This is a living document tracking which tech companies allow AI coding assistants during their technical interview rounds, which prohibit them, and which actively evaluate the candidate’s AI-collaboration skill. Last reviewed: 2026.
Policies are based on publicly stated company guidance, recent candidate reports on Glassdoor and Levels.fyi forums, and recruiter-confirmed positions where available. Policies change; verify with your recruiter before any specific interview.
Tier 1: AI-permitted and graded for collaboration
These companies allow candidates to use AI tools openly and grade the candidate’s interaction with the AI as part of the rubric.
- Anthropic — Allows AI tools (including Claude itself) in coding rounds. Explicitly tells candidates the rubric is on AI direction, verification, and integration, not on unaided coding ability. Pioneered the “AI-collaborative” interview format.
- Cursor (Anysphere) — Allows and encourages use of Cursor during interviews, since the product is the company’s. Candidates without significant Cursor experience are at a disadvantage.
- Some Google teams — Specific teams (DeepMind, some Cloud AI groups) permit AI tools. Other Google teams do not. Confirm with your recruiter.
- Replit — Allows AI use, particularly within Replit’s own AI features.
- Glean — Allows AI tools and looks for candidates who can integrate AI output into clean enterprise-grade code.
- Vercel — Allows AI tools in coding interviews; the v0 product means candidates likely have AI fluency baked in already.
- Many YC startups (2024+ batches) — As a general pattern, post-2024 YC companies skew toward AI-permitted interviews because their internal engineering already assumes AI use.
Tier 2: AI-allowed but evaluated for over-dependence
These companies say AI use is permitted but watch for candidates who lean on the AI without verifying or adding judgment.
- OpenAI — As of 2026, individual teams vary. Some allow AI tools fully, others ask candidates to demonstrate unaided ability for foundational portions and allow AI for higher-level work.
- Stripe — Has shifted to allowing AI in some interview formats while keeping the bar on the candidate’s understanding of the underlying problem.
- Shopify — Allows AI use in line with the company’s stated “AI-first” engineering culture; candidates who do not use AI tools naturally during the interview are read as out-of-date.
- Canva — Allows AI tools in some rounds; explicitly evaluates whether the candidate can verify AI output.
- Notion — Generally permissive on AI tool use during interviews.
- Linear — Permits AI tools in coding rounds.
Tier 3: AI-prohibited (foundational filter intact)
These companies prohibit AI tools during interviews, treating unaided coding ability as a foundational filter.
- Most Google product teams — Despite Google shipping AI products, the majority of product-team interviews remain AI-prohibited as of 2026.
- Meta — Generally AI-prohibited in coding rounds. Has internal AI tools but the interview format predates them.
- Apple — AI-prohibited.
- Amazon — AI-prohibited in coding rounds.
- Microsoft — Most teams AI-prohibited, with some experimentation in newer AI-focused groups.
- Netflix — AI-prohibited.
- Most quant firms — Jane Street, Citadel, HRT, Two Sigma, Optiver, SIG, Akuna, IMC, DRW, Jump Trading, Tower Research — all AI-prohibited. The probability and combinatorics work that quant interviews emphasize is precisely the kind of reasoning where unaided ability matters most.
- Most investment banks (Goldman, Morgan Stanley, JPMorgan, Citi) — AI-prohibited; regulatory considerations also discourage even discussing AI tools in detail.
- Defense primes (Lockheed, Northrop, RTX) — AI-prohibited; security-clearance environment makes AI use complicated.
- Most healthcare and regulated-industry employers — AI-prohibited or heavily restricted.
- Bloomberg LP — Generally AI-prohibited.
Tier 4: Take-home and async assessment policies
For take-home assignments, the picture is more complicated:
- “Use whatever tools you normally would”: the most common take-home policy in 2026 across mid-tier tech companies. Effectively AI-permitted.
- “No AI tools during this assignment”: some take-homes still specify this, but enforcement is impossible.
- “AI tools are allowed; we will discuss your code in a follow-up live session”: increasingly common. The follow-up live session is the actual evaluation. Stripe pioneered this format.
How to confirm a company’s policy
Recruiter screen is the right time to ask. Use language like: “What is the company’s policy on using AI coding assistants during the technical rounds?” Most recruiters will give a clear answer. If the recruiter does not know, that is itself a signal — the company’s interview process has not been updated for the AI era, and the practical default is whatever the individual interviewer decides.
How policies are likely to evolve
Three trends seem likely over the next 2-3 years:
- Convergence toward Tier 2 + Tier 1. Pure Tier 3 (AI-prohibited) is hard to enforce in remote interviews and increasingly out of step with how engineers actually work. Many companies will move into Tier 2.
- Voice / agent-driven evaluation. The next generation of interview format will likely involve the candidate directing an AI agent verbally, with the interviewer evaluating the candidate’s voice-mode prompting, verification, and debugging. This format does not yet have a name in 2026 but is being prototyped.
- The “no AI” filter will narrow. Foundational-filter use of AI-prohibited interviews will likely concentrate at quant firms, regulated industries, and a few big-tech outliers. The mainstream will be AI-permitted with grading on collaboration.
Frequently Asked Questions
Why do some AI labs prohibit AI in interviews?
Foundational-filter logic. An AI lab still wants to know if the candidate can reason about systems unaided, especially for research roles where the work involves building the next generation of AI tools, not just using existing ones.
What if I’m a strong unaided coder but new to Cursor / Claude Code?
Spend two to three weeks getting fluent before AI-permitted interviews. The verification and integration skills transfer from your unaided coding skill, but the prompting habit does not — that takes deliberate practice.
If a company says “use whatever you’d use”, does that include external help (Stack Overflow, friends, Googling)?
Generally yes for searches and documentation; generally no for getting solution help from another human. Searching for syntax or library docs is normal engineering work. Asking a friend for the answer is not.
Is this list exhaustive?
No. It covers the most-asked-about companies and the major tier patterns. Smaller companies and individual teams within larger companies may diverge from the company-level tier shown here. Always verify with your recruiter.
How recent is this data?
Last reviewed in mid-2026. Policies change quarterly. We update this tracker as new public policy statements emerge or as candidate reports indicate a company has shifted tiers.