How to Hire an AI Engineer: Skills, Pay and Assessment

When a company decides to put AI into its products or operations, someone has to build it, and the title it reaches for is "AI engineer". It is also a hard hire to get right, because the same title covers people who train new models, people who build products on existing ones and people who keep the data flowing underneath.

A clear definition comes first: what problem the engineer will solve, what data they will work with and who will run the system after launch. With that written down, you can decide what kind of engineer you need, what to pay and how to tell a strong candidate from a confident one.

This guide explains the main kinds of AI engineer, how to define the role, when to hire, grow or contract, where to find candidates, what government pay data shows, and how to assess candidates with a scorecard, a work sample and interview questions. It ends with the mistakes that most often lead to a wrong hire.

DisclosureJake Jorgovan co-founded Alpha Apex Group, which is mentioned in this article.

What an AI engineer does and where they come from

LinkedIn describes the role as engineers who "develop and implement AI models that perform complex tasks" such as problem-solving and prediction, and notes that the title is also used for machine learning engineers. AI engineers topped LinkedIn's 2026 Jobs on the Rise list (opens in a new tab) of the fastest-growing roles in the US, measured on jobs its members started from January 2023 to July 2025. It also topped the 2025 list (opens in a new tab), which covered January 2022 to July 2024.

The listed skills changed between the two years. In 2025 the most common skills LinkedIn recorded were large language models, natural language processing and PyTorch. In 2026 they included retrieval-augmented generation, a way of connecting a model to a company's own documents and data, alongside PyTorch.

LinkedIn's data also says where these engineers come from. In the 2026 list the roles AI engineers most often moved from were software engineer, data scientist and full stack engineer, and their median prior experience was 3.7 years. That gives you a place to look: software engineers with evidence of relevant AI work.

The main kinds of AI engineer

Job titles vary from company to company, so define the role by the work. LinkedIn counts machine learning engineers under the same title; this guide separates them because the work and the hire differ. Use these four groupings to describe the work you need.

If existing models can do the core task, consider an application engineer, and check whether data or platform work needs to come first. The role calls for real software engineering: the engineer has to integrate a model with your systems, test whether its answers are good enough, control cost and keep it working as the models underneath change. It needs less of the advanced mathematics that research roles demand. LinkedIn treats researchers as a separate role: AI and machine learning researchers ranked fifth on its 2026 list.

Define the role before you write the job description

An AI engineer job description can easily ask for everything: a research background, data pipelines, cloud deployment, product sense and the ability to explain it all to the board. Few people do all of that well, and the search stalls.

One role that took me a while to recruit for was scheduling guests for our podcasts, which I talked through on my podcast in 2021. The work had several parts, from writing the scripts and email campaigns to handling replies and scheduling the guests. My first job description asked one person to manage all of it, and I ended up splitting it into three separate roles. Some parts needed highly skilled people and others did not. If a role looks too hard to hand to one person, it may be several roles.

The same test applies here. If your description asks for a researcher, a data engineer and a product engineer at once, decide which part creates the value now, hire for that and cover the rest another way. Before you open the search, fill in a one-page brief.

Hire, grow or contract

You have three ways to get the work done, and the brief usually points to one.

• Grow an engineer you already have. LinkedIn lists software engineer first among the roles AI engineers moved from. For application work, consider training an engineer who already knows your systems and data. Check that they have time to learn and access to technical support.
• Contract for a defined project. A prototype, an evaluation of whether AI can do a task, or a one-off integration has an end date. A contractor or a specialist firm can do it, provided someone on your team will own the result afterwards.
• Hire. Consider a permanent hire when the workload is ongoing and you need engineering capacity inside the company. Whatever you choose, name who owns monitoring, maintenance and improvements.

Image

Zoom

The approach described in this article, shown as a visual guide.

The choices combine. One option is a single experienced hire who sets the technical direction, with existing engineers trained alongside them. When the gap is leadership rather than engineering, the role you need may be a Chief AI Officer.

Where to find candidates

For a hire, look beyond people who already hold the title. AI engineers most often come from software engineering, data science and full stack engineering roles, so look for candidates with relevant AI work; referrals from your own engineers are a good place to start. In LinkedIn's 2026 list, San Francisco, New York City and Dallas were the regions with the largest shares of AI engineers employed from January to October 2025. A specialist recruiter can run the search for you; the AI recruiters and search firms guide compares them. For a defined contract, compare independent engineers or specialist firms on relevant past work, using the same assessment criteria you would use for a hire.

What to pay an AI engineer

The Bureau of Labor Statistics' Occupational Outlook Handbook has no AI engineer occupation; its A to Z index of job titles (opens in a new tab) lists none. The government's figures for two related occupations give a reference point. BLS notes that research scientists' expertise will be needed to create new technologies related to artificial intelligence.

Application work is closer to software development, and research roles to the research scientists. Skills matter to pay as well as titles: a study of about eleven million UK job postings (opens in a new tab) from 2018 to mid-2024 found that AI skills carried a wage premium of 23%, more than any degree below a doctorate.

Image

Zoom

The study’s UK job-posting data show different relationships between degree requirements and advertised pay in AI, green, and other roles. Figure 2 combines salary trends, education requirements, and estimated degree premiums. It does not show that dropping a degree requirement causes a pay increase.arxiv.org (opens in a new tab)

Treat the government figures as a reference point, not an offer. Before you set a range, check current offers for the specific role, seniority and city, through a compensation survey or a recruiter working on similar searches, and decide how much of the package is base salary. Location is part of the offer too: in LinkedIn's 2026 data, 26.2% of AI engineer jobs posted from November 2024 to November 2025 were flagged as remote and 27.1% as hybrid.

Degrees or skills

A degree tells you someone learned the foundations. It does not tell you whether they can ship a working system. The same UK study found that from 2018 to 2023, demand for AI roles grew by 21% as a share of all postings, while mentions of university degree requirements for AI roles fell by 15%. The authors describe this as employers turning to skill-based hiring to widen the pool.

For research roles, specify the research capability you need and whether an advanced degree is necessary to show it. For application and machine learning engineering, put evidence of relevant work first: systems they built, what happened when those systems met real users, and code or write-ups you can review.

How to assess AI engineer candidates

Fluent answers about models do not establish production experience, so ask candidates for evidence of systems they have deployed. Have an experienced engineer, from your team or an outside adviser, judge the technical evidence, and use three tools in order: a scorecard agreed in advance, structured questions and a work sample.

1. Agree the scorecard before the first interview

Weight the scorecard to the kind of engineer you need. The example below is for an application engineer. For a machine learning engineer, add modeling and statistics as a competency and reset the weights to match your brief.

Define what each rating means before you meet anyone. For evaluation and quality, for example, a 1 could be a demo with no measured evaluation, a 3 a relevant test set and a clear account of its limits, and a 5 a test set plus evidence of catching regressions and handling failures. Record the evidence behind each rating, multiply each rating by its weight and add them up for a total out of 5. Agree in advance which competencies are essential, so a high total cannot hide a missing one.

2. Ask every candidate the same structured questions

The US Office of Personnel Management, the federal government's HR agency, describes a structured interview (opens in a new tab) as one where all candidates are asked the same predetermined questions in the same order and all answers are judged against the same rating scale. Write the questions and what a strong answer looks like before the first interview, and have each interviewer rate independently before the panel talks.

3. Set a work sample from your own work

A work sample asks candidates to do a task that mirrors the job. OPM's guidance on work samples (opens in a new tab) says they closely reflect real work and that performance on them relates highly to performance on the job. It also says they can be costly to develop and run, and give the best return when critical skills are needed on entry, the number of applicants to test is limited and only a few are expected to have the skills. Give the task to your shortlist, and if you plan to teach a skill after hiring, do not make mastery of it a selection requirement.

Base the task on a real problem, simplified and stripped of confidential data: improve the answers of a simple document assistant, find why a model's results fell after a data change, or structure a messy dataset so a model can use it. Keep it short, tell candidates how it will be judged, and ask them to walk you through their choices afterwards.

Image

Zoom

The approach described in this article, shown as a visual guide.

Generic online tests have a narrower use. TestGorilla's test library (opens in a new tab), for example, lists multiple-choice tests of core AI and machine learning concepts. They can screen a large applicant pool, but they cannot show whether someone can build on your systems.

If you use automated tools to screen applicants, check the rules where you hire. New York City's Local Law 144 (opens in a new tab) bars employers and employment agencies from using an automated employment decision tool unless it has had a bias audit within one year of its use, information about the audit is public and the required notices have gone to candidates. The guide to reducing bias in AI hiring tools covers the wider question.

Once you have chosen, turn the role brief into an onboarding plan: access to systems and data, the person who reviews the engineer's technical work, the first deliverable and who accepts it.

AI engineer interview questions

Choose questions that fit the kind of engineer in your brief, ask every candidate the same ones and score the answers against the scorecard.

Building and shipping

Data, quality and evaluation

Judgment and communication

For a machine learning engineer

Mistakes that lead to the wrong AI hire

• One job description for several roles. A list that asks for research, data engineering and product work in one person invites a long search or a compromise hire. Split it into roles you can fill.
• Hiring for theory when you need delivery. Questions about how algorithms work favor candidates from research. If the job is building products on existing models, test building.
• Weighting degrees over evidence. A degree requirement narrows the pool without telling you whether someone can ship. Require a degree only when you can explain why the work needs it.
• Ignoring the data. Models need reliable data. If yours is scattered, the first hire may need to be a data engineer, or the AI engineer needs time and authority to fix it.
• No owner after launch. AI systems need monitoring and maintenance. Decide before you hire who will run what gets built.
• Not testing communication. The engineer will explain limits and risks to people who are not technical. Score that skill as deliberately as coding.

If you wrote the role brief today, could you say in one sentence what your AI engineer will deliver in the first 90 days?