AI Engineer: getting AI to work reliably inside real software
What AI engineers actually build, from LLM agents and retrieval to trained models in production, what four employers' postings ask for, and the routes people take into the title.
An AI engineer builds software features that use machine learning or large language models and keeps them working in production. In current postings that means two overlapping kinds of work: training and deploying models on a company's own data, and building applications on top of LLMs, such as agents that call tools, retrieval over internal documents and structured outputs. Either way, the engineer owns the whole path from framing the problem to monitoring the feature after launch, including evaluation, guardrails, cost and failure handling.
What an AI Engineer does
AI engineering is software engineering with a probabilistic component in the middle. The engineer decides whether AI is the right tool for a problem, picks or trains the model, wires it into services through APIs and data pipelines, and builds the checks that tell the team whether it is working. Because model output can be wrong in ways normal code is not, much of the job is evaluation: test sets, automated evals, traces and human review that catch regressions before users do.
The title is new and loosely used. Some employers mean a machine learning engineer who trains models on their data; others mean an engineer who builds LLM-powered products with agents, retrieval and prompt orchestration; many want both. What the postings share is production ownership. A prototype in a notebook is the start, not the deliverable: the feature has to stay fast, affordable, secure and correct under real traffic.
- Writing an evaluation set and running it against a new prompt or model version.
- Building a retrieval pipeline over internal documents and measuring answer quality.
- Adding a tool call or API integration to an agent, with permission checks.
- Investigating a production trace where the model gave a wrong or unsafe answer.
- Cutting latency or token cost on a feature that is getting expensive at scale.
- Explaining to a product partner where AI helps and where plain code is the better answer.
O*NET lists AI Engineer (Artificial Intelligence Engineer) as a job title under Computer and Information Research Scientists (15-1221.00). That is a research occupation that does not describe production AI engineering work, so this page does not use O*NET task or skill data, and TryHired does not quote that occupation's pay for this title.
What employers ask AI Engineers to do and bring
- Employers ask AI engineers to own features from problem framing through production deployment and monitoring, not only prototypes.
- LLM application work is central in several postings: agents, retrieval, tool calling and structured outputs.
- Evaluation is a core skill: employers ask for automated evals, regression tests and guardrails for model behavior.
- Strong Python and general software engineering come first; employers expect production services, APIs and data work.
- Senior postings ask for about five or more years of software engineering, including running production systems.
- Employers value judgment about when AI is the right answer and when simpler, deterministic software is better.
- Databricks, in a posting based in Costa Rica, lists the LLM failure modes it wants an engineer to understand, including hallucination, incorrect tool use, prompt injection and excessive permissions.
- Chime's AI/ML engineer role is the model-building kind: training and deploying deep learning and classical models on financial and behavioral data.
- Anduril says its applied LLM role is not a prompt-writing job, asks the engineer to manage token usage and cost, and asks for eligibility for a U.S. security clearance.
- PagerDuty's Lisbon-based role pairs applied AI with distributed systems: the engineer reasons about throughput, fault tolerance and cost for features running on real-time event streams.
These points come from current job postings on the employers' own public job boards, read on the dates shown. Each general point is backed by at least two independent sources; points that rest on a single employer name that employer. They describe what employers commonly ask for, not legal requirements, and any one employer may differ.
| Employer | Posting | Read |
|---|---|---|
| Anduril | Applied LLM Systems Engineer | Nov 16, 2026 |
| Chime | AI/ML Engineer | Nov 16, 2026 |
| Databricks | AI Automation Engineer | Nov 16, 2026 |
| PagerDuty | Senior AI/ML Engineer | Nov 16, 2026 |
For a duty-by-duty reading of real postings, see the AI engineer job description.
AI engineer vs machine learning engineer, data scientist and software engineer
- A machine learning engineer usually focuses on training, serving and scaling models; an AI engineer title increasingly also covers building applications on top of LLMs that someone else trained.
- A data scientist answers questions with data and experiments; an AI engineer ships a feature that runs in production and is responsible for how it behaves there.
- A software engineer builds deterministic systems; an AI engineer adds a component whose output can be wrong, so evaluation, guardrails and monitoring become part of the job.
- A prompt engineer tunes model instructions; the AI engineer postings we read expect full systems work, including APIs, retrieval, permissions and cost control.
How to become an AI Engineer
- Get solid at software engineering first. Python, APIs, testing, Git and cloud basics come before any AI-specific skill in these postings.
- Build one end-to-end AI feature. For example, retrieval over a document set with an evaluation set that scores the answers, deployed somewhere people can use it.
- Learn evaluation properly. Write test sets, automated evals and traces, and be able to show a regression you caught.
- Learn the ML fundamentals. Model types, metrics, data leakage and train/serve skew still matter, especially for model-building roles.
- Move over from an adjacent seat. Backend, data and ML engineers are common routes in; ask to own an AI feature in your current team.
What AI Engineers work with
- Languages: Python above all, plus SQL; some teams use Go or TypeScript for services.
- Model and LLM tooling: PyTorch for model work, LLM APIs and agent frameworks for application work.
- Retrieval: embeddings, vector search and document pipelines that ground answers in trusted sources.
- Evaluation and observability: eval harnesses, tracing, logs and dashboards for model behavior.
- Cloud and infrastructure: AWS, GCP or Azure, containers and Kubernetes, data pipelines and orchestration.
Credentials: what exists and what employers rely on
- AI engineering is not a licensed occupation, and none of the four postings we read asks for an AI certification. They look at shipped production systems, programming depth and evaluation experience.
- Cloud and data-platform vendors run their own certification programs, such as the AWS certifications; they can show platform knowledge, but they are not a substitute for a working AI feature you can explain.
Certifications are optional unless an employer or regulator says otherwise; read the posting, and check eligibility on the issuer's own site before you apply.
Where this role leads
- Machine Learning Engineer: for AI engineers who want to go deeper into training and serving models.
- Data Scientist: for people who prefer experiments, analysis and modeling questions.
- Software Engineer: a common starting point and a fallback with the widest choice of roles.
- Data Engineer: for people who enjoy the pipelines and data platforms AI features depend on.
What AI Engineers earn
TryHired does not print a salary for AI engineers. The U.S. Bureau of Labor Statistics publishes pay for occupations, not job titles. O*NET 31.0 lists the title under Computer and Information Research Scientists (15-1221.00), which BLS publishes as Computer and Information Research Scientists (BLS 15-1221); as this page explains, that occupation does not describe the title's day-to-day work, so TryHired does not quote its pay here.
Pay for this title depends on employer, industry, region and level, so ask how the role is graded and compare offers against the posting's own range where one is published.
Applying for AI Engineer jobs
- Show one AI feature you built end to end, with the evaluation numbers before and after a change.
- Be ready to explain a failure you found, such as a hallucination or a bad tool call, and how you fixed it.
- Talk about cost and latency, not only accuracy: interviewers want to know you can run it at scale.
- Match the posting's flavor: model training for some teams, LLM applications and agents for others.
Two free tools start from this job title, so you only add your own experience: draft a resume summary for an AI Engineer role or write a LinkedIn headline for an AI Engineer role. Neither fills in skills or results for you: list only the work you have actually done.
What real postings list, duty by duty, and how to answer them.
Open →ResumeSoftware Engineer resumeStructure, sections and examples you can adapt to this title.
Open →Cover letterSoftware Engineer cover letterA letter that names the team, the problem and your evidence.
Open →InterviewSoftware Engineer interviewQuestions to rehearse before the interview.
Open →SkillsAI Engineer skillsThe skills to show and how to show them honestly.
Open →Interview questionsAI Engineer interview questionsPractice questions and how to answer them.
Open →Disclosure: TryHired makes HIRED, a paid job-search program, so we have a conflict of interest when we mention it. Everything on this page is free to use without buying anything.
Questions people ask
What does an AI engineer do?
Is an AI engineer the same as a machine learning engineer?
How do I become an AI engineer?
How this page is built
The employer section summarizes current postings from the employers' own public job boards, saved on the read date shown; every general point is backed by at least two independent sources and single-employer points name the employer; each maps to short verbatim quotes in this page's claims ledger. Employers' preferences are not legal requirements. The job, difference, path, tools and hiring sections are TryHired's explanation and advice, labeled as such. TryHired does not print pay for job titles that BLS does not publish separately.
- Pay for this title: BLS publishes occupations, not titles.
- Whether a particular employer requires a degree, a certification or a license for this title.
- How a specific company levels or names this role.
- Four postings from four employers are sampled here, all at technology companies, two of them outside the United States (Costa Rica and Lisbon), and the title is used inconsistently across the industry.
- Postings outside the United States: Databricks (Costa Rica); PagerDuty (Lisbon). Asks in other countries can differ from those of US employers.
- Chime, “AI/ML Engineer” job posting, read Nov 16, 2026.
- PagerDuty, “Senior AI/ML Engineer” job posting, read Nov 16, 2026.
- Databricks, “AI Automation Engineer” job posting, read Nov 16, 2026.
- Anduril, “Applied LLM Systems Engineer” job posting, read Nov 16, 2026.
- U.S. Bureau of Labor Statistics publishes pay by occupation, not job title; this page names the occupations and does not quote their figures.