Employer resource · Updated August 2026
AI engineer job description: templates that actually work
Most AI job postings are wish-lists that repel the engineers they’re meant to attract. Here are two copy-paste templates — AI engineer and ML engineer — built from what we see working across hundreds of briefs and placements, with the salary bands and screening questions to go with them.
Why most AI job descriptions fail
They describe a department, not a job. Ten technologies, three research areas, and “wear many hats” tells a strong candidate you don’t know what you need — and strong candidates have options. The fix is naming the one problem the hire will own in their first six months.
They demand impossible experience. “5+ years building with LLMs” filters for people willing to lie. Ask for outcomes — shipped systems, production operation, measured results — not years in a field younger than the requirement.
They hide the money. The engineers you want know their market value to the dollar (they read the same salary data you do). No band, or a band 30% under market, means your posting is screening you out — silently.
Template 1: AI engineer (LLM applications)
Copy, then replace every [bracket] — the brackets are where your specificity goes
Title: AI Engineer — [product/team]
Compensation: $[150–220]k base + [equity/bonus], [location/remote policy]
The job: You will own [the specific system — e.g. “our customer-support copilot, from retrieval quality through production monitoring”]. Success at 90 days looks like [one measurable outcome].
What you’ll do: Build and ship LLM-powered features ([RAG / agents / structured extraction — pick what’s real]); design and run evaluations so we know quality before users do; own inference cost and latency; work directly with [the actual stakeholders].
What we require: You have shipped at least one LLM application to production and can walk us through what broke; you have built evaluation harnesses, not just demos; you write production-grade [Python/TypeScript]; you can explain a technical trade-off to a non-technical owner.
Nice to have: [2–3 items maximum — every extra line here costs you candidates.]
Stack: [The truth, including the boring parts.]
Template 2: Machine learning engineer
For trained-model work — forecasting, ranking, fraud, vision — not LLM product features
Title: Machine Learning Engineer — [domain]
Compensation: $[160–240]k base + [equity/bonus], [location/remote policy]
The job: You will own [the model and the pipeline around it — e.g. “demand forecasting across 40 warehouses, from feature pipeline through retraining and drift monitoring”]. Success at 90 days looks like [one measurable outcome against the current baseline].
What you’ll do: Improve on [the existing baseline — name it]; build the boring machinery that keeps models honest: versioned data, automated retraining, drift alerts, rollback; deploy and operate models serving [real scale numbers]; decide when classical methods beat ML — and say so.
What we require: You have operated production ML systems through retraining cycles and incidents; you reach for the simplest model that clears the bar and can defend that choice; you are fluent in [Python + the data stack you actually run]; you measure business impact, not just offline metrics.
Nice to have: [2–3 items maximum.]
Stack: [The truth, including the boring parts.]
Calibrate the bands against the full AI engineer salary guide and ML engineer salary guide — including the location adjustments if you pay by geography.
What to screen for once the applications arrive
| Signal | What it proves | Ask this |
|---|---|---|
| Production ownership | Systems running today, and what broke | “Walk me through an AI system you shipped that is still in production. What failed first?” |
| Evaluation discipline | How they know the model works | “How did you measure quality before launch — and how do you catch regressions now?” |
| Cost awareness | Inference economics, not just accuracy | “What did your system cost to run per month, and what did you do about it?” |
| Scoping judgment | Knowing when NOT to use AI | “Tell me about a problem where you recommended against an ML solution.” |
This is the same bar we vet against for network admission: verified production work beats credentials, portfolios, and interview polish. Candidates who clear all four questions are rare — which is the honest reason AI hiring takes months through open postings.
Or skip the posting entirely
A great JD still means months of sourcing and screening. Brief us instead: we match you with two or three pre-vetted engineers — contract or full-time — whose production record has already cleared the bar above.
Frequently asked questions
What should an AI engineer job description include?+
Five things: the specific problem the hire will own (not a technology wish-list), the stack they will actually touch, 3-5 hard requirements tied to production experience, a real salary band, and what success looks like at 90 days. Cut everything that reads like a research lab if the job is applied product work.
What is a realistic salary range to put in an AI engineer job posting?+
For US roles in 2026: $150k-220k for mid-level AI/LLM engineers, $220k-300k senior, with ML engineers roughly 5-15% above that. Postings with no band or a fantasy band ($120k for a senior LLM engineer) get ignored by exactly the candidates you want. Our full salary guide breaks this down by role and city.
Should the job title be AI engineer or machine learning engineer?+
Match the work. LLM application work — RAG, agents, integrating foundation models into product — is "AI Engineer" in 2026. Building and operating trained models — forecasting, ranking, fraud — is "Machine Learning Engineer". Using them interchangeably attracts the wrong pipeline and wastes everyone’s screening time.
How many years of experience should we require?+
Fewer than you think, described differently. "5+ years of LLM experience" is impossible (the field is younger than that) and signals an unserious posting. Require outcomes instead: shipped AI features, production ML systems operated through incidents, measurable results. A strong engineer with 2 years of intense applied-AI work outperforms 8 generic years.
Do we need a full-time AI engineer, or a contractor?+
For a first project, often a contractor: you pay senior rates only for the weeks you need instead of carrying a $250k+ hire plus months of recruiting — and the JD writes itself once the first system exists. Hire full-time when the roadmap shows continuous AI work beyond about six months. We broker both, and will tell you honestly which your brief calls for.