LinkedIn Coding with AI Interview: 4 Experiences

Shared by janebeingjane ·
Company
LinkedIn
Reports
4 interview reports
Level
Full-time
Stages
Onsite

LinkedIn's 2026 "Coding with AI" round gave each candidate one long problem in three to five parts, and nobody coded every part in time.

At a glance

  • Process: one AI coding round at the onsite, in an editor whose agents edit files directly (03, 04).
  • Asked recently: one long, multi-part task; in 03 the AI proposed multi-source BFS before a billing rule was added.
  • Difficulty: no one finished: 01 coded two of three tasks, 02 talked through part 4, 03 never reached part 3, and 04 stopped mid part 3. 04 lost time to planning and long agent replies; 02 waited for an interviewer who never came.

Coverage: Four experienced candidates changing jobs, all at the onsite stage of full-time LinkedIn loops. Three were software engineers: an infrastructure engineer in Bellevue with about a year of experience, a mid-level infrastructure engineer in the Bay Area, and a Senior Staff engineer. The fourth interviewed for Staff AI Engineer. Interviews fall in October–November 2026, and the reports went up between October 27 and November 12, 2026.

Software engineers and the AI engineer are covered separately:

  • Software engineers (reports 01–03): how many parts each problem had and how far each candidate got, plus the one problem described in full: a fire-station grid with a cost follow-up.
  • Staff AI Engineer (report 04): a repo of a few files in an editor with coding agents built in, and how a plan-first habit ate the clock.

Software engineers: LinkedIn's AI coding round

What came up in the software engineer AI coding rounds

All three software engineers describe the same shape: one problem, several parts, and a last part that never got coded. Only the Senior Staff report says what the problem was.

Topic Reports Last seen
One problem split into several parts (three or four reported) 3 of 3 Nov 2026
The last part only talked through or never reached 3 of 3 Nov 2026
Two interviewers scheduled, one never joined 1 of 3 Nov 2026
An AI that edits the code itself, so nothing is copied and pasted 1 of 3 Nov 2026 (report date)
Grid shortest paths around lakes, then a cost per home by distance 1 of 3 Nov 2026 (report date)
A follow-up on the memory used by the AI's code 1 of 3 Nov 2026 (report date)

A candidate can count toward more than one row.

Questions we found in the software engineer rounds

AI Coding

Question Reports Last seen
Fire stations, lakes and homes on a grid: each home's distance to the nearest station, then a distance-based cost per home with a cap; a third part wasn't reached (Fire Station Response Bill) 1: 03 Nov 2026 (report date)

Reports 01 and 02 don't describe their problems.

3 detailed candidate timelines, loops, and screens

01: Mid-level infra, Bay Area: three tasks, two written

Interview date: Last onsite round Nov 6, 2026 · Report date: Nov 12, 2026 · Role: Infrastructure engineer, mid level · Location: Bay Area · Result: Still waiting

Onsite: AI coding

  1. The round had three tasks. The candidate wrote code for two and explained the third out loud.
  2. Looking back, they thought the AI round went okay.

This is a short account and doesn't say what the tasks were.


02: Bellevue infra, one year in: four parts and an empty interviewer seat

Interview date: Onsite finished Friday, Nov 6, 2026, about a week later than planned · Report date: Nov 11, 2026 · Role: Software engineer, infrastructure team · Location: Bellevue · Experience: About one year · Channel: Applied online · Result: Waiting for a decision

The onsite was five rounds; the candidate names coding, system design, hiring manager and AI coding. One interviewer couldn't make the original date, which pushed the last round back.

Onsite: AI coding

  1. The round was supposed to have two interviewers. One never showed up, and the start of the hour went to small talk while they waited.
  2. The problem came in four parts and was long. The candidate discussed part 4 but didn't finish coding it, and that left them nervous about the result.

The rest of this candidate's onsite, including a first-ever system design round, is in 2026 LinkedIn Experienced Infrastructure Software Engineer Onsite.


03: Senior Staff: the AI proposed multi-source BFS, then a billing rule

Report date: Nov 10, 2026 · Interview period: 2026 Q4, five rounds over two days · Role: Senior Staff software engineer · Degree listed: PhD · Channel: Headhunter · Result: Offer

Onsite: coding, with AI allowed

  1. Setup: the AI was allowed to write the code, and the candidate chose a model they call "GPT 6.1 Sol." The AI changed the files directly, so there was no copying code back and forth. The candidate worked in Go.
  2. Part 1: a grid where 0 is a fire station, -2 is a lake and -1 is a home. There are several stations. For every home, find the distance to the closest station, and trucks can't drive through lakes. FastPrep's practice version is Fire Station Response Bill. It folds parts 1 and 2 into one function and uses its own cost table.
  3. The candidate's first idea was a BFS from each station, keeping the smallest distance per home. Asked about complexity, they said O(N) for a fixed number of stations, where N is the grid size, and O(M·N) if the M stations vary too. The interviewer didn't comment.
  4. Next they told the AI that plan and asked for something better. The AI suggested multi-source BFS: put every station in the queue at distance 0 and run one BFS, which is O(N). The candidate agreed, the AI wrote it, and the candidate read the generated code line by line.
  5. Part 2: the city pays for each truck trip, and the price depends on distance. Distances 0 and 1 get a discount off a base price, distance 2 pays the base price, and from distance 3 on a percentage is added, up to a cap. The candidate didn't remember the exact rates. They talked it through with the AI and let it write the code.
  6. The interviewer asked about space. The candidate said O(N): the code kept adding to the queue and only moved an index forward instead of removing items, so the whole grid ended up in the queue. They told the AI to drop each item once it was visited, by slicing it off the front of the queue in Go, and the AI made the change.
  7. There was a Part 3, but time was short and the interviewer didn't move on to it.

The other four rounds were two design rounds and two behavioral rounds, one of them with the hiring manager, a director. The candidate thinks the second design round went badly, but the offer came anyway.


Staff AI Engineer: LinkedIn's Coding with AI round

What came up in the Staff AI Engineer round

The single AI engineer report describes the setup but not the problem: a small repo, an editor with several coding agents, and four or five parts.

1 detailed candidate timeline

04: Staff AI Engineer: plan-then-execute cost more time than the problems did

Report date: Oct 27, 2026 · Interview period: 2026 Q4 · Role: Staff AI Engineer (machine learning) · Degree listed: PhD · Channel: Headhunter · Result: Offered Senior instead of Staff, and declined

The onsite was five one-hour rounds: AI modeling and fundamentals, AI system design, a data coding round, a leadership round, and "Staff Coding with AI."

Staff Coding with AI (1 hour)

  1. The editor had a range of coding agents built in. The candidate got a repo with a handful of files.
  2. The problem had four or five parts. They got about halfway through part 3.
  3. None of the parts were hard, and the candidate knew how to solve each one.
  4. What slowed them down was their usual Claude Code routine: plan first, then execute. Here that added a lot of overhead, and the agent's replies were so long that there wasn't time to read them. In hindsight they think a simpler model and a faster, more direct approach would have done better.

Patterns across the reports

  1. Nobody finished every part. This holds for all four reports, across both roles. Report 01 coded two of three tasks, report 02 talked through part 4 without coding it, report 03 never reached part 3, and report 04 stopped halfway through part 3 of four or five. Two of them say where the minutes went, and the causes differ: report 02 waited for an interviewer who never came, and report 04 spent them on planning and on long agent replies.

  2. The agent lives inside the editor. In report 03 the AI edited the code itself, and in report 04 the editor came with coding agents already built in and a repo to work on. Reports 01 and 02 don't describe the environment.

Common questions

Can you finish the LinkedIn AI coding interview?

None of the four candidates coded every part (Patterns, point 1).

Keep practicing

AI coding rounds at other companies

More companies with AI rounds on FastPrep:

Across companies: 2026 AI Coding OAs at 15 Companies: 32 Reported Problems and 2025–2026 AI Coding Phone Screens and Onsites at 43 Companies: 76 Reported Problems.

Sources: Anywhere useful interview info can be found.

0 total reactions. Your reactions: none.

Comments

0

No comments yet — be the first to share what you know.