By Vadym Lobariev, founder of MindHunt and Claude Certified Architect — recruiting engineers across Europe and Ukraine since 2011.
Quick answer
A good AI engineer job description in 2026 describes production work on top of foundation models — LLM features, retrieval, agents, evaluation, cost and reliability — not the generic "develop machine learning models" boilerplate that most templates still use. It names which kind of AI engineer you need, says what the role is not, asks for one to three years of real LLM production experience on top of solid software engineering (not "5+ years of LLM experience," which almost nobody has), and states a salary range. A copy-and-adapt template is below.
Search for an AI engineer job description template and the top results will hand you something like this: "Develop machine learning models and AI solutions. Experience with machine learning, deep learning, NLP and computer vision. Proficiency in Python, Java and R. Knowledge of TensorFlow or PyTorch."
That describes a 2019 machine learning engineer. Post it today and you will attract data science graduates, repel the engineers who actually ship LLM products, and spend your interviews discovering the mismatch.
What the Job Actually Is in 2026
The best evidence comes from real postings rather than opinion. AI Shipping Labs analysed 889 AI engineer job listings collected in January 2026 across Berlin, Amsterdam, London, Los Angeles and New York. The findings are hard to square with the standard template:
- 70% of roles were "AI-first" — direct work with LLMs and retrieval. Another 28.5% were AI platform and infrastructure roles. Fewer than 2% were traditional machine learning or deep learning jobs.
- 95.6% were production-focused. Only 4.4% were research.
- The most requested skills were RAG (35.9% of postings), prompt engineering (29.1%) and LLM integration (25.4%).
- Python appeared in 82.5% of postings — but TypeScript was in 23.4% and React in 14.8%. Cloud and delivery skills were everywhere: AWS 40.1%, Docker 31.0%, CI/CD 29.3%, Kubernetes 29.1%.
Read that list again. It describes a software engineer who builds products with language models — not a statistician who trains them. Your job description should too. If you genuinely need someone to train models, you are hiring an ML engineer, and the difference matters: AI Engineer vs. ML Engineer.
First Decide Which AI Engineer You Need
Real postings cluster into three archetypes. One job description cannot serve all of them, and trying is the most common reason a search attracts the wrong people.
| Archetype | What they do | Typical titles |
|---|---|---|
| Product AI engineer | Builds user-facing LLM features end to end: retrieval, agents, structured outputs, evaluation, the API and often the UI around them | AI Engineer, Applied AI Engineer, Founding AI Engineer |
| AI platform / enablement engineer | Builds the shared layer other teams use — model gateways, MCP servers, evaluation and observability tooling, cost controls — or runs the AI tooling your own engineers depend on | AI Platform Engineer, AI Platform Lead, AI Enablement Engineer, MLOps Engineer |
| Forward-deployed / solutions AI engineer | Works with customers or internal business teams to turn their workflows into working AI solutions quickly | Forward Deployed Engineer, Solutions AI Engineer, Member of Technical Staff |
Titles vary more than most employers realise. Among the postings reviewed for this guide, the same work appears as "AI Engineer," "Applied AI Engineer," "Member of Technical Staff, AI" and "Software Engineer (AI)." Candidates search for all of them. Pick the one your target candidates use for themselves — for product roles, "AI Engineer" or "Applied AI Engineer" is the safest choice.
What Strong Real Postings Have in Common
I looked at how companies with strong engineering reputations describe this role. Four patterns stand out.
1. They say "production" and mean it. Stripe asks for "experience building and deploying AI agents, LLM-powered tools, or workflow automation beyond basic prompt engineering" and four or more years of shipping production systems. The phrase "beyond basic prompt engineering" does more filtering than a paragraph of buzzwords.
2. They name evaluation as a responsibility. WITHIN requires "experience designing evaluations for LLM-powered systems," including test cases and regression testing. Future lists "creating evaluation frameworks and quality scoring mechanisms" among the core duties, and names observability tools — Langfuse, OpenTelemetry, Datadog — outright. Generic templates say "monitor and debug."
3. They put engineering fundamentals first. GitLab's posting puts it best: "AI is a powerful part of your toolkit, but it sits on top of solid engineering fundamentals, not in place of them." It also asks candidates to "confirm whether AI is the right intervention" before building — judgment, written into the job description.
4. They are honest about requirements and pay. GitLab tells candidates that many successful applicants do not meet every requirement. Several postings publish the range: Stripe $126,600–$189,800, Future $215,000–$250,000 plus equity, WITHIN $90,900–$254,100 across five levels.
AI Engineer Job Description Template
This template is for the most common case — a product AI engineer. Replace everything in square brackets, and delete any line that is not true for you. A shorter honest description beats a longer impressive one.
[Applied] AI Engineer · [Location / Remote within time zones] · [Full-time / Contract] · [Salary range]
About the role
[One paragraph: what your product does and what AI does inside it today. Be concrete — "our assistant answers 40,000 customer questions a month using retrieval over our help centre" tells a candidate more than "we are an AI-first company."]You will build and run the LLM-powered parts of [product]: designing them, shipping them, measuring whether they work, and keeping them fast, reliable and affordable. This is a hands-on software engineering role. It is not a research position, and it does not involve training models from scratch.
What you will do
- Design, build and ship LLM-powered features end to end — [retrieval over our data / agents that take actions in X / document extraction / …]
- Build the evaluation that tells us whether those features work: test sets from real usage, automated scoring, regression checks before every prompt or model change
- Choose models for each task and manage the trade-off between quality, latency and cost
- Make model output dependable for the rest of the system: structured outputs, validation, fallbacks
- Instrument what you ship — tracing, quality monitoring, cost tracking — and respond when it degrades
- Design for safety: prompt injection, data privacy, human approval for actions that cannot be undone
- Tell us when an LLM is the wrong tool for a problem
What you bring
- Strong software engineering background: [4+] years shipping production systems in [Python / TypeScript / your stack]
- Hands-on production experience with LLM features that real users depended on — not only prototypes or side projects
- Working knowledge of the current toolkit: [retrieval and vector search / tool use and agents / MCP / structured outputs / prompt caching]
- Experience evaluating LLM output systematically, beyond manual spot checks
- Practical familiarity with at least two model providers [Anthropic, OpenAI, Google, open-source] and opinions about when to use which
- Solid delivery habits: testing, CI/CD, observability, cloud infrastructure [AWS / Azure / GCP]
- Clear written and spoken English; able to explain risk and trade-offs to non-engineers
Nice to have
- [Domain experience: fintech / healthcare / regulated data]
- [Fine-tuning experience — only if you will really use it]
- [Front-end skills, if the engineer will own UI]
- [Observability tooling: Langfuse, OpenTelemetry, Datadog]
What we offer
[Salary range. Contract type. Remote policy and required time-zone overlap. Who they will work with. What is already built and what is greenfield. Interview process and how long it takes.]You do not need to match every line above. If you have shipped LLM features to production and care about whether they work, we want to hear from you.
Adjusting the Template by Seniority
| Mid-level | Senior | Lead / Platform | |
|---|---|---|---|
| Engineering experience | 2–4 years | 5+ years | 7+ years, including leading a team |
| LLM production experience | Has shipped at least one LLM feature | 1–3 years; has designed and run a system end to end | Has built AI services that other teams depend on |
| Responsibility wording | "Build and ship features within an established design" | "Own the design, evaluation and operation of…" | "Define the platform, its interfaces and its roadmap; mentor engineers" |
| Add for this level | Who will mentor them | Decision-making scope; on-call expectations | Hands-on versus leadership ratio (for example 70/30); stakeholders |
The early-career track exists too — some companies hire engineers with under two years of experience into AI roles. If you do, say so plainly, and be realistic about the senior time needed to support them.
The Tool Salad: One Line That Gives You Away
The mistake I see most often in the job descriptions clients send me is a single requirements line like this: "Experience with Claude Code, Cursor, Copilot, n8n, Make, LangChain, LangGraph."
To a strong candidate, that line says the person who wrote it does not know who they are looking for. Those seven names belong to three different categories, and being good at one says almost nothing about the others.
| Category | Examples | What it actually is | How to handle it in the description |
|---|---|---|---|
| AI coding tools | Claude Code, Cursor, GitHub Copilot | How the engineer writes code. Even here the tools differ: Claude Code is an agentic coding system that runs in the terminal, as an IDE extension or as a desktop app and carries out multi-step tasks; Cursor is a code editor — a fork of VS Code — with AI built into the editor itself; Copilot is an assistant inside the editor you already use. | One line: "fluent with AI-assisted development." Do not require a specific tool — a good engineer switches in a week. |
| LLM application frameworks | LangChain, LangGraph, LlamaIndex, provider and agent SDKs | Libraries your product is built with — code that ships to users. These are current, not legacy: LangChain and LangGraph reached version 1.0 in late 2025 and run in production at large companies. | List them only if your codebase really uses them, and add "or similar." Many strong engineers build directly on provider SDKs. |
| No-code automation platforms | n8n, Make, Zapier | Visual workflow builders from before the LLM era — Make began as Integromat in 2012, n8n in 2019 — that later added AI steps. Mostly used by operations teams and automation specialists. | If this is the core of the job, you are hiring an automation specialist, not a software engineer. An engineer who is fluent with an agentic coding tool will build the same integration in code and rarely needs them. |
Putting LangChain on the same line as Claude Code confuses the libraries your product runs on with the tool an engineer types into. Putting n8n next to either confuses two different professions.
The fix: group tools by what they are for, ask for the capability, and name tools only as examples — "has built agents with tool use (LangGraph, a provider agent SDK, or similar)." If you cannot say which category a tool on your list belongs to, take it off the list and ask someone who can before you publish.
The Years-of-Experience Trap
Building products on top of large language models became a distinct discipline around 2023. "5+ years of LLM experience" is therefore a requirement that excludes nearly everyone qualified — and signals to the ones who are that the company does not understand the field. None of the real postings reviewed for this guide asked for it. They ask for years of software engineering (four or more at Stripe, for example) plus demonstrated production work with LLMs.
Write it the same way: general engineering seniority as a number, LLM experience as evidence — "has shipped LLM features that real users depended on."
Should You Publish the Salary?
Yes, where you can. Three of the seven real postings reviewed here state a range. Several US states already require one in the posting, and the EU Pay Transparency Directive — due in national law from June 2026 — obliges employers to tell candidates the pay range before the interview. For a role where candidates typically hold more than one offer, a missing range reads as a low one.
Publish a band wide enough to be true across the levels you would hire, and say what decides where in the band someone lands. What the market pays — US, Western Europe, Poland and Ukraine, including total employer cost — is in our AI engineer salary guide.
Seven Mistakes That Sink AI Job Descriptions
- Machine learning boilerplate. TensorFlow, PyTorch, "deep learning, NLP and computer vision" for a role that will call model APIs. You will interview the wrong people.
- The tool salad. Coding tools, product frameworks and no-code platforms on one line, as described above. It tells candidates the author does not know the field.
- The kitchen sink. Training models, building agents, MLOps, data engineering and front end in one role. Strong candidates read that as "they do not know what they need."
- Impossible tenure. Five or more years of experience in a three-year-old discipline.
- No mention of evaluation or production. The engineers you want look for those words. Their absence tells them the company is still at the demo stage.
- Not saying what the role is not. One sentence — "this is not a research role" — saves everyone weeks.
- A degree requirement. Many of the strongest AI engineers are backend engineers who moved into LLM work. A mandatory MSc or PhD filters them out for no benefit in an application role.
Once the description is attracting the right people, the next risk is the interview: 15 AI engineer interview questions, with the red flags. The whole process from brief to offer is in How to Hire an AI Engineer.
How MindHunt Helps
Every search we run starts with the brief, because most AI searches that fail were lost there. We will tell you which archetype you are describing, whether the requirements match the salary, and what the same budget reaches in Ukraine, Poland or Western Europe — before anyone is contacted. Candidates are then technically screened by a founder who is a Claude Certified Architect.
→ Talk to us about your AI engineer brief
Frequently Asked Questions
What does an AI engineer do?
An AI engineer builds software products on top of foundation models: LLM-powered features, retrieval (RAG), agents and tool use, evaluation, and the monitoring, cost control and safety work needed to run them in production. An analysis of 889 real job postings found 95.6% of AI engineer roles are production-focused rather than research.
What skills should an AI engineer job description list?
Strong software engineering in Python or TypeScript; production experience with LLM features; retrieval and vector search; agents, tool use and structured outputs; systematic evaluation of model output; model selection and cost management; observability; cloud and CI/CD; and awareness of prompt injection and data privacy. In real postings, RAG, prompt engineering and LLM integration are the most frequently requested skills.
What is the difference between an AI engineer and a machine learning engineer job description?
A machine learning engineer description centres on training and deploying models: statistics, data pipelines, frameworks such as PyTorch, model evaluation. An AI engineer description centres on building products with existing models: retrieval, agents, evaluation of LLM output, cost and reliability. Using the first template for the second job is the most common mistake.
How many years of experience should I require?
Ask for general software engineering experience as a number — typically four or more years for a senior hire — and for LLM experience as evidence, such as having shipped LLM features to real users. The discipline is about three years old, so "5+ years of LLM experience" is unrealistic.
What job title should I use?
"AI Engineer" or "Applied AI Engineer" for product roles; "AI Platform Engineer" or "AI Platform Lead" for shared infrastructure; "Forward Deployed Engineer" or "Solutions AI Engineer" for customer-facing work. Use the title your target candidates would search for.
Should an AI engineer job description include a salary range?
Yes where possible. Several leading employers publish ranges, pay transparency rules increasingly require it, and candidates with multiple offers tend to skip postings without one. US postings reviewed for this guide ranged from about $127,000 to $250,000 depending on level and company.
Does an AI engineer need a degree in AI or machine learning?
Not for an application-focused role. Many strong AI engineers are backend or full-stack engineers who moved into LLM work. A mandatory advanced degree makes sense only for research or model-training positions.
Written by
Vadym Lobariev
MindHunt is an AI powered recruitment firm for founders, C-level and hiring managers who are tired of posting and praying. We execute a proven sourcing process for your hardest roles and show you the work every week — so you can make hires with confidence, not hope.
