In an AI interview, some companies fail you for opening Copilot

The research tracks candidate loops at most of the companies on this list. The question that comes up most right now isn't about dynamic programming. It's: "am I allowed to use AI in this interview?"

Here's the uncomfortable truth. The industry has not reached consensus. Two companies I track will grade you on how well you use AI tools. Seven will disqualify you for touching them. Four have said nothing publicly. And the irony writes itself — some of the companies that build the most powerful AI tools in the world are the ones banning AI hardest from their own interview rooms.

So I stopped giving one generic answer and started mapping it company by company. What follows is the map.

---

The three camps

Diagram

Copy source

Rendering diagram...

The first branch decides your whole prep strategy. Prep for an AI-embracing company and you need to practice with the tools. Prep for an AI-prohibited company and you need to practice without them. A generic prep mode is underprepared in at least one direction.

The real question here is not whether AI is "fair" in interviews. That ship sailed. The real question is whether you know what THIS company is measuring — and then whether you can rehearse that exact mode.

---

Camp 1: AI-EMBRACING

DoorDash — AI is the interview

DoorDash replaced the traditional coding round with a 60-minute AI-assisted working session. You bring your own IDE (Cursor, Claude Code, VS Code + Copilot — free tier is fine), you join a realistic project, and you start contributing.

The interviewer is not watching you write from scratch. They are watching you ramp into an unfamiliar codebase with an AI at your shoulder. Source: DoorDash Engineering puts the grading dimensions in plain English: getting unblocked in an unfamiliar domain, validating with rigor (you read logs, you write checks, you don't blindly trust AI output), and communicating the why of your decisions.

The common failure mode: candidates read the format, panic, and rehearse prompt tricks instead of practicing the workflow. Prompt poetry is not what DoorDash grades. The strong candidate opens the repo, spends 90 seconds reading, writes a one-sentence plan, then uses the AI for narrow subtasks they could've done themselves but don't want to type from memory.

What this means for your prep. Practice in a real IDE with a real AI coding agent on a real multi-file task. Not a LeetCode widget. If you have never used Cursor or Claude Code in anger, you are not ready for DoorDash.

Meta — the model chooser in the IDE

In October 2025 Meta started piloting an AI-enabled coding round that replaces one of the two onsite coding interviews. It's a CoderPad environment with a chat sidebar and a model dropdown. Source: Hello Interview lists the confirmed models available in the round: GPT-4o mini, Claude Haiku 3.5, Llama 4 Maverick. You can switch between them mid-interview.

The critical detail: Source: interviewing.io consistently shows Meta is testing whether you can use AI without blindly trusting it. The candidates who fail this round are not the ones who avoid AI — they're the ones who accept every suggestion without reviewing it. Meta wants to see that you treat the AI as a tool, not as an oracle.

The nuance: this is one specific coding round. Meta's other rounds (behavioral, system design, the other coding round) don't have an AI component. And this pilot is getting rolled broader through 2026, but don't assume it's universal yet — ask your recruiter which loops apply to your role.

What this means for your prep. Practice with AI assistance, but force yourself to catch the AI's mistakes before you run the code. If you never turn the AI off in practice, you'll never develop the review instinct Meta is grading.

---

Camp 2: AI-PROHIBITED

Amazon — banned across the board

Amazon explicitly prohibits AI tools during every stage: HackerRank OA, Livecode phone screen, every onsite round. Amazon's Livecode environment doesn't even have code execution — you can't run your own code, let alone an AI's. A strong candidate just types, reasons out loud, and hand-simulates.

Google — banned, and restructuring around the ban

Google explicitly prohibits AI tools in interviews. But there's a deeper story: Google's 2026 shift back toward in-person interviewing is partly motivated by concerns about AI-assisted cheating in remote settings. They're not just banning AI — they're restructuring the interview format to make AI assistance harder to use undetected.

You're already coding in a Google Doc with no IDE, no autocomplete, nothing. Adding AI was never on the table. The ban is explicit regardless.

OpenAI — the irony that writes itself

OpenAI — the company that ships ChatGPT, Codex, and GPT-4 — strictly prohibits AI during live coding interviews. This is arguably the single most ironic policy on the list. You're interviewing to build the most powerful AI coding tools in the world, and you're not allowed to use them to demonstrate your skills.

What this means for your prep. Practice raw coding with no autocomplete. OpenAI's progressive problems are custom and can't be found on LeetCode — you'll solve them with a text editor and your brain, nothing else.

Anthropic — Claude for prep, not for the interview

Anthropic has the most nuanced policy in the entire set. Claude is explicitly allowed for two things — refining your application materials after your own first draft, and general interview preparation. Source: Anthropic Careers spells that out.

But in live interviews the wording is unambiguous. Source: Anthropic candidate AI guidance: "This is all you–no AI assistance unless we indicate otherwise. We're curious to see how you think through problems in real time." The coding screen explicitly permits Google and Stack Overflow. It does not permit Claude, ChatGPT, or Copilot.

The strong candidate names this trade-off out loud in the first minute: "I understand Google is allowed but AI isn't — I'll narrate my searches and reasoning so you can see how I work." That framing buys you a ton of signal. The weak candidate doesn't say a word, gets caught glancing at another tab, and eats a red flag.

What this means for your prep. Use Claude to prep (Anthropic literally encourages this). Walk into the live interview ready to reason without it.

Netflix — named and banned

Netflix's interview guidance names coding assistants, Copilot, and ChatGPT specifically. No ambiguity, no exceptions. When a policy names the tools, you know it's been tested.

Microsoft — prep yes, live interview no

Microsoft draws a pragmatic line: AI tools are encouraged for interview preparation, prohibited during live assessments unless an interviewer explicitly permits them. Their official guidance asks candidates to "demonstrate their own skills."

This is the most honest middle path. Microsoft ships Copilot — they're not anti-AI. They just want the interview to measure you, not your tools.

Stripe — written ban across the live loop

Stripe has a consistently reported written ban on AI tools in all live interview rounds: coding, debugging, integration sessions. Internet access and docs may still be allowed depending on the round. AI assistants are explicitly excluded regardless.

---

Camp 3: NO POLICY

Airbnb — silence

Airbnb has not published an explicit public policy on AI tools during interviews as of April 2026. Their interview rounds use CoderPad (runnable code), and any AI restriction is inferred rather than publicly stated.

Default assumption. Don't use AI unless your recruiter explicitly says you can.

Apple — consistent with their secrecy culture

Apple has issued no public AI interview policy. Different teams run different loops — a universal Apple-wide policy would be hard to enforce anyway. Interviewers have reportedly shifted toward more real-world, team-specific questions, which naturally makes AI overlay tools less useful.

Default assumption. Ask your recruiter directly. Apple's recruiter is the single most reliable source for interview prep logistics.

Databricks — practically restricted, not formally banned

Databricks has no explicit published AI policy. Their assessment environments effectively limit outside help, and some candidate reports mention "no Google or AI" during assessments — but that's an unstated practical restriction, not formal public policy.

Default assumption. Don't use AI. The environments are designed to prevent it regardless of policy.

Uber — heavy internal AI adoption, public silence

Uber uses AI extensively internally but has published no policy on AI use during interviews. Candidates are advised to assume AI is not permitted in live rounds unless a recruiter says otherwise. That's caution, not policy.

Default assumption. Assume prohibited. If Uber wanted you to use AI in interviews, they'd say so.

---

Where this fits in the broader industry shift

To be clear, this isn't a fringe question. Source: CoderPad 2026 State of Tech Hiring reports that 82% of developers now find GenAI at least somewhat useful in their actual work (up from 76% in 2025), and 54% say they'd take a measurable productivity hit if AI tools were removed tomorrow. Technical assessments are up 48% globally versus mid-2023 — hiring teams are interviewing more, under pressure, and they're figuring out in real time whether AI belongs in the room.

Canva was the first household name to flip fully pro-AI. Source: Canva Engineering made the case explicitly in June 2025: their engineers use AI assistants daily at work, so their interviews should measure how well candidates collaborate with AI, not whether they can pretend it doesn't exist. Source: The Register covered the switch the same week. Canva's own interviewer guide (Source: Canva Engineering) spells out the grading rubric: clarifying questions, strategic AI use, critical review of AI-generated code. It's basically the DoorDash rubric a year earlier.

DoorDash and Meta followed. Google is going the opposite direction. That divergence is the story. It's why a generic "practice with ChatGPT open" prep plan fails at both ends of the spectrum.

---

The full map

---

What this means for you

Three rules for 2026 interview prep.

1. Research the policy before you practice. Prepping for DoorDash or Meta? Practice with AI tools. Prepping for Google or OpenAI? Practice without them. A single "generic" prep mode is underprepared in at least one direction.

2. "No policy" means "no." If a company hasn't published an explicit AI policy, assume AI is prohibited. The risk is asymmetric — using AI when it's allowed costs you nothing, and using AI when it's not allowed costs you the interview. Default to the safe side, and ask your recruiter if you need clarity.

3. The trend is splitting, not converging. A year ago, most of us (including me) assumed the industry would converge on AI-assisted interviews. It hasn't. Instead, we're watching a split harden: companies that build AI into the interview format (DoorDash, Meta) versus companies that are actively restructuring interviews to make AI harder to use (Google's return to in-person, Netflix's named bans, OpenAI's strict no). Both ends are committing harder to their position, not drifting toward each other.

The strong candidate names the trade-off out loud. "I know AI is banned in this round — let me show you how I reason through this from first principles." Or: "I know AI is required here — let me narrate what I'm choosing to offload and what I'm keeping in my own head." Either framing turns an awkward question into interviewer signal.

The bottom line: there is no "standard" AI interview policy in 2026. Every company on this list has made a different choice, and those choices are structurally different enough that your prep strategy needs to account for them. Check the dossier before you practice.