Which companies allow AI in coding interviews? A sourced candidate guide

What counts as an AI-assisted interview here

This guide covers candidates using AI to complete assessed engineering work. An employer using AI to shortlist applicants, transcribe conversations, or conduct an automated interview is a different category. An engineering team using AI at work also does not establish permission during hiring. The useful question is whether the candidate may use a tool in the round they are about to take.

We distinguish live coding rounds, online assessments, and longer projects because their preparation differs. We count a company when a first-party policy, engineering account, or identifiable leader confirms an actual permitted stage. We do not count an unspecified exception clause as proof of an operating format. This is a researched selection, not a census of employers or a guarantee that every role has the same process.

Companies with documented live AI coding formats

These employers have direct evidence of candidate AI use in technical interviews. The sources establish a format or practice; your invitation establishes whether you will receive it. Read the linked company guide for a preparation exercise and the limits of the evidence.

• Meta: its current hiring FAQ describes built-in AI in many technical interviews and prohibits outside AI tools. Official hiring FAQ. Read the Meta guide.

• Canva: its engineering team documents a dedicated AI-Assisted Coding interview. Engineering announcement. Read the Canva guide.

• DoorDash: its engineering hiring announcement describes a 60-minute AI-assisted coding session on the candidate’s machine. Engineering interview announcement. Read the DoorDash guide.

• CoderPad: its employer policy permits transparent AI use and expects Product and Engineering candidates to demonstrate it. This is about hiring at CoderPad, not every employer using its platform. Candidate policy. Read the CoderPad guide.

Optional AI, selected rounds, and leadership accounts

Omnea permits AI during pair programming, including full coding agents, while also accepting candidates who choose to work without it. Omnea engineering account, May 26, 2026. Read the Omnea guide.

Bumble says some software engineering interviews include an announced AI collaboration exercise. Its default live-interview rule otherwise prohibits real-time AI responses. Bumble candidate policy, updated July 24, 2026.

Shopify’s Head of Engineering has described candidates using AI during interviews. This is direct leadership testimony; it does not establish an identical loop for every role. Farhan Thawar interview. Read the Shopify guide.

Vercel CTO Malte Ubl describes both AI-required and no-AI interviews in an event recording. Lovable recruiting leader Tim Duong describes an agentic coding interview in a September 10, 2026 talk. These direct accounts confirm practices without supplying full candidate rules. Vercel leadership recording, Lovable recruiting talk.

For preparation, treat optional use as a choice to explain. Decide which task benefits from assistance and what you can complete manually. If the exercise is explicitly designed to assess AI collaboration, rehearse that collaboration rather than planning to avoid the tool altogether. These are our recommendations, not a shared employer rubric.

Companies allowing AI in projects or take-home assessments

PostHog expects AI use in its paid SuperDay project, but its engineering handbook prohibits AI during the debugging session. Engineering hiring handbook. Read the PostHog guide.

Anthropic explicitly permits AI for its documented performance-engineering take-home. Its general candidate policy still requires separate permission for other assessments and live interviews. Performance-engineering account, general candidate guidance.

Nitro encourages AI for Engineering take-home tasks, with disclosure and an explanation of the work; live follow-up interviews are unaided. Nitro candidate policy.

Practice the handoff as well as the implementation. Close the assistant and explain one important branch, one rejected approach, and one failing test you repaired. Then change a requirement manually. This rehearsal exposes gaps that a working final repository can conceal. Keep a short decision record while working so your explanation is grounded in what actually happened.

Why Amazon, Microsoft, and Stripe are outside this list

Amazon confirms an assessment with an interactive AI assistant, but the announcement does not establish a standard AI-assisted live coding interview. We have kept it outside this interview directory. Amazon’s assessment announcement.

Microsoft’s published code of conduct permits interview assistance only when explicitly authorized. That exception alone does not confirm an operating AI coding round. Microsoft hiring guidance.

We did not establish Stripe’s current candidate AI format from a sufficiently clear first-party source in this review. Unconfirmed is an evidence status, not a claim that the company prohibits AI everywhere. Search results and candidate stories can identify questions to investigate, but we do not promote them into company-wide rules.

When two accounts conflict, compare the role, interview stage, source date, and tool being discussed before choosing one. A no-AI phone screen and an AI-enabled project can coexist. So can a pilot in one team and a different process elsewhere. Preserve that scope instead of compressing the evidence into a single yes-or-no label.

Turn the invitation into a concrete tool policy

Before preparing the environment, write down the stages in your own process and the permission attached to each one. Record the permitted tool, whether an account is supplied, what material can be shared with it, and whether prompts must be visible. If any field is missing, ask the recruiting contact to resolve that specific uncertainty. Avoid asking only whether the company is “AI friendly.”

A useful clarification request names the round and the tool you intend to use. Keep the response with the invitation. If the interviewer changes the instructions on the day, confirm the new expectation before starting a tool. This prevents an assumption from becoming the most consequential decision in the session.

For my upcoming coding round, could you confirm:
1. Is candidate AI use allowed, optional, or expected?
2. Must I use an integrated assistant, or may I use my own editor and agent?
3. Which task materials may I share, and should I show my prompts?
4. Does the permission also apply to debugging and follow-up rounds?
5. What should I do if the tool or environment fails?

Choose a rehearsal that matches the assessed work

For a live AI round, ask a partner to supply a small unfamiliar repository and observe your session. Begin with an example of expected behavior, inspect the relevant files, and ask the assistant for one bounded change. Pause to explain why you accept or reject it. Reserve time to run a regression and describe what remains unverified. The partner should interrupt once with a requirement change.

For a project or take-home, add reproducibility. Start from a clean checkout, document the setup command, and verify that another person can run the result. Keep the final diff smaller than your first draft if necessary. A new dependency or abstraction should solve a stated problem; it should not survive merely because the assistant introduced it.

For an unaided stage in a mixed process, switch AI off during practice. Diagnose a small failure with the debugger or tests, narrate the evidence, and implement the correction yourself. Practicing both modes makes the transition less distracting and helps distinguish a gap in engineering knowledge from unfamiliarity with a tool.

Use the company guides as preparation, not question predictions

Each linked nrml guide separates documented employer statements from our original preparation exercises. The exercises are designed to train useful behaviors, such as finding a counterexample or defending a small patch. They are not leaked questions, reconstructions of private assessments, or claims about an unpublished scoring system. Choose one exercise because it addresses a skill you need, rather than because its story resembles the employer’s product.

After a rehearsal, write three sentences: what you initially believed, what evidence changed your mind, and what you would do differently next time. Then repeat the weak part on a different repository. That produces a more useful preparation record than counting generated lines, memorizing prompts, or attempting to reproduce somebody else’s reported interview.

For concrete starting points, read the published interview examples. Try Rebuild State from Out-of-Order Events to work through an event-driven feature, or Stop Double Claims in the Job Scheduler to practice ownership and concurrent work. Each exercise has its own brief; neither is presented as a company’s private question.