CodeSignal and CoderPad AI Assessments: Candidate Guide

If a company sent you a CodeSignal or CoderPad link, the test may let you use AI. Both vendors now build an AI assistant into the editor. What their marketing pages rarely say to candidates is the other half: your conversation with that assistant is saved, and the employer can read it. This guide puts each vendor's AI modes and what the employer sees side by side, from the vendors' own documentation. For how AI rules differ across companies, see our AI coding interview overview.

AICodingRound is not affiliated with CodeSignal or CoderPad. We have no access to their tests. Every page cited below was opened on October 29, 2026; vendor docs change, so check the date on each.

Practice the skill these tests record. On AICodingRound you work real tickets in multi-file apps with your own Claude or Codex as a coach, and the report includes how you used the AI. Try Save a cart for later (medium, TypeScript, about 50 min). The beta is free and places are limited: get a place.

First, find out what you are walking into

The invite email or the test URL tells you the vendor. It does not tell you the setup, and on both platforms the company controls it. Before the test, ask your recruiter three questions:

1. Is AI allowed, and if so, which assistant: the built-in one only, or an agent like Claude Code or Codex?
2. Can the AI edit my code, or only answer questions?
3. Will the interviewer review my AI conversation afterward?

A clear answer saves you from guessing during the test.

CodeSignal: Cosmo and agentic assessments

AI-assisted assessments. On May 28, 2025, CodeSignal launched AI-assisted coding assessments and interviews built around Cosmo, its assistant "embedded directly in the IDE" (CodeSignal). The press release names two modes:

• Full AI Co-Pilot Mode "enables developers to tackle advanced challenges with real-time AI collaboration."
• Guided Support Mode "provides limited AI help focused on platform navigation, syntax, and documentation."

Both are available "for both Assessments and Interviews." CodeSignal's announcement post says Full Co-Pilot makes some tasks possible that would be unsolvable in the time allowed without AI (CodeSignal blog). So if you get Full Co-Pilot, expect a task sized for it. If you get Guided Support, expect to write most of the code yourself.

What the employer sees. The press release says: "Alongside session replays, hiring leaders gain access to transcripts of all AI interactions." The announcement post says every AI-assisted assessment or interview "comes with a full transcript of candidate-AI interactions, along with a session replay" (CodeSignal blog).

Agentic assessments. CodeSignal also sells assessments where candidates work with "agentic tools like Cursor, Claude Code, and Codex." Its data sheet says they give "a strong signal of skill from how candidates extract requirements, build a solution, and explain their reasoning" (CodeSignal). On October 29, 2026, the pricing page listed "Agentic Assessments with Claude & Codex" on its Build plan and did not mention Cursor (CodeSignal pricing). The pages we opened do not say what the employer can review from an agentic session, how long one runs, or how it is scored. Ask.

CoderPad: AI Assist in interviews and Screen tests

CoderPad has two products candidates meet: Interview (a live pad with an interviewer) and Screen (a test you take alone).

In a live Interview pad. CoderPad's docs say "AI Assist is available on all pads by default"; the company's setting only decides whether it starts on or off, and the interviewer can switch it per pad (CoderPad docs). The models listed are GPT, Claude, Gemini and Llama. AI Assist has three modes:

• Ask: the AI answers but has no write access to your code.
• Edit: the AI can change code, including the file structure.
• Plan: the AI outlines a step-by-step approach before answering.

In multi-file languages, Claude Code and Codex can appear as agent options, but the docs say they "are currently only available for Enterprise plans."

Two lines in the same doc matter most to candidates. First: "All pad users' prompts and AI output will be saved for review in playback mode." Second, there is a Private AI Assist that is "not visible to the candidate." It is for the interviewer, who can use it to analyze your code and generate follow-up questions.

Playback. CoderPad "keeps track of every keystroke of every participant within a pad" and the interviewer can replay it (CoderPad docs). A yellow dot marks each paste from outside the pad; the doc notes that copying from the AI Assist tab can also trigger it. Orange dots mark times your browser was not focused on the editor. Code you run in the shell does not appear in playback. Separately, CoderPad alerts interviewers when a candidate "Pastes code from an external source" or "Leaves the IDE (e.g., tab-switching)" (CoderPad docs).

In a Screen test. AI Assist is "only available for Project questions," and the company turns it on in the test settings. In the report, "All of the candidate's prompts and AI output will be saved for review in playback" (CoderPad docs). CoderPad's page for employers sums up its approach: "allow AI use, evaluate how it's used; monitor with code playback, proctoring, and similarity checks." Its integrity tools include "AI-assisted webcam proctoring" and "Code similarity checks, IDE exit tracking, randomized questions" (CoderPad).

Side by side

What this changes about how you work

On both platforms, assume the AI conversation is part of your answer. An interviewer reading it sees whether you understood the task before you asked, whether you checked what came back, and whether you could have done the work without it.

A few habits that read well in a transcript:

• State your plan before you ask. "I think the bug is in how the total is rounded; show me where the total is computed" reads better than "fix this."
• Check, then say so. After the AI suggests something, run the tests or a quick input, and note what you saw. The replay shows the run; your note shows the reasoning.
• Push back when it is wrong. A transcript where you catch the AI's mistake is strong evidence. One where you accept everything is not.
• Keep the scope small. Ask for one function or one explanation at a time. Large pasted answers are hard to defend in the follow-up.

Practice for an AI-allowed test on AICodingRound

A transcript reads well when you understood the code before you asked and checked what came back. That skill comes from working in code you did not write, with an AI you have to check, and it is what we built AICodingRound for. You get multi-file apps in Python, TypeScript and SQL, tickets written like real bug reports, automatic checks on your work, and a short debrief about what you did. You bring your own Claude or Codex as a coach: it reads code and explains, but it does not write the fix. The report gives a level per skill (Strong, Solid or Developing), including how you used AI. If your test is agentic, also spend a session letting Claude Code or Codex edit files, since our coach does not.

A practice set that matches these formats:

• Save a cart for later (medium, TypeScript, about 50 min): extend an existing codebase from a short spec.
• Refund reasons you can correct later (medium, Python, about 45 min): a feature with a webhook.
• Stop rescheduling from regressing (medium, about 40 min): write tests that catch a regression.
• A forwarded password-reset email still changes the password (medium, about 40 min): a bug from a customer report.

Run the transcript habits above on each one. The beta is free and places are limited: get a place. If your test is on HackerRank instead, read our HackerRank Chakra guide, and for the round format in general, our guide to the AI coding round.