Skip to content
Zarif Automates

Career path

AI Engineer

An AI engineer builds software on top of models someone else trained: retrieval, tool use, output contracts and the evaluations that show whether a feature works. The title also covers customer-facing deployment jobs and some modeling jobs, so start by reading what a posting owns. Use this path to understand the work, build an evaluated project and prepare for the interview.

A 10-guide reading path

  1. AI Engineer: A Practical Career and Portfolio Guide

    Understand what the role owns and see a retrieval project with eight test cases.

  2. AI Engineer vs ML Engineer vs Software Engineer

    Tell an AI engineer posting from an ML or software engineer posting before applying.

  3. How to Become an AI Engineer: A Path With Proof

    Order what to learn from a software or data background and specify the portfolio project.

  4. AI APIs for Beginners: Make a Request and Handle the Response

    Treat a model call as an unreliable network call and validate what comes back.

  5. Three Prompt Contracts You Can Check with Evaluation Fixtures

    Turn an output that looks right into a contract with fixtures that pass or fail.

  6. Build a Small RAG Pipeline and Measure What It Misses

    Filter by access before ranking, check citations and measure what retrieval misses.

  7. Evaluate a Ticket Agent: Dataset, Scoring Code and Failure Report

    Keep format, evidence and correctness as separate scores without dropping failures.

  8. Run a Bounded Tool-Calling Loop in Python

    Separate a model's proposed tool call from an authorized action.

  9. How to Deploy AI Agents to Production

    Plan hosting, state, cost limits and rollouts for a model-powered service.

  10. AI Engineer Interview Guide

    Prepare the coding, system design, evaluation and debugging rounds.

Compensation in current employer postings

Checked September 26, 2026. Annual ranges from 6 specific employer postings. These are not market averages, comparable levels or guaranteed offers. Read the linked posting for the current range and package.

Employer-posted AI Engineer compensation checked September 26, 2026
Employer and roleLocationPosted rangeWhat the range covers
Baseten
AI Engineer
San Francisco (hybrid)$220K – $260KThe posting's compensation field, plus equity. It does not say whether the range is base salary.
Ramp
Applied AI Engineer
New York / San Francisco (hybrid)$204.4K – $352KThe posting's compensation field, plus equity. Final pay depends on location and level. It does not say whether the range is base salary.
Brex
AI Engineer, Product
San Francisco (hybrid)$171,000 - $240,000 USDExpected salary range; starting base pay depends on location, skills and experience. Equity and other compensation may be part of the total package.
Gusto
Enterprise Application AI Engineer
San Francisco (hybrid)$225,000-$250,000 /yrTargeted cash compensation for San Francisco, Seattle and New York; the posting targets $185,000-$205,000 /yr in Denver and most remote locations. Equity (RSUs) is separate. An internal IT role, not a product role.
Cognition
Applied AI Engineer
San Francisco / London / New York City (on-site)$180K – $225KEstimated base salary, plus equity, for a customer-facing role in Customer Engineering. The posting does not give a range by location.
Samsara
AI Engineer, Customer Success
Remote, United States (outside the SF Bay, NYC and DC metro areas)$105,910—$160,200 USDAnnual base salary. Performance bonus or variable pay and equity for eligible roles are separate.

Interview practice bank

Original practice prompts, not leaked questions or a claim about any employer’s interview loop.

Design a help-center assistant for a product with free and enterprise tiers.

Scope who asks and when the system declines, filter documents by tier before ranking, validate citations against retrieved evidence and name the first eval cases.

A model upgrade improved a public benchmark, but your assistant got worse. What do you do?

Reproduce changed answers on frozen inputs, compare retrieval, citation and answer scores separately, and explain why the benchmark measures a different task.

Build an eval set for a feature that has no labeled data yet.

Write the labeling policy first, mix representative and rare high-cost cases, hold back a set that is never tuned on and explain how incidents become new cases.

Your eval score went from 82 to 88 percent. Is the change better?

Ask for the denominator and which cases flipped in each direction, keep failures in the count and treat a few flips on a small set as inconclusive.

The agent sometimes calls the refund tool twice.

Find the missing idempotency key, separate the model's proposal from execution, require approval for irreversible calls and add a replay test.

When would you not use a model for this feature?

Name a rule-based alternative, compare error cost, latency and maintenance, and choose the simpler system when it meets the requirement.

Other AI engineering roles

All career paths and the guides every role draws on