Where can candidates use Claude Code and Codex?

Claude Code and Codex run in the Codility assessment environment across Screen, Interview and Skills Intelligence. Every interaction is saved to the record. Account and assessment settings control access.

What happens if our policy prohibits AI?

An administrator can turn all AI features off across the account. Assessments can also disable the AI Assistant individually. Integrity checks remain available for human review.

How does Codility surface undisclosed AI use?

Behavioral and similarity checks can surface signals of undisclosed assistance. Reviewers inspect the evidence behind the risk rating. No signal proves misconduct on its own, and a reviewer makes every decision.

Does the amount of AI use affect the score?

AI usage metrics and the AI skills the report shows do not change automatic task scores. Selected task types use fixed scoring rules. In a live interview, the human interviewer assesses the candidate against the rubric.

Can we assess our existing engineers this way?

Yes. Skills Intelligence uses real-world tasks and the same AI controls to assess existing engineers. The results inform skills development and staffing decisions.

What does AI use cost?

Plans include a token allowance for AI use. Extended AI Usage adds a further block of Codility tokens for assessments that need more.

AI coding assessments with the real Claude Code and Codex

Candidates work in a real VS Code environment. See how they direct the agent and check its output.

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-service Recording

› the discount rule fails for stacked coupons. find out why

● Percentage coupons apply after fixed ones.

Fix: swap the order on line 46.

› wait. that breaks fixed-only orders. run the tests first

2 failed test_fixed_only, test_stacked_coupons

› the bug is apply_cap() running twice. remove that

● Edited pricing.py -1

12 passed in 0.42s

AI skills evidence07:15

AI Output Validation

“wait. that breaks fixed-only orders. run the tests first”

Tested the agent’s fix before accepting it and caught a regression.

Quoted word for word from the transcript

Read the failing test

Directed the agent

Rejected a flawed fix

Tests pass, 12 of 12

› the discount rule fails for stacked coupons. find out why

● Percentage coupons apply after fixed ones.

Fix: swap the order on line 46.

› wait. that breaks fixed-only orders. run the tests first

2 failed test_fixed_only, test_stacked_coupons

› the bug is apply_cap() running twice. remove that

● Edited pricing.py -1

12 passed in 0.42s

AI skills evidence07:15

AI Output Validation

“wait. that breaks fixed-only orders. run the tests first”

Tested the agent’s fix before accepting it and caught a regression.

Quoted word for word from the transcript

Read the failing test

Directed the agent

Rejected a flawed fix

Tests pass, 12 of 12

A correct solution no longer shows what the candidate understands

01

An assistant can produce working code. The code alone shows little of what the candidate understands.

02

If you don’t allow AI, you never see how a candidate uses it.

03

Allowing AI with no record of its use leaves reviewers guessing how the work was done.

See the engineer’s judgment in the work

Decide whether candidates can use AI

Allow the AI Assistant or turn it off for each task, and every prompt and response is saved for review. Account administrators can also switch off all AI features. Integrity signals support human review in either case.

Codility settings showing the account-wide AI switch and, beside it, the AI Assistant setting for a sample assessment.

Codility Interview session showing the shared VS Code environment with the Claude Code terminal open and both participants listed in the session.

Put Claude Code and Codex into the assessment

Candidates use Claude Code or OpenAI’s Codex CLI in the VS Code terminal alongside project tools, so they work with the AI tools they already use. The full interaction is saved to the report and covered by Replay. In Interview, both participants can use it live.

Choose tasks that test how engineers use AI

Ask candidates to debug, investigate or build with an assistant. Some tasks include deliberate assistant errors to check and correct. For AI engineering roles, assess candidates building AI or machine learning systems.

Codility Work with AI task in VS Code showing the task brief, an assistant suggestion containing an error, and the candidate's corrected code.

Codility Screen assessment report for a sample candidate showing task scores, the submitted code and the integrity signals panel.

Review the solution and how it developed

Replay the code and AI exchanges to see what candidates asked, accepted and fixed. Each AI skill comes with a quote from their prompts.

Giving the AI the right context
Iterating on its answers
Checking what it produces
Deciding what to hand it
Spotting where it goes wrong

Explore Screen for asynchronous hiring

Explore Interview for live technical sessions

Make AI collaboration part of the assessment

A conventional coding exercise can show whether the final solution works. An AI detection tool addresses a different question: whether assistance may have been used.

Codility puts the assistant into the task and preserves the interaction. Reviewers examine the candidate’s work and decisions, supported by documented assessment methodology.

Explore the assessment methodology

What a score gives you

74%

Python · 41 min

Did they write it, or did the agent? Would they catch the same bug on the job? The number cannot say.

What Codility gives you74%

Codeorder-service, 4 files changed, tests passing

ReplayAll 41 minutes of work, in the order it happened

Coding agentClaude Code or Codex: 7 exchanges, saved word for word

AI skillsAI Output Validation: “wait. that breaks fixed-only orders”

IntegrityIdentity verified, no similarity match, reviewer decides

MethodScored by fixed rules, documented in the Technical Manual

Candidates know when AI is allowed. Reviewers see every prompt.

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 tasks were relevant to real programming problems and focused on practical problem-solving rather than memorization.”
Candidate, Codility post-assessment survey

Candidate feedback survey, October 2025 to April 2026

Assessing engineers with AI in the workflow

See how a candidate works with Claude Code or Codex

Review a task and the evidence it gives your hiring team.

Talk to us