Which coding agents can candidates use?

Claude Code and Codex run in the VS Code terminal, and candidates launch either with one click from the toolbar. The AI Assistant in the editor offers Agent, Chat and Plan modes with models from OpenAI, Anthropic and Google. Claude Code and Codex are available across Screen, Interview and Skills Intelligence.

Do candidates need their own account or API key?

No. Claude Code and Codex come preconfigured in the environment. AI use draws on the Codility tokens in your plan. Administrators can cap tokens per candidate, and Extended AI Usage adds a further block of Codility tokens for longer sessions.

Can the same loop include rounds without AI?

Yes. Switch AI on or off for each interview, and an administrator can switch off all AI features across the account. In a Screen assessment, AI can be switched on or off for each task. With AI off, integrity signals reach the interviewer live during the session.

Does an AI score the candidate?

No. A human interviewer runs the session and scores against the rubric. Every AI interaction is saved, so the panel can review how the candidate worked. The AI skills the report shows are information for the panel and do not change any score.

Does Codility Interview work with Ashby?

Yes. Interviews can be created from Ashby, and the report syncs back to the candidate record. Codility connects with more than 80 applicant tracking systems, human resources and scheduling tools.

Run agentic coding interviews on a real codebase

Candidates and interviewers share one VS Code environment with Claude Code or Codex in the terminal. You keep every prompt, command and code change.

Request a free trial

Take the product tour

AI tools included, no API keys needed Validated assessments since 2009 SOC 2 Type II ISO 27001

Claude CodeCodex

~/order-serviceRecording

+412

lines the agent wrote

4

decisions worth asking about

Interactions per tool · Sample

Claude Code11

Codex3

AI Assistant2

Session recordSample

112:40Interviewer adds a requirement: limit per API key

216:30Scopes the agent only touch ratelimit.py

321:40Rejects the agent’s edit to db.py

427:50Runs the tests: 23 passed

Every prompt, command and code change, in order

Engineers ship with coding agents now. Most interviews still can’t show who directs them well.

01

A single-file exercise leaves the hardest part of the job untested: steering an agent through a real codebase.

02

An agent can write hundreds of lines in minutes. Interviewers need to see what the candidate asked for, checked and changed.

03

Most loops mix rounds with AI and rounds without it. Running them in separate tools splits the evidence.

Run agentic coding interviews in a real engineering environment

Start from a real project

Save the interview as a multi-file project with its dependencies and services, such as a database or cache, already in place. The candidate starts on the task as soon as the session opens. Build tasks from your own code with the task creation MCP server.

Codility task library showing a saved multi-file interview project, its file tree and attached database and queue services, sample data.

Claude Code running in the Codility assessment terminal beside the saved interaction record.

Give candidates the real Claude Code and Codex

Claude Code and Codex run in the VS Code terminal. One click launches either, with no account or API key needed.

Claude Code is from Anthropic, and the Codex CLI is from OpenAI
Both read files, run commands and edit code in the project
In the editor, the AI Assistant adds Agent, Chat and Plan modes with models from OpenAI, Anthropic and Google

Decide what AI is allowed and steer the session

Switch AI on or off for each interview and choose which AI Assistant modes and models the candidate can use. Every prompt and response is saved for review. Write AI Instructions that the AI Assistant, Claude Code and Codex all follow during the interview. Interviewer and candidate share the terminal, so the interviewer can add a requirement and watch how the candidate takes it on.

Codility Interview settings showing the AI Assistant enabled for the interview and an AI Instructions field with sample guidance.

Codility Interview report showing rubric scores, the AI usage panel with tools and interactions per model, code playback and the transcript, sample data.

Review how the candidate directed the agent

Replay the session with every prompt, terminal command and code change in order. The report shows which AI tools the candidate used, telling Claude Code and Codex apart, and how many interactions went to each model, next to the code, the transcript, integrity signals and the interviewer’s rubric scores. It also shows which AI skills the candidate used, with quotes from their own prompts as proof.

Explore Codility Interview

Assess engineers who use AI across hiring

Run every round of the loop in one environment

Codility Interview runs rounds with AI and rounds without it, and when AI is off, integrity signals are designed to reach the interviewer during the session.

Every round produces the same record, rubric and report, backed by almost two decades of assessment science and a documented methodology.

See how integrity signals work

Candidates know what AI is allowed before they start

90%

of candidates say they received clear instructions on AI tool usage.

Codility candidate survey, October 2025 to April 2026. 34,532 responses to this question, across Codility assessments.

“The assessment was practical, well-structured, and closely aligned with real-world problem-solving. The tasks encouraged thoughtful data handling rather than rote coding.”
Candidate, Codility post-assessment survey

Candidate feedback survey, October 2025 to April 2026

Running an agentic coding interview

Run your next agentic coding interview in Codility Interview

Send us one task. We will set it up with Claude Code or Codex and walk your interviewers through the session and the record.

Talk to us