Snowflake AI Coding Interview, Experienced Hires: 3 Experiences

Shared by janebeingjane ·
Company
Snowflake
Reports
3 interview reports
Level
Full-time
Stages
Onsite

Snowflake's 2026 AI coding round had candidates design a rate limiter or a retrying task queue while the AI wrote the code.

At a glance

  • Process: one AI coding round at the onsite, next to a behavioral round (01, 02); the AI Engineer's onsite was in person on their own laptop (03).
  • Asked recently: design and implement a rate limiter (01, 03) and a retrying task queue (02).
  • Difficulty: neither engineer found the task hard, yet 01 was rejected and 02 ranked it the hardest of five rounds; 02's AI took about three minutes per answer. The AI Engineer passed and declined the offer (03).
  • Graded on: how the candidate prompted the AI and weighed trade-offs, as 02's interviewer said afterward.

Coverage: Three experienced hires who were changing jobs: two software engineers (one at IC1, one whose level wasn't given) and one machine learning engineer interviewing for Snowflake's "AI Engineer" title. All three describe the onsite, and two also mention the rounds before it. The interviews took place in July–September 2026, and the reports went up between September 17 and October 23, 2026. One onsite was in person at the office and one was virtual; the third didn't say.

The candidates called the round by different names: "AI-enabled coding," "AI Augmented coding," or just "AI coding." In every case it was a single round of the onsite where the candidate could use an AI assistant to write the code.

Roles in this article

  • Software engineer, 2 reports: a rate limiter and a retrying task queue in the AI coding round, plus a key-value store, two system design questions and behavioral questions about using AI.
  • AI Engineer (machine learning), 1 report: the same rate limiter in an in-person AI coding round, then a design round on agent skills and agent memory.

Software engineer: Snowflake AI coding onsite (2 reports)

What came up

Both engineers had one AI coding round at the onsite, and in both the AI produced the code while the candidate worked out the design. Neither called the task hard. One thought the round went well and was still rejected; the other found the task simple but only learned at the very end what the interviewer was grading.

Topic Reports Last seen
One AI coding round at the onsite 2 of 2 Oct 2026
AI wrote the code while the candidate handled the design 2 of 2 Oct 2026
System design, at the phone screen or the onsite 2 of 2 Oct 2026
Behavioral round at the onsite 2 of 2 Oct 2026
Rate limiter 1 of 2 Oct 2026
Key-value store with transactions at the phone screen 1 of 2 Oct 2026
Project deep dive 1 of 2 Oct 2026
Task queue with backoff retries and a dead-letter queue 1 of 2 Sep 2026
Graded on prompting and trade-off judgment 1 of 2 Sep 2026
Behavioral questions about using AI at work 1 of 2 Sep 2026

A report can count toward several rows.

Questions we found

AI Coding

Question Reports Last seen
Design and implement a rate limiter (Sliding-Window Rate Limiter) 1: 01 Jul–Sep 2026
Queue plus worker that retries failed tasks with exponential backoff and moves them to a dead-letter queue after the limit (Design a Fault-Tolerant Cloud Queue Service) 1: 02 Jul–Sep 2026

The Sep 22 date in the rate limiter's FastPrep history is the AI Engineer report further down.

Algorithms / Data Structures

Question Reports Last seen
Key-value store with transactions, at the phone screen (Key-Value Store with Transactions) 1: 01 Jul–Sep 2026

System Design

Question Reports Last seen
Load data from S3 into a database, at the phone screen 1: 01 Jul–Sep 2026
Task scheduler (Design a Scheduled Job Execution System) 1: 02 Jul–Sep 2026

Behavioral (BQ)

Question Reports Last seen
How do you use AI in your work? 1: 02 Jul–Sep 2026
How do you check that what the AI gives you is right? 1: 02 Jul–Sep 2026

2 detailed candidate timelines

01: A rate limiter the AI mostly wrote, and a rejection anyway

Report date: Oct 23, 2026 · Interview period: Jul–Sep 2026 · Role: Software engineer, level not given · Experience: changing jobs, referred · Result: Rejected

Phone screen

  1. System design: load data from S3 into a database. No further detail was shared.
  2. Coding: a key-value store that supports transactions. The practice version is Key-Value Store with Transactions.

The onsite was booked very soon after the phone screen.

Onsite: AI-enabled coding

  1. Design and implement a rate limiter. Practice version: Sliding-Window Rate Limiter. That exercise fixes one algorithm; the report doesn't say which one this candidate chose.
  2. The candidate added later that the round was mostly about design, and that the AI wrote all of the basic code.
  3. They felt they answered the rate limiter well.

The report doesn't say whether the rate limiter had to run as a service called over HTTP or only as core logic.

Onsite: project deep dive and behavioral

  1. A deep dive into one of their past projects.
  2. A behavioral round. No questions were named.

02: IC1, a retrying task queue, and an AI that took three minutes per answer

Report date: Sep 17, 2026 · Interview period: Jul–Sep 2026 · Role: Software engineer, IC1 · Experience: changing jobs, referred · Format: virtual onsite · Result: Not known when posted; the candidate expected a rejection

Onsite format

  1. Three rounds back to back: AI coding, behavioral, then system design.
  2. The candidate called the AI round the hardest of the five rounds in their process. The other two weren't described.

Onsite: AI coding

  1. Build a queue and a worker that takes tasks from it. A failed task is retried with exponential backoff, and once it goes past the retry limit it moves to a dead-letter queue (a separate queue that holds tasks that keep failing). The practice version is Design a Fault-Tolerant Cloud Queue Service, a design exercise built on the same queue.
  2. The interviewer spent most of the time on the retry strategy and asked directly about jitter (adding some randomness to each wait) and a cap on the delay.
  3. Later the interviewer added that client jobs can have a priority. The candidate compared two layouts: one priority queue where the client passes a number, or separate queues for three fixed levels where the client picks one of the three.
  4. They had the AI write the code but ran out of time to test it.
  5. The task itself felt simple. What the candidate couldn't work out was what the interviewer wanted. Only at the end did the interviewer explain it: they look at how you ask the AI questions, how you write prompts, and how you judge trade-offs.
  6. The AI was the one built into CoderPad, and it was very slow, about three minutes per answer. The interviewer kept asking the candidate to query it anyway, so there were long, awkward waits.

Onsite: behavioral

  1. Mostly questions about AI: how they use it at work, and how they check what it produces.
  2. Very little about their resume.

Onsite: system design

  1. A task scheduler. Practice version: Design a Scheduled Job Execution System.
  2. Not many follow-ups on scaling. The candidate said the interviewer was good.

AI Engineer (machine learning): Snowflake AI-augmented coding onsite (1 report)

This role has one report, an in-person onsite. Its AI coding round was the same rate limiter as report 01, followed by a design round about AI agents and a behavioral round with the hiring manager.

Questions we found

AI Coding

Question Reports Last seen
Design and implement a rate limiter (Sliding-Window Rate Limiter) 1: 03 Jul–Sep 2026

System Design

Question Reports Last seen
Pipeline that automatically extracts and distills reusable agent skills for enterprise users 1: 03 Jul–Sep 2026
Common components of agent memory (Design Memory for an AI Assistant) 1: 03 Jul–Sep 2026

1 detailed candidate timeline

03: AI Engineer who passed, then turned the offer down

Report date: Sep 22, 2026 · Interview period: Jul–Sep 2026 · Role: Machine learning engineer; the title at Snowflake is AI Engineer, on an agent-focused team that hires directly · Experience: changing jobs, applied online · Result: Passed; declined the offer

Before the onsite

  1. An ordinary algorithm coding round. The candidate no longer remembers which problem it was.
  2. An ML fundamentals round.

Onsite format

  1. In person at the office, which the candidate said was rare in their own experience. Three rounds.

Onsite: AI Augmented coding

  1. Candidates were asked to bring their own laptop with any AI tools already installed. This candidate uses Codex and Claude. It turned out they could also use the interviewer's laptop, so they did, to save their own tokens.
  2. The recruiter mentioned two possible formats, refactor-and-debug and design-and-build. They first said it would be the first one, then later said the second.
  3. The task: design and implement a rate limiter, the same task as report 01. Practice version: Sliding-Window Rate Limiter.
  4. The AI was allowed to write the code. The candidate had to explain which algorithm they picked and why, what the other options were, and the trade-offs between them.
  5. Then they demoed the AI-written code live and walked the interviewer through how they had talked to the AI, how they reviewed its code, and how they tested it.
  6. They found it easy and enjoyable, and expected anyone who already works with AI day to day to feel the same.

Onsite: Expertise (design)

  1. A system design round tied to the team's work. First: design a pipeline that automatically extracts and distills reusable agent skills for enterprise users.
  2. Then: what are the common parts of an agent's memory? The practice version is Design Memory for an AI Assistant.

Onsite: behavioral

  1. With the hiring manager. No questions were named.

After the onsite

  1. The offer depended on a reference check with two references.
  2. The candidate passed and then declined the offer.

Patterns across the reports

These patterns come from the two software engineer reports (01 and 02). Report 03 is a different role, so it stays out of the comparison.

  1. The AI wrote the code, and the candidate owned the decisions. In 01 the round leaned toward design and the AI wrote all of the basic code. In 02 the AI produced the code too, and the interviewer said afterward that the grade came from how the candidate prompted it and weighed trade-offs.

  2. An easy-feeling task didn't predict the result. Neither engineer found the task hard. The candidate in 01 felt the rate limiter went well and was rejected. The candidate in 02 called the task simple, yet ranked the round the hardest of their five. They weren't told what was being graded until the end, and the AI took minutes to answer.

  3. The AI round sat next to a behavioral round. Both onsites had one. In 01 it came with a project deep dive; in 02 it came with a system design round, and the behavioral questions themselves were mostly about AI.

Common questions

What does the Snowflake AI coding interview ask?

A rate limiter (01, 03) or a retrying task queue (02), where the candidate designs and the AI writes the code.

How is the Snowflake AI coding round graded?

Candidate 02 was told it came from how they prompted the AI and weighed trade-offs, not from the code alone.

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.