Datadog AI Coding Interview, Experienced SWE: 5 Experiences

Shared by janebeingjane ·
Company
Datadog
Reports
5 interview reports
Level
Full-time
Stages
Onsite

Datadog's AI coding interview in 2025–2026: review a large pull request with AI, or build a small client, often a Snowflake query client, with your own AI.

At a glance

  • Process: an onsite round (01–04) or a phone screen (05), with any AI tool and a shared screen.
  • Asked recently: a Snowflake query client in 02, 04 and 05; a Python PR review with a missing fallback (01); a Java build in a self-configured environment (03).
  • Difficulty: candidates wrote more about setup trouble than code: package compatibility took most of 04's round, and 05 finished too fast. Follow-ups after the first version, such as secrets and batching (02), decided the round.
  • Practise first: Build a Snowflake Query Client.

Coverage: 4 experienced full-time SWE onsites, interviewed January–September 2026 and reported April–September 2026 · plus 1 full-time phone screen whose hiring level isn't stated, kept in a separate supplement (reported November 13, 2026; its header lists a July–September 2025 interview season) · AI-assisted code review and feature development rounds only · These accounts don't establish the internship environment

What came up

Three of the four experienced candidates built something with AI help, and one reviewed a Python change. Two of the builds were a Snowflake query client. None of the four names a Python application framework.

Topic Reports Last seen
AI-assisted implementation 3 of 4 April–June 2026 interview
Follow-up questions after the first version 3 of 4 April–June 2026 interview
Snowflake query client 2 of 4 April–June 2026 interview
Dependency or package setup problems 2 of 4 April–June 2026 interview
Python PR review with AI help 1 of 4 July–September 2026 interview

These counts cover the 4 experienced reports. A candidate can appear in more than one row.

Questions we found

AI Coding

Question Reports Last seen
Build a Snowflake Query Client 3: 02, 04, 05 April–June 2026
Review a Batched Processor Pull Request 1: 01 July–September 2026

One experienced candidate described the AI round and their Java setup without naming the task. Additional FastPrep dates and the supplement report don't add to the counts above.

4 detailed candidate timelines from the AI rounds

01: A Python PR with a missing fallback

Interview period: July–September 2026 · Report date: September 4, 2026 · Role: Full-time SWE · Stage: Onsite

The candidate reviewed a large PR for a Python application that failed under heavy load. The change replaced one-at-a-time processing with batches of 40 requests. The original synchronous processor was still in the repository.

They found two concrete issues in the batched processor review:

  1. If the batch processor couldn't be created, the new code returned None instead of falling back to the existing synchronous processor.
  2. The PR claimed capacity of 6,000 requests per second per instance, while the README described peak demand of 8,000 per instance.

The candidate stressed how big the PR and its documentation were. In their account, AI was expected to help find and rank the critical bugs, with the important suggestions checked against the code. Python is explicit; the framework and dependency list aren't disclosed.


02: A query client, with secrets and batching as follow-ups

Interview period: April–June 2026 · Report date: May 14, 2026 · Role: Full-time SWE · Stage: Onsite

The task was to build a Snowflake query client: connect to Snowflake, start a query and check its status. The candidate used Claude Code and described the first implementation as quick.

Follow-ups covered secret management and batching. The report doesn't give the programming language, a Python framework or the exact batching requirement.


03: Java in a self-configured environment

Interview period: April–June 2026 · Report date: April 27, 2026 · Role: Full-time SWE, moving from a current job · Applied: Referral · Stage: Onsite · Result: Rejected

This candidate could use their own AI tool, any one they liked, and shared their screen so the interviewer could watch how they worked with it. The problem itself wasn't hard: the AI wrote the code in one pass. Dependencies were what slowed them down, and they were stuck on environment setup for a while. They came away feeling that solving the problem wasn't the real test, but couldn't tell what the interviewer wanted to see.

In a May 7 clarification, the author said they used Java in their own, self-configured environment. The actual coding task wasn't disclosed, so this account shows one candidate's setup without identifying a Python framework.

The candidate later reported a rejection. System design had gone well; feedback on the AI round was "mixed" and mentioned follow-ups that weren't covered. They had finished ten minutes early and answered both follow-ups the interviewer asked, and felt any further ones should have been asked too.


04: Package compatibility consumed most of the round

Interview period: January–March 2026 · Report date: April 30, 2026 · Role: Full-time SWE · Stage: Onsite

Another candidate got a Snowflake client task and could use local AI tools. Package compatibility problems took more than 30 minutes. They connected successfully with about ten minutes left and had little time for follow-ups.

In an author reply, they listed client extensions the interviewer had suggested: pagination, multiple tenants, authentication, long-running queries, recovery, cancellation and throttling. These were possible directions, not features they finished. The author thought the task was about a query client rather than building a backend service.

Neither the original post nor that clarification names a Python framework.


Supplement: hiring level not reported

This full-time report doesn't say whether the candidate was experienced, new grad or something else, so it sits outside the counts above.

05: A Snowflake API client on a phone screen, finished too fast

Report date: November 13, 2026 · Interview season stated in the report header: July–September 2025 · Role: Full-time SWE, level not stated · Applied: Referral · Stage: Technical phone screen · Result: Rejected

  1. The round ran on a website that opens a virtual environment much like VS Code. Any AI tool was allowed, including a command-line coding agent.
  2. The task was to implement read and write operations against a Snowflake API and handle errors, close to Build a Snowflake Query Client.
  3. The candidate let the AI write it. It was done very quickly, and then they weren't sure what to do next while the interviewer kept asking follow-up questions.

Looking back, the candidate would rather have added features one at a time and kept control of the pace. As in reports 02 and 03, the first version came quickly, and follow-ups filled the rest of the round.

Python framework status

The full-time reports give useful task mechanics, but they don't establish FastAPI, Flask or Django as the interview framework. The supplement report doesn't name one either. The SWE intern framework also remains unconfirmed in the material available so far.

An intern discussion contains an unanswered question about FastAPI. It doesn't include a candidate's confirmation after finishing the round.

The FastAPI and pytest labels in the linked FastPrep practice workspaces describe the exercise setup we authored. They don't identify dependencies supplied by Datadog. The original repositories and their dependency manifests were unavailable.

Patterns across the reports

These cover the 4 experienced reports.

  1. The Snowflake query client. Reports 02 and 04 both got a Snowflake client task. Report 02's follow-ups were secrets and batching; report 04's interviewer suggested pagination, multiple tenants, authentication, long-running queries, recovery, cancellation and throttling as directions.

  2. Setup, not code, took the time. Reports 03 and 04 both lost time to dependencies or package compatibility, more than 30 minutes in report 04. Report 03 said the AI-written code itself came in one pass.

  3. Follow-ups after the first version. Reports 02, 03 and 04 all describe follow-ups once a first version existed: secrets and batching in 02, two follow-ups in 03 whose coverage the feedback questioned, and suggested client extensions in 04, where little time was left.

Common questions

What does the Datadog AI coding interview ask?

Mostly building a Snowflake query client with your own AI tool (02, 04, 05), or reviewing a Python PR (01).

What slows candidates down in the Datadog AI round?

Environment setup: package compatibility took most of 04's round, and 03 configured their own Java environment.

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.