Free guide · 5 steps · about 6 min · Updated Nov 3, 2026

AI Engineer interview prep

What AI engineers really build with LLMs, retrieval, agents and evals, 30 interview questions with model answers, real pay data for India, the US and Europe, and who is hiring.

Step 1 of 5 · 1 min

Where the AI Engineer role came from

The role in 30 seconds

An AI engineer builds products on top of foundation models such as GPT, Claude and Gemini. The work is mostly software engineering: prompts, retrieval (RAG), tool-calling agents, evaluation pipelines, latency and cost control, and safe deployment. Unlike a machine learning engineer, an AI engineer rarely trains models from scratch. Interviews test coding, LLM application design, evals and debugging.

The AI engineer was named in June 2023, when Shawn “swyx” Wang’s essay “The Rise of the AI Engineer” argued that foundation models had created a new kind of engineer. By January 2025 it was the fastest-growing job on LinkedIn’s US Jobs on the Rise list.

  1. June 2023
    The role gets a name

    The Latent Space essay argues that work which once needed “five years and a research team” can start from “API docs and a spare afternoon”.

  2. Oct 2023
    First AI Engineer Summit

    The community holds its first summit, and the title starts appearing in job posts.

  3. Jan 2025
    Top of LinkedIn’s list

    AI engineer takes the top spot on LinkedIn’s 2025 Jobs on the Rise list for the US.

  4. Jan 2025 → now
    72,786 people take the title

    72,786 people started an AI Engineer or GenAI Engineer title since January 2025, including 21,239 in India.

Why it took off

  1. 01

    Work that once needed a research team can now start from an API, so every product team needs engineers who build on models.

  2. 02

    Shipping AI reliably takes new skills: retrieval, tool calling, evals, and latency and cost control.

  3. 03

    It reuses software-engineering foundations, so developers can switch in without a PhD.

Up next · Step 2 · 2 minWhat the job really is
Step 2 of 5 · 2 min

What an AI Engineer really does

The AI Engineer community describes the role as a new kind of engineer at the intersection of software engineering and AI: work that once needed a research team can now start from an API. Our count shows 72,786 people started an AI Engineer or GenAI Engineer title since January 2025, including 21,239 in India.

Day to day, you design how a model is used: what context it receives, which tools it can call, how outputs are checked, and how the system behaves when the model is wrong. A typical posting, such as Fluency's AI Engineer role in San Francisco, asks for shipping LLM features end to end, designing evals that catch regressions, and choosing models for quality, latency and cost.

The title covers a wide range. At IT services firms it often means building GenAI solutions for clients; at product companies and startups it means owning AI features in production; at AI labs it overlaps with applied research and forward deployed engineering.

Skills you need

Software engineering

  • Python and/or TypeScript in production
  • APIs, async code and streaming
  • Databases, including vector search (pgvector)
  • Testing, logging and deployment

LLM application design

  • Prompting and structured outputs
  • Retrieval: chunking, embeddings, reranking
  • Tool calling, agents and MCP
  • Choosing models for quality, latency and cost

Evaluation

  • Building eval sets from real data
  • Model-graded and human evaluation
  • Regression testing across prompt or model changes
  • Error analysis by failure type

Production and safety

  • Caching, batching and rate limits
  • Prompt injection and data-leak defences
  • Monitoring quality in production
  • Fallbacks when models fail

What companies expect: B2B vs B2C

Most open roles in 2026 are B2B. Read that column first.

B2BMost roles today

B2B (most roles, including IT services): reliable AI inside business workflows

  • Build retrieval over company documents with permissions, citations and freshness, so answers can be trusted and audited.
  • Build agents that take actions in business systems (CRM, ticketing, ERP) with approvals for risky steps.
  • Design evaluation sets with domain experts and report accuracy per task before every release.
  • Meet enterprise requirements: data residency, no training on customer data, audit logs and access control.
  • Control cost per task, because enterprise usage is priced and budgeted.
  • At IT services firms, work across several client projects and present solutions to client stakeholders.
Metrics you will own
Task accuracy on the eval setHallucination or unsupported-claim rateCost per taskp95 latencyAutomation rate of the workflowIncidents and rollbacks
B2C

B2C: AI features millions of people use

  • Ship features such as chat, search, recommendations, writing help or voice for consumer apps, at large scale.
  • Optimise latency and cost per user aggressively, because usage is free or cheap for most users.
  • Handle abuse, jailbreaks and harmful content with safety filters and monitoring.
  • Run A/B tests and use feedback signals (thumbs, edits, retries) to improve quality.
  • Support many languages and low-end devices, which matters a lot in India.
Metrics you will own
Engagement and retention liftPositive feedback rateTime to first tokenCost per daily active userSafety incident rate

How it differs from a classic Engineer role

  • Compared with an ML engineer: you mostly use and adapt existing models instead of training them.
  • Compared with a software engineer: outputs are probabilistic, so testing means evals, not only unit tests.
  • Model choice, prompts and context are part of the codebase and need versioning.
  • Cost and latency per request are design constraints from day one.
  • The field changes monthly, so evaluating new models and tools is a regular task.
Up next · Step 3 · 1 minPay and who is hiring
Step 3 of 5 · 1 min

What a AI Engineer earns, and who is hiring

Levels.fyi tracks AI Engineer as its own title. It pays less than Machine Learning Engineer on Levels.fyi, partly because many AI Engineer roles sit in services companies and early-career positions.

India median
₹17.8 LPA
Levels.fyi, AI Engineer, India
US median
$156K
Levels.fyi, AI Engineer, US
Germany median
€81.1K
Levels.fyi, AI Engineer, Germany
Netherlands median
€65.0K
Levels.fyi, AI Engineer, Netherlands

India

INR, lakh per annum (total compensation)
  • Typically fresher to early career (0–2 yrs)
    25th percentile
    ₹10.9 LPA
  • Typically mid-level (2–5 yrs)Median
    Median
    ₹17.8 LPA
  • Typically senior (5–8 yrs)
    75th percentile
    ₹34.5 LPA
  • Typically lead or staff (8+ yrs)
    90th percentile
    ₹52.6 LPA

Source: Levels.fyi, AI Engineer, India (Updated Nov 3, 2026). AI Engineer title, India. Levels.fyi does not publish the sample size for this page. Its data leans towards product companies, so many services-company offers sit below these bands.

For comparison: Levels.fyi reports a US median of $287,500 for Machine Learning Engineers (Nov 2, 2026), and a recent AI Engineer posting at Fluency in San Francisco listed a base salary of $180,000–$250,000 (posting closed July 2026). Strong AI engineers at product companies often negotiate towards ML-engineer pay.

See every posted AI Engineer offer, with in-hand pay (36 job posts that state a salary).

Got an Indian offer? Turn the CTC into monthly in-hand pay on 2026-27 tax rules, or compare two offers.

Who is hiring the most in 2025–26

Companies with the most people who started an AI Engineer, GenAI Engineer or LLM Engineer title in 2025–26. IT services firms lead because they are staffing GenAI projects for clients at scale.

  • Tata Consultancy Services
    IT services (India)
    758new
  • IBM
    Enterprise tech
    393new
  • Deloitte
    Consulting
    333new
  • Accenture
    Consulting and IT services
    242new
  • Cognizant
    IT services
    162new
  • Microsoft
    Big Tech
    143new
  • Infosys
    IT services (India)
    140new
  • Amazon
    Big Tech
    55new

How we counted: Source: Apollo.io people database, searched Nov 3, 2026. Titles counted: AI Engineer, Senior AI Engineer, Applied AI Engineer, Generative AI Engineer, GenAI Engineer and LLM Engineer, held for 670 days or less. Across all companies, 72,786 people match, 21,239 of them in India. Big Tech often uses titles such as "Software Engineer, AI", so their counts here are a lower bound. (Apollo.io people search, Nov 3, 2026.)

Up next · Step 4 · 30 questionsCrack the interview
Step 4 of 5 · 30 questions

Crack the interview

First see the rounds you will face. Then practise every question out loud before you open the answer.

The rounds, by company type

Typical loops. Each company varies, so confirm the rounds with your recruiter.

AI startups and product companies

  1. 1Recruiter or founder call
  2. 2Practical coding in Python or TypeScript
  3. 3Build exercise: a RAG app or agent, often a take-home
  4. 4LLM system design (e.g. a support assistant at scale)
  5. 5Deep dive on a project you shipped, plus behavioural

Big Tech

  1. 1Recruiter screen
  2. 2Data structures and algorithms rounds
  3. 3ML or LLM system design
  4. 4Applied AI round: evals, retrieval and failure analysis
  5. 5Behavioural or leadership round

IT services and consulting (TCS, Infosys, Accenture, Deloitte)

  1. 1Online assessment (aptitude and coding)
  2. 2Technical interview: Python, GenAI concepts and your projects
  3. 3Case: propose a GenAI solution for a client scenario
  4. 4Managerial round
  5. 5HR round

AI labs

  1. 1Recruiter screen
  2. 2Coding with real-world framing
  3. 3Take-home or onsite build with their models
  4. 4System design and deployment trade-offs
  5. 5Values and mission interviews

30 AI Engineer interview questions

Each answer shows how you could respond and what the interviewer is listening for. Say yours out loud first.

Answer out loud, reveal, then rate yourself. Weak answers come back for review.0/30

Common opening questions

Show that you have shipped, measured and learned, not just followed tutorials.

1Walk me through an AI feature or project you built.
How you could answer

Cover the user problem, why an LLM was the right tool, the architecture (model, retrieval, tools), how you evaluated it, what failed and how you fixed it, and the result in numbers: accuracy, latency, cost or user adoption. Interviewers dig into decisions, so be ready to defend each one.

What the interviewer is listening for

Real decisions, measurement and honest failures.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

2What is the difference between an AI engineer and an ML engineer?
How you could answer

ML engineers typically train, fine-tune and serve models, working with datasets, training pipelines and model architecture. AI engineers mostly build applications on top of existing foundation models: prompts, retrieval, agents, evals and product integration. The roles overlap, especially for fine-tuning and evaluation.

What the interviewer is listening for

A clear, non-dismissive distinction.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

3How do you keep up with new models and tools?
How you could answer

Describe a system, not a feed: you follow a few primary sources (model release notes, papers or engineering blogs), and when something new appears you test it on your own eval set before adopting it. Give an example of a model switch you made or rejected based on your evals.

What the interviewer is listening for

Evidence-based adoption.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

4Why do you want to work on AI here?
How you could answer

Connect the company's AI product or clients to problems you have worked on. Mention something specific you tried or noticed in their product and one idea you would explore.

What the interviewer is listening for

Specific interest and preparation.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

LLM fundamentals

Explain clearly, then connect each concept to a product decision.

5What are tokens, and why do they matter for cost and latency?
How you could answer

Tokens are the chunks of text a model reads and writes; one word is often one to two tokens, and many non-English scripts use more. Pricing is per input and output token, and output tokens are generated one at a time, so long outputs increase latency. Shorter prompts, smaller contexts and concise outputs cut both cost and latency.

What the interviewer is listening for

Linking tokens to cost, latency and languages.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

6What does temperature do, and when would you set it to zero?
How you could answer

Temperature controls how random the model's choice of next token is. Low values make outputs more consistent, which suits extraction, classification and evals. Higher values add variety for brainstorming or creative text. Even at zero, outputs can vary slightly, so do not rely on exact repeatability.

What the interviewer is listening for

Correct intuition and practical defaults.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

7What is a context window, and what happens when you exceed it?
How you could answer

It is the maximum number of tokens the model can consider at once, including the prompt, retrieved documents, conversation history and the answer. If you exceed it, the request fails or content must be truncated. Even within the limit, very long contexts can cost more and reduce attention to important details, so retrieve and summarise selectively.

What the interviewer is listening for

Limits plus the quality trade-off of long contexts.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

8When would you fine-tune instead of using prompting or RAG?
How you could answer

Use prompting first, add retrieval when the model needs your facts, and fine-tune when you need consistent format, style or a narrow skill that prompting cannot reach, or a smaller, cheaper model to match a larger one on one task. Fine-tuning needs good labelled examples and an eval set to prove it helped. It is not the right way to add frequently changing knowledge.

What the interviewer is listening for

Ordered decision-making with evidence.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

9How do you get reliable structured output (JSON) from a model?
How you could answer

Use the provider's structured-output or function-calling feature with a schema where available, validate the output against the schema in code, retry with the validation error if it fails, and keep the schema simple. Log failures to improve prompts.

What the interviewer is listening for

Schema plus validation, not trust.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

10What are embeddings and what are they used for?
How you could answer

Embeddings are lists of numbers that represent the meaning of text, so similar meanings end up close together. They power semantic search, retrieval for RAG, clustering, deduplication and recommendations. Quality depends on the embedding model and on how you chunk the text.

What the interviewer is listening for

Clear definition with applications.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

Retrieval (RAG) design

The most common AI engineer system-design topic.

11Design a RAG system for a company's internal knowledge base.
How you could answer

Ingest documents with metadata (source, team, date, access group). Chunk by structure (sections, headings) with some overlap, embed and store in a vector index, ideally alongside keyword search (hybrid). At query time: rewrite the question if needed, retrieve with permission filters, rerank, and pass the top chunks to the model with instructions to cite sources and say "I don't know" when unsupported. Evaluate retrieval (did the right chunk come back?) and answers (correct and grounded?) separately, and keep the index fresh with incremental updates.

What the interviewer is listening for

Permissions, hybrid retrieval, citations and two-level evaluation.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

12How do you choose a chunking strategy?
How you could answer

Follow the document's structure: split on headings or sections rather than fixed character counts, keep chunks small enough to be specific but large enough to carry context, add overlap or parent-section context, and attach metadata. Then test: compare retrieval hit rate on your eval questions for two or three strategies.

What the interviewer is listening for

Structure-aware and empirically tested.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

13Your RAG system retrieves the right document but the answer is still wrong. Why?
How you could answer

Possible causes: the relevant passage was cut off or ranked low within the context, too many irrelevant chunks distracted the model, the prompt did not force use of sources, the question needed reasoning across several chunks, or the source itself was ambiguous. Fix with reranking, fewer but better chunks, clearer instructions and a citation requirement, then confirm on the eval set.

What the interviewer is listening for

Separating retrieval failure from generation failure.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

14What is hybrid search, and why use it?
How you could answer

Hybrid search combines keyword search (exact terms such as product codes, names and error messages) with vector search (meaning). Each catches what the other misses. Results are merged and often reranked by a cross-encoder or a model.

What the interviewer is listening for

Knowing why exact matching still matters.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

15How do you handle documents that update often?
How you could answer

Track document versions and timestamps, re-embed only changed documents, delete stale chunks, store the effective date as metadata and prefer the latest version at query time. Monitor index freshness as a metric.

What the interviewer is listening for

Incremental indexing and freshness.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

Agents and tool use

Know when an agent helps and when a simple workflow is better.

16Design an agent that resolves customer refund requests.
How you could answer

Tools: look up order, check refund policy, issue refund (with limits), escalate to human. Flow: identify the customer and order, check eligibility against policy, refund automatically under a threshold and ask for human approval above it, then confirm to the customer. Log every tool call, cap the number of steps, and evaluate on historical tickets for correct decisions and policy compliance.

What the interviewer is listening for

Bounded autonomy, approvals and evaluation on real cases.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

17When is an agent the wrong choice?
How you could answer

When the steps are known and fixed, a deterministic workflow with one or two model calls is cheaper, faster and easier to test. Agents add value when the path depends on what is discovered along the way. Start simple and add autonomy only where it measurably improves results.

What the interviewer is listening for

Pragmatism over hype.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

18How do you stop an agent from looping or taking harmful actions?
How you could answer

Limit steps and budget, give it least-privilege tools, require confirmation for irreversible or high-value actions, validate tool inputs in code, detect repeated identical calls, and provide a clear "stop and escalate" path. Test with adversarial scenarios.

What the interviewer is listening for

Guardrails in code, not just in the prompt.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

19What is the Model Context Protocol (MCP)?
How you could answer

MCP is an open standard for connecting AI applications to tools and data sources. An MCP server exposes capabilities, such as querying a database or creating a ticket, that any compatible AI client can use. It reduces custom integration work and makes permissions and logging consistent across apps.

What the interviewer is listening for

Accurate definition and why it matters.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

20How would you evaluate an agent, not just a single response?
How you could answer

Define end-to-end tasks with known correct outcomes, run the agent in a sandbox, and score success, the number of steps, cost, time and any policy violations. Review traces of failures to see which step went wrong, and keep the task set as a regression suite.

What the interviewer is listening for

Task-level success plus trace review.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

Evaluation and production

The skill that separates senior AI engineers from the rest.

21How do you build an eval set for a new AI feature?
How you could answer

Collect real inputs (logs, support tickets, user requests), cover common, hard and edge cases, write expected outputs or a grading rubric with domain experts, and include examples that should be refused. Start with 50–200 examples, grow it with every production failure, and run it on every prompt or model change.

What the interviewer is listening for

Real data, rubrics and continuous growth.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

22Can you use an LLM to grade another LLM's output?
How you could answer

Yes, with care. Give the grader a specific rubric, ask for reasoning plus a score, check its agreement with human labels on a sample, and watch for bias towards longer or similar-sounding answers. Use code-based checks wherever possible (exact match, schema, citations present) and keep humans in the loop for subjective quality.

What the interviewer is listening for

Calibrated model grading.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

23Latency is 8 seconds and users are leaving. What do you do?
How you could answer

Measure each step first. Then: stream the response, cut prompt and context size, run retrieval and tool calls in parallel, cache frequent answers, use a smaller model for simple requests or first drafts, and pre-compute where possible. Track time to first token and p95 latency.

What the interviewer is listening for

Measure first, then several concrete levers.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

24Your monthly LLM bill tripled. How do you investigate?
How you could answer

Break cost down by feature, model, customer and token type. Common causes: a prompt that grew, longer conversation history, retries in a loop, a switch to a pricier model or abuse. Fix with caching, context limits, routing easy requests to cheaper models, batch processing for non-urgent jobs and per-user limits.

What the interviewer is listening for

Attribution before optimisation.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

25How do you defend against prompt injection?
How you could answer

Treat all retrieved or user-supplied content as untrusted data, separate it clearly from instructions, limit the tools and data the model can reach, require confirmation for sensitive actions, filter outputs that could leak secrets, and keep attack examples in your eval suite. No single defence is enough.

What the interviewer is listening for

Defence in depth.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

26How do you safely switch to a new model version?
How you could answer

Run the full eval suite on the new model, compare quality, latency and cost by task, review differences manually, then roll out gradually behind a flag with monitoring and a quick rollback. Pin model versions in production so behaviour does not change unexpectedly.

What the interviewer is listening for

Eval-gated, staged rollout.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

Coding and behavioural

Expect practical coding plus classic behavioural questions.

27Write a function that splits a long document into overlapping chunks.
How you could answer

Take the text, a chunk size and an overlap. Move a window forward by (size − overlap) each step and slice, preferably on sentence or paragraph boundaries rather than mid-word. Handle edge cases: overlap greater than or equal to size, empty text and a final short chunk. Mention that in production you would count tokens rather than characters.

What the interviewer is listening for

Correct windowing and edge cases.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

28Implement retries with exponential backoff for an LLM API call.
How you could answer

Wrap the call in a loop with a maximum number of attempts. On rate-limit or server errors, wait base × 2^attempt plus random jitter, then retry; on client errors such as invalid requests, fail immediately. Respect any retry-after header, set an overall timeout and log each attempt.

What the interviewer is listening for

Knowing which errors to retry.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

29Tell me about a time an AI system you built failed in production.
How you could answer

Describe the failure, how you detected it (ideally monitoring, not a customer complaint), the root cause, the fix, and what you added so it cannot recur: an eval case, an alert or a guardrail.

What the interviewer is listening for

Ownership and systemic fixes.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

30Tell me about a time you pushed back on using AI for something.
How you could answer

A strong answer shows judgement: the task was better solved with rules, search or a simple model, or the risk was too high. Explain the evidence you used and the outcome.

What the interviewer is listening for

Judgement over enthusiasm.

Tick what your answer covered.

How did your answer compare?

Your rating decides when this question comes back for review.

Reading answers is not practice.

Answer these out loud to an AI interviewer and get feedback in minutes.

Start a mock interview
Up next · Step 5 · 2 minYour prep plan and resources
Step 5 of 5 · 2 min

Your prep plan and resources

Pick the plan that matches the time you have. Then learn the AI skills this role now expects, and go deeper with the best free resources.

Pick your plan: 7, 14 or 30 days

Tick tasks off as you go (saved on this device), and put the sessions in your calendar so they actually happen.

7 daysInterview next week0/7

45 minutes a day at 8 pm, starting tomorrow. Works with Google, Apple and Outlook calendars.

14 daysTwo weeks to build proof0/4

45 minutes a day at 8 pm, starting tomorrow. Works with Google, Apple and Outlook calendars.

30 daysA month to switch into AI engineering0/4

45 minutes a day at 8 pm, starting tomorrow. Works with Google, Apple and Outlook calendars.

AI skills for an AI Engineer

Most interviews now ask how you use AI. Learn what LLMs are, practise with the free OpenAI and Anthropic courses, then build one small agent or automation you can show.

Why it matters for this role
This is the AI role. Go beyond prompting: build a small RAG app or agent with evals, and be ready to explain your latency, cost and failure cases.
  1. Start hereAndrej Karpathy
    Intro to Large Language Models (1-hour talk)

    The clearest plain-English explanation of what an LLM is, how it is trained and where it fails.

    Key idea: An LLM predicts the next token. Everything else (chat, tools, agents) is built on top of that.

  2. Start hereOpenAI
    OpenAI Academy: AI Foundations, Applied AI, Agents and Workflows

    Free, self-paced courses of about 60–90 minutes each, with a completion certificate (needs a ChatGPT account).

    Key idea: Use AI on your real work every day. Interviewers now ask how you already use it.

  3. Start hereAnthropic
    Anthropic Academy: AI Fluency, the Claude API, MCP and Claude Code

    Anthropic’s own courses, from working well with AI to building on the Claude API, with certificates.

    Key idea: Being fluent means knowing when to delegate to AI, how to describe the task and how to check the output.

  4. BuildAnthropic
    Prompt engineering overview (Claude docs)

    Practical techniques: clear instructions, examples, structure and letting the model think step by step.

    Key idea: Define what “good” looks like and test against it before you tweak the wording.

  5. BuildAnthropic (Dec 2024)
    Building effective AI agents

    The most cited guide to workflow vs agent patterns, written from Anthropic’s work with customers.

    Key idea: Start with the simplest thing that works. Add agent autonomy only when results show it helps.

  6. BuildOpenAI (Apr 2025)
    A practical guide to building agents (PDF)

    When to build an agent, how to design its tools and instructions, and how to add guardrails.

    Key idea: An agent is a model plus tools plus instructions. Guardrails decide whether you can trust it.

  7. Go deepern8n
    n8n Academy: build automations and AI workflows

    Hands-on courses from your first workflow to APIs and AI, without needing to be a full-time engineer.

    Key idea: Ship one small automation that saves someone an hour a week. That is your best proof of work.

All free AI and interview resources
Free · No signup to start

Practise out loud. It’s free.

A AI Engineer mock interview with an AI interviewer, with feedback on every answer.

Start a free mock interview

Prepare for another new-age role

AI Engineer: frequently asked questions

What does an AI engineer do?

An AI engineer builds applications on top of foundation models: prompts, retrieval over company data, tool-calling agents, evaluation pipelines and production systems that control cost, latency and safety. Most AI engineers use existing models rather than training their own.

What is the salary of an AI engineer in India?

Levels.fyi reports a median total compensation of about ₹17.8 lakh a year for AI Engineers in India, with the 25th percentile near ₹10.9 lakh and the 90th near ₹52.6 lakh (Nov 2026). Product companies and AI startups usually pay more than IT services firms.

Which companies hire the most AI engineers?

In our count of people who started an AI Engineer or GenAI Engineer title in 2025–26, Tata Consultancy Services, IBM, Deloitte, Accenture and Cognizant lead, followed by Microsoft, Infosys and Amazon. Startups collectively hire a large share too.

Do AI engineers need a machine learning degree?

No. Strong software engineering, an understanding of how LLMs, retrieval and evaluation work, and shipped projects with measured results matter more. ML knowledge helps for fine-tuning and evaluation work.

What questions are asked in an AI engineer interview?

Expect LLM fundamentals (tokens, context windows, fine-tuning versus RAG), RAG and agent system design, evaluation, latency and cost optimisation, prompt injection defence, practical coding, and a deep dive on a project you built.