Your engineers build on Claude Code and Codex. They steer agents, validate output, debug hallucinations, and ship working software by orchestrating AI, not writing every line by hand.
Then you interview candidates on a locked-down IDE with autocomplete disabled. You are testing for a job that does not exist anymore.
Why we built this
Traditional interviews test two things: can this person recall algorithms under pressure, and can they write syntax from memory. Neither tells you how they actually work.
The take-home version is worse. Candidate works in private, sends back a zip file. You grade the output. Now every submission is flawless because an AI wrote it. 91% of hiring managers suspect AI-generated answers in their process (Greenhouse, 2025). The format did not break. The assumption that output equals skill broke.
Most companies went the other direction. They added more proctoring, built "AI-resistant" questions, started a detection arms race with tools like Cluely and Interview Coder. Fabric found 38.5% of candidates show cheating behavior across 19,000+ interviews (Fabric HQ, 2026). That number will keep climbing as long as the industry treats AI as something to catch rather than something to evaluate.
The fix is not banning AI. The fix is letting candidates use it, and watching how.
What you get
Fairground's Interview Canvas gives candidates a real, multi-file project with CLI agents built in: Claude Code, Codex, Gemini CLI, OpenCode, and others. Real codebase, real documentation, real problems.
You see:
How they decompose problems into prompts an agent can handle
Whether they validate output or blindly accept suggestions
When they steer the agent away from a bad approach
When they stop using the agent and write code themselves
That is four signals you get zero of from a whiteboard or a zip file.
Two ways to use it
Live, with an interviewer. Your engineer joins the session and watches the candidate build in real time. They see every prompt, every agent interaction, every decision. The interviewer can ask follow-up questions while the candidate works. You get the full picture of how this person collaborates with AI tools under real conditions.
Async, as a take-home. Send an invitation. The candidate works on their own schedule, inside the same Interview Canvas with the same tools. No human time spent until you have results. You evaluate AI skills across 50 candidates before a single engineer sits down for a live round.
Both modes produce the same output: a full session recording, a complete transcript of every candidate-AI interaction, and an AI-Readiness Scorecard with dimensional breakdowns of technical skill and AI judgment. Not a single pass/fail number. Evidence.
Everything is proctored. 20+ integrity signals captured automatically. Full audit trail.
Get started
Meta, Anthropic, and Canva already test candidates with AI tools enabled. They built that evaluation internally because no platform offered it. You do not need 1,000 engineers to do the same.
Fairground's Interview Canvas gives your team the same capability. Configure your questions, pick live or async, and see how candidates actually build. Start free, 100 credits. No credit card. No sales call.
Related: Take-Homes Were Never the Problem. Black Boxes Were. | Every Interview Loop Needs an Async Round
Get started with Fairground in just few mins.
Plug and Play. Works well with your existing ATS.
100 Free Credits
