity signalsWhat you learn about how the work happened | Timestamped facts: tab switches, fullscreen exits, paste events, full keystroke playback | None — you see only the final zip |
| AI policyWhether candidates may use AI | Each org decides per assessment; AI use is a fact to evaluate, and reviewers read the agent transcript beside the code | Unknown — you cannot tell either way |
| Live interviewsPairing with the candidate in real time | Live coding from one link: a shared sandbox (code, terminal, running app), whiteboard, built-in video call with screen sharing, and candidate self-booking for scheduled interviews | None — a separate call, if any |
| Team reviewHow the hiring team decides | Reviewer verdicts and pass/advance/reject decisions with notes, multiple reviewers, inline comments, live collaborative review with a built-in video call | Ad-hoc — whoever reads the zip decides |
| AI agentsAI assistance inside the sandbox | Claude Code or Codex in the sandbox terminal, enabled per assessment or per live session by the org on its own API key, with a token budget per attempt. Take-home screens can instead let candidates use their own account with any of seven agents (Claude Code, Codex, GitHub Copilot, Cursor, Cline, OpenCode, Antigravity), at no cost to the org. Reviewers can watch the terminal live | Whatever the candidate uses, invisibly |
| PricingWhat drives the bill | Flat plans, never per candidate — Starter $79/mo; Pro $400/mo billed yearly ($500 monthly) with unlimited candidates, interviews and reviewers; free trial, self-serve with a work email | Free — paid for in your team’s review time and candidate drop-off |
| Data protectionWhere candidate data lives | EU-hosted (Frankfurt), GDPR-first: signable DPA, export and erasure, no trackers in the product | Depends entirely on how you handle emailed submissions |
What take-homes get right
Take-homes are popular for good reasons: candidates work on realistic problems, in relative calm, without an interviewer watching them type. Kendor is built on the same belief — real work over puzzle performance — and keeps that shape: project challenges with a real codebase, a terminal, and a running app.
What a take-home cannot tell you
Whose code is it? How long did it really take? Does it even run outside the candidate’s machine? A traditional take-home answers none of these. Kendor answers all three: the workspace records a timestamped activity timeline and a keystroke playback, the environment is identical for every candidate, and the code runs there, not on the candidate’s laptop — "works on my machine" is not a factor, and pasted-in code shows up on the timeline.
Logistics: from manual email to automatic
Take-homes run on manual effort: emailing instructions, chasing submissions, cloning repos, installing dependencies, and writing feedback from scratch. On Kendor, moving a candidate to a stage in Ashby, Greenhouse or Lever can send the screen automatically, submissions land in a review queue with the workspace, transcripts and timeline attached, and your team reviews the actual workspace together — no setup on either side.
Respecting the candidate’s time
Unbounded take-homes quietly cost candidates evenings and weekends, and the best candidates decline them. Kendor challenges are scoped and timeboxed, need zero local setup, and let candidates work in a familiar IDE in the browser. Respecting candidate time is not just kindness — it protects your funnel.
When a classic take-home is still right
If you are hiring a very senior engineer for open-ended design work, an unconstrained take-home (or better, paid trial work) on their own machine may tell you more than any platform — and some candidates will rightly prefer it.
Frequently asked questions
Are take-home assignments still worth it now that AI can solve them?
The format is worth keeping; the blind trust is not. A realistic project is still the best predictor of job performance, but an emailed zip can no longer prove who — or what — produced it. A shared environment plus a timestamped work timeline keeps the realism while making the process visible.
How do you know a take-home was not written by AI or a friend?
On Kendor you see how the work happened, not just the final file: tab switches, paste events, and full keystroke playback, plus the terminal and AI-agent transcripts. Organizations that allow AI can enable Claude Code or Codex in the sandbox terminal and judge the result on its merits; organizations that do not can weigh the timeline evidence.
How long should a take-home assignment be?
Aim for 2–4 hours of focused work, scoped so a strong candidate can finish. Kendor challenges are timeboxed by design, and because the environment is pre-provisioned, none of that budget is spent on setup.
Do candidates need to install anything to do a Kendor screen?
No. The workspace — a VS Code-based IDE with a terminal, dependencies, a live preview, and a PostgreSQL or Redis-compatible database when needed — runs entirely in the browser. Candidates open a link and start.
More comparisons
Kendor vs CoderPad · Kendor vs Codility · Kendor vs HackerRank · Kendor vs CodeSignal · Kendor vs Coderbyte
See it on the Kendor homepage or read the engineering-hiring blog.