How to get a job in AI in 2026: a step-by-step guide

To get a job in AI, pick one specific path, learn the skills that postings for that path actually list, and prove them with work that runs in the real world. Then find openings on company career pages, apply with evidence, prepare for interviews that test judgement as well as code, and negotiate the offer. The steps are the same for students, career changers and software engineers. Only the starting point changes.

This guide is built on the roughly 770 active AI jobs at 79 companies listed on Tokn as of 12 October 2026, all taken from the companies' own career pages. Each step shows what those postings ask for.

In short

• Python appears in 51% of AI job postings on Tokn. No other skill comes close.
• Shipping beats credentials: at least 41% of requirement lists ask for production or shipped work, about 20% mention a degree and 7% a PhD, mostly in research.
• Entry level is narrow: only about 9% of jobs are junior roles or internships, including 40 internships at 18 companies, 23 of them in Paris.
• Where a minimum is stated, the median is 5 years of experience. Apply anyway when you can show the work.
• Companies post on their own career pages first. Watch those pages and set alerts.

Can you still get a job in AI in 2026?

Yes, you can still break into AI. Demand is real, but most of it sits above entry level. According to LinkedIn's Jobs on the Rise ranking, as reported by Euronews, AI Engineer was the fastest-growing job in the UK, France and Spain, and Head of AI topped the list in Germany. That ranking covers hires from January 2023 to July 2025.

AI work is also spreading beyond tech teams. Indeed data analysed by Euronews shows that most AI-labelled job titles sit outside tech occupations: 59% in Germany, 54% in France and 54% in the UK. That is good news if you come from marketing, operations or consulting.

The catch is seniority. On Tokn, a third of postings carry no level. Among those that do, about seven in ten are senior or lead roles, and 14% are junior roles or internships.

Step 1: Pick a path, not "AI"

"AI" is not a job title. Pick one family of roles that matches what you already do well, because each one asks for different skills and a different kind of proof. Our overview of the 8 AI career paths describes each in depth; here is the short version with live counts.

Active jobs on Tokn as of 12 October 2026.

Which path fits your starting point

• Student in CS, maths or engineering: machine learning, LLM engineering, or research if you are heading for a PhD. Internships are your main way in (see Step 4).
• Software engineer: LLM and GenAI engineering or MLOps is the shortest move. You already have the production habits employers ask for; add LLMs, retrieval and evaluation.
• Career changer or domain expert: AI product, design, marketing and content together account for 138 jobs on Tokn (18%). Deployment and solutions roles also value people who know the customer's field.

We would not plan a career around "prompt engineer": some guides still sell it as an entry job, but it does not stand out as a title in the postings we track.

Step 2: Build the skills employers list

Learn Python first, then the stack of your chosen path. Python appears in about half of the jobs on Tokn (51%); the next skill, AWS, appears in 17%.

Skills extracted from active postings on Tokn, as of 12 October 2026.

The pattern matches the wider market: the Stanford AI Index 2026 data summarised by Lightcast also lists Python as the top specialised skill, based on US job postings. After Python, choose by path:

• ML engineering and research: PyTorch, which appears in 117 postings against 33 for TensorFlow.
• MLOps and infrastructure: Kubernetes, Terraform and one cloud (AWS, GCP or Azure).
• LLM product engineering: TypeScript alongside Python, plus RAG and agent frameworks.
• Data science: SQL and statistics.

Good free starting points: the Hugging Face LLM Course, fast.ai's Practical Deep Learning, and for complete beginners the University of Leeds' short How to Get Into AI course on Coursera.

The 2026 additions: LLMs, evals and AI tools

Three newer themes show up across requirement lists. The shares below are lower bounds, based on up to five summarised requirement lines for about two thirds of the jobs.

• LLMs, GenAI, RAG or agents: mentioned in at least 30% of requirement lists.
• Evaluation ("evals"): at least 11%.
• Using AI tools yourself: at least 7% ask that you already work with coding agents or AI tools. A Brevo product engineering posting in Paris put it plainly: "Coding agents already central to how you work. Claude Code, Cursor, Codex, something equivalent."

For a closer look at the LLM side, browse the LLM and GenAI jobs on Tokn.

Do you need a degree, a master's or a PhD?

For most AI jobs, no PhD is needed, and a degree is mentioned in only about one in five requirement lists. Research is the exception.

Jobs on Tokn with an extracted requirements list, as of 12 October 2026. Lower bounds.

Degree mentions cluster in two places: client-facing consulting and solutions roles, and French internships that ask for a Bac+5 or an engineering school. Even in research, several requirement lines add "or equivalent research experience". Across all jobs, at least 9% of requirement lists say "or equivalent experience" outright.

We think the time you might spend on a certificate is usually better spent on a project, because postings ask for shipped work far more often than for any credential.

Step 3: Prove it with projects that look like production

Build one or two things that actually run, have users or a measured result, and are written up. "Production", "shipped" or "deployed" appears in at least 41% of requirement lists, twice as often as any degree. In LLM and GenAI roles the gap is 65% against 8%.

Employers spell this out. A Shift Technology graduate data scientist posting asked for "having developed some apps with actual users – writing scripts for models and notebooks is not enough". A Photoroom ML posting wanted "You've shipped ML into production". Granola asked for "work you can show us".

What a good AI portfolio project looks like

• It is deployed. Someone can use it through a URL, an API or a package, not just a notebook.
• It is measured. An evaluation set, a benchmark, latency and cost numbers, or real usage.
• It has a short write-up. The problem, what you tried, what failed, what you would do next.
• The code is public and readable, with a README that lets someone run it in minutes.

Open source and public work

Only 6% of requirement lists name a portfolio, GitHub or open source explicitly. But "track record" or "proven" appears in at least 28%, and "demonstrated" or "work you can show" in another 8%. A merged contribution to a well-known library is shipped work that strangers reviewed: strong evidence.

Step 4: Find the openings where companies post them

AI companies publish jobs on their own career pages first, usually through an applicant tracking system (ATS). Other job boards copy those listings, sometimes late. Tokn reads the career pages directly, every day. Most of the career pages we track run on Ashby or Greenhouse, with Lever third.

Make a shortlist of employers whose products you would happily build, then check their career pages, or follow them on our list of companies hiring in AI.

Internships, new-grad roles and graduate programmes

There are 40 internships on Tokn at 18 companies, and 24 of them are in France (23 in Paris). The UK has 8, Germany 4, Spain and the Netherlands 2 each. We list no apprenticeships at the moment.

Tokn, as of 12 October 2026. Nine other companies have one internship each.

Timing matters. Many internships are already for 2027 (winter 2027, summer 2027, February starts), so apply four to nine months ahead. Some research internships require you to be enrolled in a PhD.

Junior permanent roles are rarer: 29 in total, and 14 of those belong to one graduate scheme, Celonis's Orbit programme for associate AI consultants and value engineers, mostly based in Madrid or Munich. To find the rest, search titles for "new grad", "graduate", "associate" or "Engineer I". Most internships are in Paris, so start with AI jobs in Paris.

Remote AI jobs: what "remote" really means

About one job in six on Tokn is fully remote (132 of about 770, as of 12 October 2026), and those roles are usually remote within a given country rather than from anywhere. Hybrid is more common (28%), and 40% of postings state no policy. The big labs, OpenAI, Anthropic and Mistral AI, hire almost entirely on-site or hybrid; remote-first employers include ElevenLabs, Cohere and n8n. Browse remote AI jobs and read the location line before applying.

Set alerts so you apply early

Postings can close without notice. Set up a job alert for your path and city to hear about new ones as soon as we pick them up.

Step 5: Apply so a human reads it

Mirror the posting's requirement lines, lead with shipped work and numbers, and apply even if you are short on years. Map each requirement line to one piece of proof.

Years of experience are stated on only a little over a third of jobs (37%). Where they are, the median is 5 years, and no job asks for 0.

Tokn, as of 12 October 2026, 278 jobs with a stated minimum.

For junior roles that state a minimum, the median is 1 year (25 jobs). That year can come from internships, freelance work or a serious project with users. If you meet most lines and can show strong work on the core one, apply.

• Put two or three shipped projects or results at the top, each with a number.
• Link the deployed project and the code, not just a GitHub profile.

Can you use AI to write your application?

Yes, but we think you should write the first draft yourself and use AI to tighten it, because a generic AI-written letter reads like hundreds of others. Some companies publish rules on AI use during hiring, especially for take-home tests and live interviews, so check the careers page before you start.

Applying across Europe

English is the working language in most postings. French appears in 12 requirement lists and German in 5, mostly for client-facing roles in those markets. If you need a work permit, check the posting's right-to-work line first.

Step 6: Prepare for AI interviews

Prepare for four things: coding, system design for ML or LLM systems, a deep-dive on your own project, and behavioural questions. Ask the recruiter what the loop looks like.

• Coding: practise practical Python on data and model tasks, not only algorithm puzzles.
• System design: how you would build, evaluate and serve an ML or LLM feature, including cost and latency.
• Project deep-dive: expect hard questions on your choices. This is where a real, deployed project pays off.
• Behavioural: requirement lists mention communication or stakeholders in at least 26% of cases, ownership or ambiguity in 21%, and customer-facing work in 15%. Prepare a story for each.

Step 7: Negotiate the offer

Know your level, get the range early and compare the whole package. Only about one in eight jobs on Tokn shows a salary, so ask for the range in the first recruiter call.

• Check the level. The same title can be junior or senior at different companies, and the level drives the pay band.
• Compare total compensation, including equity at startups, not just base salary.
• Negotiate more than money: start date, remote days, learning budget and title.

How to get a job in AI with Tokn

Start with the path you picked in Step 1: machine learning jobs, LLM and GenAI engineer jobs, AI research jobs or AI deployment and solutions roles. Then create a job alert so that new openings from company career pages reach you the day we find them.

FAQ

Can I get a job in AI with no experience?

Yes, but the door is narrow. On Tokn, as of 12 October 2026, about 9% of jobs are junior roles or internships, and nearly half of the 29 junior permanent roles belong to one Celonis programme. Internships, graduate schemes and adjacent roles are the realistic routes, backed by a project that shows real work.

Do I need a PhD to work in AI?

Only for research. A PhD is mentioned in 38% of research requirement lists on Tokn, against 7% across all AI jobs. Other roles care more about shipped work.

What programming language should I learn for AI?

Python. It appears in 51% of AI job postings on Tokn. Add TypeScript if you want to build AI products, and SQL if you are heading towards data science.

Which AI jobs don't require coding?

AI product, design, marketing and content roles, which make up 138 jobs on Tokn (18%). Many deployment and solutions roles are client-facing too, though they expect you to be comfortable with technical detail.

Can I work in AI remotely from Europe?

Partly. About one job in six on Tokn is fully remote, usually within a given country, and another 28% are hybrid. Read the location line carefully: "remote, France" usually means you are expected to live and work in France.

Do I need to speak French or German to work in AI in Paris or Berlin?

Usually not. English is the working language in most postings, and only a handful of requirement lists on Tokn mention French (12) or German (5), mostly for client-facing roles.