Meta's loop changed more between October 2025 and early 2026 than in the five years before. One of your coding rounds now hands you an AI assistant and grades how you use it, and the behavioural round alone can drop you from E5 to E4. Practise both with a senior interviewer on Intervue before your onsite.
Book a Meta-style mock interview → E5 or E6? Practise design + behavioural →
Meta runs a recruiter screen, a 45-minute technical phone screen in CoderPad, then a virtual onsite. What happens on the onsite depends on your level.
[[table]]
Level | Coding | Design | Other
E4 | Standard + AI | 1 round | Behavioural
E5 | Standard + AI | 1 deeper round | Behavioural
E6 | Standard + AI | 2 rounds | Behavioural, leadership
[[/table]]
"Standard + AI" means one traditional coding round and one AI-assisted coding round. E4 and E5 loops usually end with an unscored team-matching chat that still feeds the decision.
An online assessment shows up mainly for E5 and above, though some E4 pipelines include it too. E6 candidates also get a leadership assessment between the phone screen and the onsite.
[[mock-interview-banner]]
The Meta AI-Assisted Coding Round
This is the biggest change to Meta's loop in years, and the round most candidates have never practised. Starting in October 2025, Meta began replacing one of the two traditional onsite coding rounds with a 60-minute round where you get an AI assistant. Through 2026 it has rolled out across back-end and ops-focused roles.
Who gets it
[[table]]
Level | Coding rounds on the onsite
E3 to E6 | 1 traditional + 1 AI-assisted
E7 and above | 1 round, AI-assisted
[[/table]]
The rollout isn't complete everywhere. Ask your recruiter which format your onsite uses, and ask for the practice CoderPad environment with AI enabled. Recruiters can share it, and it's the closest thing to a dress rehearsal.
What the round looks like
You work in CoderPad with a chat sidebar that gives you a choice of AI models, and you can switch between them during the round. The task feels like real work instead of a puzzle. Several candidates describe getting a small existing project with multiple files, then being asked to understand it, fix or extend part of it, and reason about the result.
Candidates and prep guides describe the hour in roughly three phases:
- Understand. Read the code and the task. Ask clarifying questions. Decide what you'll do yourself and what you'll hand to the AI.
- Build. Use the AI for focused pieces, like a helper function, a test or a boilerplate parser, while you own the overall design.
- Verify and explain. Check what the AI wrote, run tests, fix its mistakes, and talk through runtime, edge cases and trade-offs.
The format deliberately mixes coding with non-coding questions. Interviewers may ask why you'd change a function signature, what the runtime is, or which data structure you'd swap in and why.
[[mock-interview-banner]]
How it's graded
You're graded on how you use the AI, and the AI's output matters far less. Interviewers look for:
- Problem decomposition. You break the task into pieces before prompting, instead of pasting the whole problem in.
- Judgement. You catch wrong or sloppy AI output and say why it's wrong.
- Verification. You test what gets written, including edge cases the AI missed.
- Communication. You explain your choices as you go, the same as any Meta coding round.
- Ownership. By the end, you can explain every line in the solution.
One E5/E6 candidate on Glassdoor listed unfamiliarity with the AI round as one of the reasons the loop didn't go their way. Most people's first attempt is slow and messy, because prompting under time pressure while someone watches is a separate skill.
How to practise
Run this loop several times before your onsite. Take a small multi-file project, write a task for yourself, prompt an AI for one piece, review the output line by line, find what's wrong, fix it, and explain the fix out loud. Time yourself to 60 minutes. Then do the same with a person watching, which is the part you can't simulate alone.
Good habits to build:
- Prompt for small, specific pieces with clear inputs and outputs.
- Read generated code before you run it, and say what you're checking.
- Write at least one test of your own that the AI didn't suggest.
- Keep a running note of decisions so you can explain them at the end.
The AI-assisted round rewards judgement under pressure, which only improves with live practice. Do a Meta-style AI-assisted coding mock with a senior engineer →
Meta Levels and Pay
Average yearly pay for Meta software engineers in the US, in thousands of dollars:
[[table]]
Level | Base | Stock | Bonus | Total
E3 | 146 | 33.6 | 6.1 | 186
E4 | 189 | 88 | 16.6 | 293
E5 | 225 | 179 | 22.1 | 427
E6 | 272 | 381 | 45.4 | 699
[[/table]]
These are Levels.fyi averages, updated August 2026. Meta stock vests evenly over four years, 25% a year. Pay in London runs lower, at about £143K for E4 and £272K for E5.
Meta E4 Interview
E4 is Meta's mid-level band, usually for engineers with 2–5 years of experience. The onsite has one traditional coding round, one AI-assisted coding round, one system design or product architecture round, and one behavioural round.
The traditional coding round moves fast. Expect two LeetCode-medium problems in 45 minutes, so you have about 20 minutes each, including talking through your approach and stating complexity. Arrays, strings, trees, graphs and hash maps come up most.
For design at E4, expect a contained product system such as a news feed, a messaging feature or event invitations. Interviewers want clear requirements, a sensible data model and an honest discussion of trade-offs. They aren't expecting org-wide architecture.
Meta E5 Interview
E5 is Meta's senior band and where most experienced hires aim. The rounds look like E4, but each one goes deeper. Design interviewers expect you to drive the whole conversation: scope it, pick an approach, and defend it under pressure on scale and failure.
The behavioural round matters most at E5. Meta's interviewers push hard on whether your stories show real senior scope or E4-level work described in bigger words. Behavioural performance alone can drop an E5 candidate to an E4 offer, even with strong coding and design.
Prepare four or five stories with clear scope. Say how many engineers were involved, what you decided, what changed because of it, and what you'd do differently. Use numbers wherever you can.
Meta E6 Interview
E6 is staff level. Before the onsite there's an extra leadership assessment covering how you've led projects and people across teams. The onsite then has one traditional coding round, one AI-assisted coding round, an architecture round, a design round and a behavioural round.
At E6, design and leadership carry most of the decision. The architecture round looks at systems that span several teams, and interviewers want to hear how you'd sequence the work and handle disagreements between teams. Your behavioural stories should show influence beyond your own team, ideally across a whole organisation.
At E5 and E6, the behavioural round can decide your level on its own. Practise a Meta-style behavioural and design round with a senior engineer →
More prep: FAANG behavioural interview guide, FAANG system design mock interview and SDE 2 coding questions on graphs and DP.
FAQs
What is Meta's AI-assisted coding round? A 60-minute onsite coding round where you can use AI models in a CoderPad sidebar. It replaces one of the two traditional coding rounds, and you're graded on how well you direct, check and explain the AI's work.
Which levels get the AI-assisted round? E3 to E6 get one traditional and one AI-assisted coding round. E7 and above get a single coding round, and it's AI-assisted.
Can I practise the AI round before the interview? Yes. Ask your recruiter for the practice CoderPad environment with AI enabled.
What is the Meta E5 salary? Levels.fyi puts the US average at about $427K a year: $225K base, $179K stock and $22K bonus.
Can I be downlevelled from E5 to E4? Yes. A weak behavioural round can do it alone, even when coding and design go well.
How do I prepare for the Meta interview? Drill two-mediums-in-45-minutes coding, practise the AI-assisted loop on a real codebase, prepare design for your level, and write out four or five scoped behavioural stories. Then take a Meta-style mock interview to put it all together.
Get Ready for Meta: Take a Mock Interview
Meta's 2026 loop tests four separate skills: fast traditional coding, working with an AI assistant under time pressure, design at your level, and behavioural stories that prove your scope. A 1:1 mock interview with a senior engineer on Intervue lets you rehearse each one and ends with a written feedback report.
Book your Meta-style mock interview →
Still comparing options? Here's our list of the best mock interview platforms. Preparing for other senior loops? Read our Netflix interview experience and Google mock interview guide.
