Which coding agents can candidates use?

Candidates use the full Claude Code and Codex command-line tools, set up in the terminal of the VS Code environment, with no separate account or API key. AI usage is included without throttling. Candidates can also use the Codility AI Assistant. Allow or turn off AI for each task in Screen and Skills Intelligence, and for each interview in Interview.

Can we see how a candidate worked with the agent?

Yes. The report keeps every prompt, every change the agent made and the final code, with a replay of the session. Reviewers see the process and the result together.

Can we use our own take-home task?

Yes. Teams can add custom tasks. Confirm the required environment and scoring support with our team before launch.

Does an AI model score the candidate?

No. Fixed rules score task work, and interviewers score live interviews against a rubric.

Do you assess machine learning and AI engineers?

Yes, on the engineering work the role requires. Some tasks ask candidates to build an AI or ML system, such as a retrieval pipeline or an anomaly detector, within the time limit. For research roles, scope the assessment with our team before you select tasks.

Can business teams in our company be assessed too?

Yes. Business teams in your company can be assessed on practical AI skills, on the same platform. Candidates and employees complete a business task with an AI assistant, and your team reviews the result and the prompts behind it. Assess AI skills in business roles

Your engineers code with AI agents. Hire the ones who use them well

Codility gives candidates the real Claude Code and Codex, without throttling, in take-home assessments and live technical interviews. You review every prompt and every change the agent made.

Request a free trial

Talk to us

Validated assessments since 2009 SOC 2 Type II ISO 27001

PYtranscode.py PYframes.py PYtest_transcode.py agent session · recorded

21# frame header: magic u32, ts u64, len u16

22HDR = struct.Struct("<IQH")

23def frames(buf, off=0):

24 while off < len(buf):

25 if len(buf) - off < HDR.size:

26 break # skip the partial frame

26 raise TruncatedLog(off)

27 magic, ts, n = HDR.unpack_from(buf, off)

$ pytest -q

FAIL test_partial_frame 3 frames lost, no error

› raise on a partial frame. never drop data

PASS 14 of 14

Coding agent31:12

Caught the agent dropping frames without an error

Wrote a failing test for the partial frame, then told the agent to raise an error.

Your hiring process still tests how engineers worked before AI

01

Every candidate says they code with AI. Few can show how well they use it.

02

A take-home that doesn’t allow AI, or gives candidates a stripped-down assistant, tests a job your engineers no longer do.

03

A take-home done with an agent returns the final code and none of the decisions behind it.

See how each candidate works with the agent

Assess engineering skills

Give candidates the real Claude Code and Codex

Candidates run the full Claude Code and Codex command-line tools in the terminal of a VS Code environment, the same tools your engineers use every day. AI usage is included without throttling, with no separate account or API key. Allow or turn off AI for each task in Screen, and for each interview in Interview.

Terminal · claude

$ claude

✻ Claude Code

› add tests for list_orders

● Wrote tests/test_orders.py

✓ 4 passed

Terminal · codex

$ codex

>_ Codex

› why is list_orders slow?

● Ran EXPLAIN on orders

→ add an index on customer_id

AI usage includedNo throttlingNo API key

Claude Code session in the Codility assessment terminal: the candidate corrects the agent and asks for the failing test first, and the session is saved to the candidate record

Review how they directed the agent

The report keeps the code, every prompt and every change the agent made, with a replay of the session. An AI usage summary shows which tools each candidate used and how often, so reviewers know where to start. Some tasks plant mistakes in the AI’s output, so you see who catches them.

Screen for engineering

Run it live or as a take-home

Use Interview for a live session in a shared VS Code environment. Your interviewer and the candidate both see the agent work in the terminal, with a whiteboard, a rubric and a full session record. Use Screen when candidates work on their own time.

Explore Interview

Senior engineering hiring

Live in InterviewTake-home in Screen

Terminal · claudeSam, interviewerLeo, candidate

› add rate limiting to POST /orders

● Updated app/orders.py

✗ 1 failed: test_burst_limit

SamWhy does the burst test fail?

LeoThe limit refills every second. I will ask the agent to make it per minute.

Your take-homeSample

transcode-logs/

README.md

transcode.py

tests/test_transcode.py

AI AssistanceClaude Code and Codex

or choose from the library

Build a retrieval pipelinePython · AI system

Build an anomaly detectorPython · AI system

1,300+ validated tasks in 80+ languages and frameworks

Bring your own task, or use ours

Add the take-home task your team already uses, for example one where candidates plan, build and test a tool with an agent. Or choose from 1,300+ validated tasks in 80+ languages and frameworks, including tasks where candidates build an AI system such as a retrieval pipeline. Results reach your applicant tracking system, including Ashby and Greenhouse.

Explore integrations

No AI model scores your candidates

Assessment platforms now use AI to grade answers and run interviews. Your engineers know how confidently an AI model can be wrong.

Codility never lets an AI model score a candidate. Fixed rules score the work, and your team makes the call.

Explore the assessment methodology

Clear rules for AI in every assessment

90%

of candidates say the instructions for using AI were clear

Candidate feedback survey, October 2025 to April 2026, 34,532 responses

Codility publishes its assessment methodology in a Technical Manual. It follows the Standards for Educational and Psychological Testing, the benchmark psychologists use to judge whether a test is valid and fair. Professional Services helps your team design assessments around your own tasks and validate them on your roles.

Questions from engineering and talent teams

Start with one engineering role

Bring the role and the take-home you use today. We show you the agent session, the report and the evidence.