Blog

How to land a job at an AI lab

What 3,074 job postings say, and what the interview tests. We read the open engineering postings at six AI labs, at Amazon and at 13 large tech companies, then held a real lab posting up against a traditional resume.

By Koaches8 min read
51%of AI lab engineering postings name Python.Amazon 51% · other big tech 39%
19%of AI lab postings name Rust, two times other big tech.Amazon 13% · other big tech 9%
4%of AI lab postings name Java.Amazon 68% · other big tech 20%
41%of AI lab postings name no programming language.Amazon 20% · other big tech 34%

Where the data comes from

Koaches runs a job data service that reads postings directly from company career sites, through their public job boards. It does not scrape LinkedIn.

On 25 October 2026, we collected 6,612 open postings from these job boards and from the public Amazon Jobs search. We kept only engineering and research roles, and removed sales, solutions and recruiting roles. That left 3,074 postings in three groups:

  • AI labs (976 postings): OpenAI, Anthropic, xAI, Perplexity, Cohere and Thinking Machines.
  • Amazon (1,031 postings): our FAANG sample. Google, Meta and Apple do not publish their postings on these public job boards.
  • Other big tech (1,067 postings): Stripe, Cloudflare, DoorDash, Datadog, Coinbase, Reddit, Lyft, Pinterest, Robinhood, Airbnb, Figma, Instacart and Dropbox.

A posting counts for a language when its text names that language. One posting can name many languages.

Python is the base. Rust is the signal.

Figure 1

Languages named in engineering postings

AI labsAmazonOther big tech
0%20%40%60%80%Python51%51%39%C++11%70%11%Java4%68%20%Rust19%13%9%Go13%2%30%TypeScript10%3%12%SQL8%5%18%Ruby0%14%10%CUDA3%2%1%0%20%40%60%80%Python51%51%39%C++11%70%11%Java4%68%20%Rust19%13%9%Go13%2%30%TypeScript10%3%12%SQL8%5%18%Ruby0%14%10%CUDA3%2%1%
Share of engineering and research postings that name each language: 976 AI lab, 1,031 Amazon and 1,067 other big tech postings, collected 25 October 2026.
Show as a table
LanguageAI labsAmazonOther big tech
Python50.9%50.6%38.7%
C++10.8%70.1%11.0%
Java4.1%67.7%19.5%
Rust18.8%13.0%8.7%
Go12.5%2.4%29.8%
TypeScript9.6%2.9%12.0%
SQL8.1%4.7%17.7%
Ruby0.0%13.7%9.6%
CUDA2.7%1.5%0.5%
  • Python is the shared language. Half of lab postings and half of Amazon postings name it.
  • Rust is the lab signal. 19% of lab postings name Rust. At other big tech companies, 9% do. At xAI the number is 31%, and at Perplexity it is 44%.
  • Java almost disappears. Only 4% of lab postings name Java. At Amazon, 68% do.
  • C++ is an Amazon language. 70% of Amazon postings name it. Labs ask for it in 11% of postings, mostly for inference and GPU work.
  • Go belongs to infrastructure companies. Datadog names Go in 46% of postings and Cloudflare in 44%.

Many lab postings name no language at all

41% of lab engineering postings name no programming language. At Amazon, 20% name none. A lab posting names 1.2 languages on average. An Amazon posting names 2.2.

When a lab posting names a language, it often gives you a choice:

"Production-quality coding ability in at least one of Python, Go, Rust, or TypeScript."Anthropic engineering posting, October 2026
"Rust or C++ experience is useful but not required."OpenAI engineering posting, October 2026

The labs tell you this directly. They want strong engineers who learn a new stack fast. A keyword match on your resume does not get you the job.

What lab postings ask for instead

Figure 2

Skills named beyond languages

AI labsAmazonOther big tech
0%20%40%60%80%Evaluation32%18%16%LLMs30%21%17%Distributed systems30%44%32%Kubernetes16%1%17%AWS11%54%16%Reinforcement learning10%11%3%PyTorch8%6%6%JAX6%3%1%0%20%40%60%80%Evaluation32%18%16%LLMs30%21%17%Distributed systems30%44%32%Kubernetes16%1%17%AWS11%54%16%Reinforcement learning10%11%3%PyTorch8%6%6%JAX6%3%1%
Share of the same postings that name each skill. Evaluation includes evals of models.
Show as a table
SkillAI labsAmazonOther big tech
Evaluation31.9%17.5%15.9%
LLMs29.9%21.4%17.3%
Distributed systems29.7%43.9%31.9%
Kubernetes15.7%1.1%17.2%
AWS10.8%54.0%16.0%
Reinforcement learning9.7%10.8%3.2%
PyTorch7.9%5.8%5.8%
JAX5.6%3.4%0.6%

Lab postings name evaluation and large language models about two times as often as big tech postings. They name Kubernetes and JAX more often too. Amazon postings name AWS in 54% of cases, because AWS is Amazon's own cloud.

Most lab work is still normal engineering. 30% of lab postings name distributed systems. A strong backend or infrastructure engineer can do much of this work.

The other half of the posting is about you

Postings do not only list skills. They also tell you what kind of person the company wants. We counted these words too.

Figure 3

Values and behaviors named in engineering postings

AI labsAmazonOther big tech
0%20%40%60%80%Mission81%28%71%Safety76%20%23%Collaboration79%68%82%Communication70%45%63%Ownership60%54%60%Ambiguity39%25%28%Curiosity and learning17%35%27%Fast pace18%19%29%0%20%40%60%80%Mission81%28%71%Safety76%20%23%Collaboration79%68%82%Communication70%45%63%Ownership60%54%60%Ambiguity39%25%28%Curiosity and learning17%35%27%Fast pace18%19%29%
Share of the same postings that use each word group, for example "mission"; "safety", "safe" or "responsible AI"; "ambiguity" or "ambiguous".
Show as a table
Word groupAI labsAmazonOther big tech
Mission81.4%27.7%71.4%
Safety76.3%20.3%22.9%
Collaboration78.6%67.7%81.6%
Communication69.6%44.7%63.2%
Ownership60.1%54.0%60.0%
Ambiguity39.4%25.2%27.9%
Curiosity and learning16.9%35.4%27.2%
Fast pace18.3%19.3%28.6%

81% of lab postings talk about the mission. 76% talk about safety. At Amazon, the numbers are 28% and 20%.

Some of this text is the company's standard paragraph. That is the point. The lab puts its mission in front of every candidate, and the interview checks it.

Amazon is different in one way. Its postings ask for curiosity and learning more often (35%). This agrees with its Leadership Principle "Learn and Be Curious".

The interview tests it

Each of these companies runs a behavioral part of the interview. The difference is what they measure you against.

Figure 4

What each company says about its interviews

Values or behavior is tested here
OpenAIInterview guide
  1. Resume review
  2. Recruiter or hiring manager call
  3. Skills assessment
  4. Final loop4-6 hours with 4-6 people. Looks for collaboration, communication, openness to feedback and "alignment with our mission and values"
AnthropicCareers page
  1. Application
  2. Live codingColab or CodeSignal
  3. Experience and motivationFirst value: "Put the mission first"
AmazonInterview page
  1. ScreensVary by role
  2. Loop with 2 to 7 AmazoniansQuestions on 16 Leadership Principles, answered with STAR, with a Bar Raiser
Googlere:Work guide
  1. ScreensVary by role
  2. Onsite loopThe same questions and scale for every candidate. Looks for "emergent leadership"
A summary of each company's public description of its process, October 2026. The steps differ by role. The sources are at the end of this post.

Big tech tests behavior against a written list. At Amazon, that list is the 16 Leadership Principles. You can prepare one story for each principle.

Labs test behavior against a mission. That is harder to prepare. You must connect your own work to their mission in your own words. A memorized answer does not do this.

Anthropic's candidate AI policy makes one more point clear. In a live interview, "this is all you". But the same policy encourages candidates to use AI to prepare. The practice happens before the interview.

Case study: a traditional engineer reads an OpenAI posting

Most engineers who apply to a lab do not come from a lab. So we took a traditional software engineering resume and a real lab posting, and read them the way a Koaches job analysis does.

  • The resume: a senior software engineer with seven years of checkout and commerce APIs, in Python, TypeScript and PostgreSQL. This is John Doe, the fictional sample profile in Koaches.
  • The job: OpenAI, Software Engineer, Monetization Product & Platform, San Francisco. The posting names no programming language.
Figure 5

The job analysis

Software Engineer, Monetization Product & Platform

OpenAI · Senior · Engineering
HeadlineA founding engineer for OpenAI's ads and monetization platform. You build a new system. You do not add features to an old one.
SenioritySenior or higher. The posting asks for 7+ years and gives you a voice in the technical roadmap.
The trapMost applicants will pitch scale. The posting uses safety, privacy and trust words 17 times, and scale and reliability words 11 times.
What they hire for, most important first
  • Safety, privacy and user trustStrong17 mentionsUphold the highest levels of safety, privacy, fairness, and policy compliance across all monetization systems
  • Distributed systems at scaleStrong11 mentionsArchitect distributed systems and services that operate at massive scale with strong reliability
  • Builds new systems from nothingStrong7 mentionsThrive in 0→1 environments where systems, architecture, and user experiences must be defined from scratch
  • Works closely with Product, Design and ResearchMeaningful7 mentionsWork closely with Product, Research, and Design to translate product requirements into high-quality technical solutions
  • Partner APIs and integrationsMeaningful5 mentionsDevelop advertiser-facing APIs and tools that enable the creation, management, and measurement of ads
  • Technical strategy and roadmapMeaningful5 mentionsContribute to technical strategy, propose new directions, and help shape the roadmap
Worth the stretchStrong on reliability and partner work. Two gaps to close with stories, not with new skills.
Matched
  • Product, Design, Research: "Aligned product and operations on a gated March launch."
  • Partner APIs: "Built API integrations and improved incident runbooks."
Partial
  • Distributed systems: the checkout migration cut failures from 2% to 0.6%. Add traffic numbers to show scale.
  • Safety and trust: idempotency keys cut duplicate orders from 40 to 2 a week. No line on privacy yet.
Gaps you can close
  • New systems: the resume shows migrations and fixes, not a system built from nothing.
  • Roadmap: the resume shows delivery, not strategy.
What to do, most important first
  1. Build a story about a time you protected users or their data while you shipped. Put trust first, then speed.
  2. Build a story about a system you started with no spec: what you decided, and what happened.
  3. Prepare "Why OpenAI, and why ads?" Say how ads can keep user trust. This is the mission question.
  4. Add scale numbers to the checkout bullet: orders per day and requests per second.
The analysis reads the posting the way a hiring manager does. It counts repeated themes, reads the order of the bullets, and quotes the posting for every trait. The verdict is "Worth the stretch" because one required trait is a gap.

This engineer does not need a new language. The posting does not ask for one. The engineer needs two new stories and one strong answer to the mission question.

The rounds tell the same story:

Figure 6

Predicted rounds for this job

Tests behavior or mission
  1. Round 1Recruiter callWhy OpenAI, why monetization, and your timeline.
  2. Round 2Technical screenPractical coding in the language you choose.
  3. Round 3System designAn ads or billing system that keeps user data private.
  4. Round 4Project deep diveYour checkout migration: the decisions, the trade-offs and the result.
  5. Round 5Hiring manager and valuesCollaboration, feedback, ambiguity and the mission.
A prediction from the posting and from OpenAI's public interview guide. When you add what your recruiter told you, it replaces the prediction.

Three of the five rounds test behavior or mission. Your code gets you to the final loop. Your stories get you the offer.

How to prepare with Koaches

Koaches prepares you for the rounds that test you as a person: the recruiter call, the behavioral rounds, the project deep dive and the hiring manager. It does not cover coding or system design.

Figure 7

From posting to practice

  1. Add your resume

    Upload a PDF or Word file. Koaches reads it into sections you can edit.

  2. Paste the job

    Read who the company wants, how you fit, and what to do. Every trait shows its quote from the posting.

  3. Open the prep plan

    Get a plan for each round: likely questions, the stories to use, and the stories to build.

  4. Write a starting answer

    Generate an answer from your own resume and stories. Fill the blanks with your own facts.

  5. Practise out loud

    Answer by voice or by typing. Get scores on five dimensions, what worked, what to fix, and a stronger version of your answer.

  6. Update after each round

    Upload your notes or transcript after a real round. The next round's prep changes to match.

The labs hire engineers who can explain why.

Practise that answer before the interview. New accounts get $5 of free credits.

Start preparing

Method and limits

  • We collected the postings on 25 October 2026 from each company's public Greenhouse or Ashby job board, and from the Amazon Jobs search (US, five engineering search terms).
  • We kept postings whose title names an engineer, developer, researcher, scientist or member of technical staff. We removed sales, solutions, customer and recruiting titles.
  • We counted a language when the posting text names it. For Go, we counted "Golang", and "Go" only in a list of languages.
  • Mistral and Character.AI moved their job boards, so their postings are not in this data. Google DeepMind, Google, Meta and Apple do not publish on these job boards.
  • OpenAI and Anthropic have more open roles than the smaller labs, so they have more weight in the lab numbers.

Sources

  1. OpenAI, Interview guide: openai.com/interview-guide
  2. Anthropic, Careers: anthropic.com/careers
  3. Anthropic, Guidance on candidates' AI usage: anthropic.com/candidate-ai-guidance
  4. Amazon, What it is like to interview at Amazon: aboutamazon.com
  5. Amazon, Leadership Principles: amazon.jobs
  6. Google re:Work, Structured interviewing: rework.withgoogle.com