How to become an AI engineer in 2026: the roadmap, based on real job ads

To become an AI engineer, get solid at software engineering first, then learn to build on existing models: LLM APIs, evals, retrieval, agents and running them in production. Most job ads care about years of engineering experience, not degrees. The realistic route for most people is software engineer first, AI engineer second.

We read the 95 AI engineer job ads open on Tokn on 27 October 2026, in Europe and North America, to see what employers actually ask for. This guide turns that into a roadmap. For the wider picture, see our guide to careers in AI.

In short

• An AI engineer builds products on existing models. Training models is the machine learning engineer's job.
• It is rarely a first job: of the postings that state a minimum, 51 of 53 ask for 3+ years, mostly of software engineering. Only a handful were open to people with about 2 years or less.
• 65 of the 92 postings we could read in full don't mention a degree.
• Learn in this order: Python (then TypeScript), LLM APIs, evals, retrieval, agents, production. Fine-tuning comes later, if at all.

What does an AI engineer do?

An AI engineer builds software products on top of models someone else trained: plugging LLMs into applications, building retrieval (RAG) pipelines, agents and evaluation suites, then running them in production. The machine learning engineer, by contrast, trains, fine-tunes and serves the models themselves.

The term took off with swyx's 2023 essay "The Rise of the AI Engineer", which described a line between model-side ML and research engineers and product-side AI engineers. In it, swyx quotes Andrej Karpathy: "One can be quite successful in this role without ever training anything." Chip Huyen's book AI Engineering (O'Reilly, January 2025) uses the same framing: building applications with readily available foundation models.

Titles vary (Applied AI Engineer, AI Software Engineer, GenAI Engineer), so read the responsibilities.

For the model-training side, read how to become a machine learning engineer. See also what an LLM engineer does and our guide to the forward deployed engineer.

What AI engineer job ads ask for: the Tokn data

The ads want experienced software engineers who can ship LLM products. Few ask for a degree, and very few are open to beginners.

How we counted: Tokn counted every open job whose title says AI engineer (including applied AI engineer, AI software engineer and GenAI engineer), excluding managers and research roles. On 27 October 2026 that gave 95 postings at 32 companies: 59 in Europe and 36 in North America. Our North American sources are still being added, so don't read the regional split as market size.

Who is hiring AI engineers

The model labs dominate. OpenAI alone accounts for 24 of the 95 postings, about one in four, and together with Anthropic and Mistral AI the three labs post 44 of 95. That concentration shapes every figure below.

Product companies hire too: in Europe, Doctolib, Legora, Pigment, Aircall and Shift Technology each had two or more postings. Many lab roles are customer-facing: 54 of the 92 postings we could read in full had "Applied AI" or "Deployed" in the title, and most mention customers.

How much experience they ask for

AI engineer is rarely a first job. Of the 53 postings that state a minimum, 51 ask for 3 years or more and 36 ask for 5 or more. Mid-level applied AI roles at Anthropic and OpenAI asked for 4+ years.

The experience is mostly software engineering, with some LLM work on top. A ClickHouse ad in October 2026 asked for "5+ years of software engineering experience, including 1–2 years on LLM-powered systems or agents in production". A Doctolib senior ad asked for "4+ years of experience as a software engineer", and another Doctolib ad said: "You can succeed without training a model from scratch."

Reading the full text, only a handful of postings, all in Europe, were open to people with about 2 years of experience or less, among them several Mistral AI roles and a founding AI engineer role at Plato (Hexa). North America had none. AI engineer roles are less junior-friendly than AI jobs in general, so keep an eye on entry-level AI jobs as a way in.

Which skills the ads mention

The core is Python, TypeScript, LLM APIs, evals, agents and retrieval. Model training is a minority skill.

Method: keyword counts on the full text of 92 readable postings (Europe 57, North America 35), so treat them as indicative. "Agents" and "evaluation" can match company boilerplate, and OpenAI's 24 postings weigh heavily.

TypeScript in over half the ads sets this role apart from the ML engineer path: AI engineers ship product.

The AI engineer roadmap: skills in the order you need them

Learn software engineering first, then the LLM layer on top, in the order the job ads use it. Fine-tuning and PyTorch come last, if at all.

1. Software engineering in Python, then TypeScript. APIs, testing, Git, SQL, deployment. Python appears in 72% of ads, TypeScript or JavaScript in 54%.
2. LLM APIs and prompting. Calling models, structured outputs, tool use, cost and latency. Start with the providers' own material: Anthropic Academy has courses on building with the Claude API and on MCP, and OpenAI Academy covers the OpenAI side.
3. Evals. Build test sets, use LLM-as-judge carefully, run regression tests on every change. 77% of ads mention evaluation, and Huyen's book gives it a full treatment.
4. Retrieval (RAG). Chunking, embeddings, hybrid search, citations. 41% mention RAG or retrieval, but only 13% name vector databases: employers want the skill, not a specific product. A Legora ad also asked for judgement on when a simpler approach is better.
5. Agents and tools. Function calling, MCP (22%), orchestration frameworks such as LangGraph (24% name one of the main frameworks). The Hugging Face Agents Course is free, with a free certificate, and assumes basic Python and LLM knowledge. Browse agentic AI jobs to see how this gets used.
6. Production. Cloud (30%), observability, guardrails, security.
7. Later, optional: fine-tuning and PyTorch. Fine-tuning or post-training appears in 23% of ads and PyTorch in 13%, mostly in post-training and quality roles. This is where AI engineering meets ML engineering.

Many "AI engineer roadmaps" start with calculus, classic machine learning and deep learning, and reach LLM apps only months later. That is an ML engineer roadmap. The job ads for AI engineers don't read that way.

You need enough math to reason about embeddings, sampling and evaluation statistics, not much more. For broader reading, DeepLearning.AI short courses are one- to two-hour courses, many built with the model providers, and roadmap.sh's AI engineer roadmap is a useful checklist.

How long does it take to become an AI engineer?

It depends on where you start. From software engineering, it can take a few months. From scratch, plan on a software job first, because the ads ask for years of engineering experience.

These are our estimates, based on the years the ads ask for. From software engineering, your years count, so a few solid LLM projects can be enough. From data science, your strength is evals and data; the gap is shipping product.

Course providers sell shorter timelines. Dataquest estimates 8 to 12 months from scratch at about 5 hours a week. That may be enough to learn the skills, but we think it oversells the market: almost no posting we read was open to someone without professional experience. That doesn't close the door. It means your first target is probably a software engineering job.

Can you become an AI engineer without a degree?

Yes. Most AI engineer ads don't mention a degree at all, and many of those that do accept equivalent experience.

Method: Tokn fetched the full text of all 95 postings from the companies' public ATS boards on 27 October 2026, read 92 of them and classified the degree wording by hand.

OpenAI's 24 applied AI engineer ads mention no degree. Anthropic's all say "Bachelor's degree or an equivalent combination of education, training, and/or experience". Six of the 10 degree-required postings are at Mistral AI, whose technical-lead forward deployed roles ask for a master's or PhD as well as 7 to 8+ years. That is too small a sample to say Europe is stricter.

Certifications don't replace experience either. See the AI certifications worth your time before paying for one.

What to build: three projects that match the job ads

Build things that show the skills the ads mention most: evals, retrieval, agents and production.

• A RAG app over a real document set, with an eval report. Measure retrieval and answer quality, and show what improved.
• An agent with tools. Include an MCP server, guardrails and logged traces, so a reviewer can see how it behaves when things go wrong.
• A shipped product with real users. Track cost and latency, and write up the decisions you made.

About 37% of the ads mention GitHub, open source, side projects or a portfolio. For the details, read how to build an AI portfolio. If you want an AI coding tool to speed up the building, our guide to getting started with Claude Code covers the basics.

How to land your first AI engineer role

The most realistic route is from inside: an AI project at your current software job. After that come early-stage startups, customer-facing applied AI roles and junior roles.

• Internal move. Volunteer for your company's LLM feature. It gives you the production LLM experience the ads ask for.
• Startups. Founding and early roles can ask for less. Plato's founding AI engineer ad asked for 2+ years and said "You don't need to master every tool".
• Applied AI and forward deployed roles. Common at the labs. They reward engineers who are comfortable with customers.
• Junior roles and internships. Rare under the AI engineer title, so look at entry-level AI jobs more broadly, including data and ML.

Expect system design for LLM apps, evals and pipeline debugging in interviews; our machine learning interview questions guide covers the more model-focused side. For pay, read our breakdown of AI engineer salary in the US and Europe.

Find AI engineer jobs on Tokn

Tokn lists AI engineer roles from company career pages in Europe and North America. Browse open AI engineer jobs, or entry-level AI jobs if you are starting out. To hear about new roles first, get AI engineer jobs by email.

FAQ

What is the difference between an AI engineer and a machine learning engineer?

An AI engineer builds products on existing models: APIs, retrieval, agents and evals. A machine learning engineer trains, fine-tunes and serves the models. Titles overlap, so read the responsibilities. Our guide on how to become a machine learning engineer covers the other path.

Are there entry-level AI engineer jobs?

Very few. On 27 October 2026, only a handful of AI engineer postings on Tokn, all in Europe, were open to people with about 2 years of experience or less, and none in North America. Most beginners start as software engineers, or in junior data and ML roles, and move across.

What programming language should an AI engineer learn?

Python first: 72% of the AI engineer postings we read mention it. TypeScript or JavaScript next, at 54%, because AI engineers ship product features, not just models.

Is AI engineering a good career in 2026?

The demand is there. LinkedIn's Jobs on the Rise 2026 list, as reported by Social Media Today, ranks AI engineer first among the fastest-growing US roles, based on members' job titles rather than postings. On Tokn, we counted 95 open AI engineer roles at 32 companies on 27 October 2026. For pay, see AI engineer salary in the US and Europe.