What an AI engineer job description should cover

The core responsibilities are consistent across companies, even when the title varies (AI engineer, applied AI engineer, LLM engineer, GenAI engineer):

• Build and ship features powered by language models, from prototype to production.
• Integrate model APIs and orchestrate multi-step or agentic workflows.
• Build retrieval and context pipelines (RAG) when the product needs grounded, up-to-date answers.
• Write evaluations: test sets, automated scoring, and regression checks that run before a release.
• Monitor quality, latency, and cost in production, then iterate on prompts and pipelines.
• Work with product, design, and data to turn a requirement into a working system.

Evaluation is worth calling out. It is one of the most useful skills for the role and one of the least common on candidate portfolios, so a posting that names it tends to attract better-prepared people.

Keep the requirements section short. A long list of named tools rarely helps, because most tools show up in only a small share of postings and can be learned on the job. Focus on the durable skills.

Usually expected

• Strong Python. Comfort with a typed language such as TypeScript or Go is a plus.
• Experience integrating LLM or other model APIs into a real application.
• A working understanding of how to evaluate and test systems whose output is not deterministic.
• Familiarity with cloud deployment, observability, and the basics of running services in production.
• Clear written and verbal communication.

Nice to have

• Retrieval-augmented generation and vector databases.
• Agent frameworks and tool calling.
• Fine-tuning or post-training.
• Prompt and context engineering, guardrails, and safety.

Experience and education

• State a realistic years-of-experience range if you have one, and treat it as a guide.
• A computer science or engineering degree is common but not required for this role. Shipped work usually counts for more than credentials.

These two roles are often confused, and merging them into one posting is a common mistake. They share a core of Python, evaluation, and production skills, and they diverge from there:

• A machine learning engineer builds and trains models. The work leans on classical machine learning, statistics, data pipelines, and sometimes research.
• An AI engineer builds products on top of models that already exist. The work leans on API integration, orchestration, evaluation, and production reliability.

If you need someone to train models from scratch, write a machine learning engineer posting. If you need someone to build a reliable product around foundation models, write an AI engineer posting. Asking for both in one role tends to attract candidates who are a partial fit for each.

• Listing every framework and tool. Most named tools appear in a small fraction of postings. A long list reads as noise and screens out capable people who happen to use a different stack.
• Requiring a PhD or research background for a role that is really about product integration.
• Leaving out evaluation. It is one of the highest-signal asks and one of the easiest to forget.
• Skipping the pay range. A stated range improves the quality of applicants and is required by law in a growing number of places.
• Blending AI engineer and ML engineer requirements into a single posting.

Copy this, cut the lines you do not need, and do not add asks the role does not actually have.

Role
AI Engineer

About the role
We are looking for an AI engineer to build and ship features powered
by large language models. You will take ideas from prototype to
production, and you will own the quality, speed, and cost of what
you ship.

Responsibilities
- Build LLM-powered features and integrate model APIs into our product.
- Design retrieval and context pipelines where the product needs
  grounded answers.
- Orchestrate multi-step and agentic workflows.
- Write evaluations and regression checks, and use them to gate
  releases.
- Monitor quality, latency, and cost in production, and iterate.
- Collaborate with product, design, and data.

Requirements
- Strong Python and solid general software engineering.
- Experience integrating LLM or other model APIs into a real
  application.
- A working understanding of how to test and evaluate
  non-deterministic systems.
- Familiarity with cloud deployment and production observability.
- Clear written and verbal communication.
- [X]+ years building software.

Nice to have
RAG and vector databases, agent frameworks, fine-tuning, prompt
engineering, or safety and guardrails work.

Compensation
[state a range].

• Read past the boilerplate. Lines about collaboration and years of experience appear in almost every posting and rarely separate candidates.
• Focus your preparation on a short, high-signal set: Python, LLM API integration, evaluation, and production basics.
• A portfolio project with a real evaluation harness stands out, because most do not have one.

You can learn these skills, ranked by how often real postings ask for them, in the MLGuerrilla curriculum.

MLGuerrilla reads real AI engineer job postings in full and counts the skills they ask for. The guidance above reflects what shows up most often. See the measured breakdown at /data and the method behind it at /provenance.