The new interview map

What rounds will you actually face now, and who says so?

Until recently, a software interview meant algorithms, system design and behavioural questions, with AI strictly banned. Now two things have changed.

First, some companies put AI in the room and score how you use it. Second, teams building AI products added rounds of their own: retrieval, agents, evals, and sometimes a role-play with a customer.

Details change quickly and many companies don't publish them, so this page labels who said what. A company's own post is strong evidence. A prep vendor's guide based on candidate reports is useful, but it could be out of date. Your recruiter's answer beats all of them.

  • AI-enabled round: A live coding round where an assistant is provided or allowed. You're scored on your judgment in using it.
  • Code comprehension round: You work on code you've never seen: reading, debugging, improving.
  • Applied AI engineer: An engineer who builds products on top of models, rather than training them.
  • Loop: The full set of interviews for one job.

Steps:

  1. Find your loop. Ask the recruiter for each round's name, length and tools.
  2. Match each round to a lesson. The next seven lessons cover one round each.
  3. Score yourself on each. This feeds the study plan in the next path.
  4. Check for updates. Formats change. Re-check what you read here against what you're told.

What's been reported (Checked in November 2026. Only Canva's row comes from the company itself. Treat the rest as reports, and confirm them with your recruiter.)

Who Round What's reported Source
Canva AI-assisted coding Candidates are expected to use tools such as Copilot, Cursor or Claude. It uses realistic, ambiguous problems. Scored on using AI well, breaking down requirements, technical decisions, fixing AI code, and production standards. The company, June 2025
Meta AI-enabled coding 60 minutes in CoderPad with an AI chat, replacing one coding round. Described as stages: fix a bug, build a feature, then optimise. Reported to be scored on problem solving, code quality, verification and communication. Prep vendors. Not confirmed by Meta.
Google Code comprehension A pilot using Gemini for select US teams hiring early and mid-career engineers, planned for the second half of 2026. Scored on AI fluency, including prompting, validating output and debugging. Press, citing an internal document. Sources disagree on the start date.
Shopify, Rippling Live coding Candidates may bring their own AI copilots. One press mention, no detail.
Applied AI roles Mixed Often 4 or 5 rounds: screen, LLM phone screen, take-home or live coding, system design, sometimes a customer conversation. Prep blogs. Varies widely.
Eval-focused roles Evals A 2 to 4 hour take-home to build a small harness, plus eval system design and eval data rounds. One blog, no sources given.

Prompt, Ask your recruiter about the rounds:

Could you share the rounds in my loop, with each one's length and format? For any coding rounds: what tools are allowed, which languages, and will I work in an existing codebase or start from scratch? For any design rounds: is there a particular product area I should expect?
  • Failure, Trusting a prep site's rubric as official: Vendors report what candidates told them. It's often right, but it isn't a promise.
  • Failure, Preparing for yesterday's loop: A guide from a year ago may describe a round that's changed. Ask.
Why might a Meta guide be wrong even when it's written carefully?
Meta hasn't published the details, so guides rest on candidate reports and can be out of date or differ by team.
What's the most reliable source on your own loop?
Your recruiter. Ask for each round's format and what tools are allowed.

Sources: