AI Training 2026 for Developers and Engineers in France: Schools, Bootcamps, Certifications and Funding

For a developer or an engineer, training in AI in 2026 runs through 4 paths: a degree, a certifying intensive program, an online course, or a cloud vendor certification. The French state funds 9 AI research and training hubs, the IA Clusters, to the tune of €360M (France 2030, ANR, May 2024). Since February 2026, the CPF (Compte personnel de formation, France's individual training account) caps funding at €1,500 for certifications on the Répertoire spécifique (RS, register of specific certifications), not for RNCP (France's national register of professional certifications) titles, and every enrollment costs a €150 co-payment since 2 April 2026 (Decrees No. 2026-127 and No. 2026-234). In the AI hiring we do, what a candidate has shipped to production outweighs the name of their training.

The AI engineer role took shape with generative AI, in 2023 and 2024. Training supply followed, at every level: master's degrees, bootcamps, online courses, vendor certifications. For a developer who already codes, the real difficulty is knowing which AI engineer training will actually weigh in a hiring decision, and how to fund it.

This resource maps the recognized paths in France, with their level, duration, cost and funding. It adds what we observe recruiting ML engineers, AI engineers and research engineers for AI start-ups and labs.

Key takeaways

• 4 paths lead to AI roles for a technical profile: a degree (master's, engineering school, PhD), a certifying intensive program, an online course, a cloud vendor certification. They do not carry equal weight with a recruiter.
• The French state structures the academic offering around 9 IA Clusters (state-backed AI research and training hubs), funded with €360M by France 2030 (ANR, May 2024).
• The CPF now distinguishes RNCP and Répertoire spécifique: since 26 February 2026, RS certifications are capped at €1,500 of funding rights, RNCP titles are not (Decree No. 2026-127).
• Every CPF-funded enrollment costs €150 as a co-payment since 2 April 2026, with exceptions (Decree No. 2026-234, Service Public).
• Cloud AI certifications change fast: AWS retired its Machine Learning Specialty certification on 31 March 2026, Microsoft retired the Azure Data Scientist Associate (DP-100) on 1 June and the Azure AI Engineer Associate (AI-102) on 30 June 2026, replaced by the AI-103.
• A France Compétences (France's certification registry authority) record should be checked before enrolling: its status (active or not) and its expiry date tell you whether the title will still be awarded.
• In the AI hiring we do, a system shipped to production weighs more than the name of the training: evaluations, costs, monitoring, choices defended in an interview, the ability to judge code produced by an agent.

Methodology

What this resource covers. Training that leads to the roles of AI engineer, ML engineer, LLM engineer and MLOps engineer, for developers, engineers and computer science students. It also speaks to managers who fund their team's upskilling.

What it does not cover. General-public AI literacy, AI use in non-technical roles, and training abroad, apart from online courses. Training for the data analyst role has its own page: data analyst training.

Inclusion criteria. A national or state-recognized degree, a title registered with the RNCP or a certification on the Répertoire spécifique, a public and detailed program, a Qualiopi-certified provider (France's training-quality certification label) for fundable programs. For the academic offering, we start from the 9 state-labeled IA Clusters.

Presentation rules. Providers are listed alphabetically, with no ranking. Durations, prices and placement rates are those announced by each provider, checked in November 2026 and attributed as such. This page contains no figures from Bluecoders' own placements: our consultants' input here is qualitative.

What are the different AI training paths for a developer?

Sources: ANR 2024, France Compétences, provider program pages, checked in November 2026, Decree No. 2026-127.

What AI skills are recruiters asking for in 2026?

Digital, engineering and consulting firms are already hiring for roles dedicated to AI. According to the BVA survey for OPIIEC (2024, published April 2025), 28% of firms in the sector have created roles dedicated to AI, and close to 45,000 jobs could be created within 3 years under AI's influence.

To build AI solutions, OPIIEC identifies 6 expected skills: data science, data governance, mathematics, infrastructure, monitoring AI solutions, and AI project management. That list already sketches the spec sheet of a serious training program.

The same report adds a useful nuance before you choose. Among initial training programs that fold in AI skills, only 25% lead to AI-specific roles. The rest simply add an AI tint to a cross-functional role. A program's "AI" label therefore says nothing about whether it trains AI engineers.

What AI roles are emerging in tech teams?

With AI, we see tech teams tightening: fewer levels, roles hybridizing, an AI agent next to every engineer. When AI writes part of the code, value shifts to decisions, specs, review and evals. The roles showing up in our searches reflect that:

• AI Engineer and Applied AI Engineer, who ship AI features to production from existing models;
• Agent Engineer and LLM Engineer, specialized in language models and their orchestration;
• MLOps Engineer and AI Platform Engineer, who keep models running over time;
• Forward Deployed Engineer and Evals & AI Quality, two titles still rare, that we see emerging.

In defense and industry, the shift is slower: clearance, safety and certification requirements keep teams more structured.

What technical skills keep showing up in job postings?

For an AI engineer role, the expected foundation is that of a solid backend engineer: Python, code structure, tests, deployment. Model-specific skills are layered on top:

• the APIs of the major LLMs and open source models;
• RAG architectures and vector databases;
• evaluation of outputs (eval sets, regression tests, LLM-as-judge);
• cost and latency optimization;
• production monitoring: drift, hallucinations, user feedback.

For an ML engineer or MLOps engineer role, the center of gravity shifts toward training, versioning and shipping models to production (CI/CD, containers, orchestration).

How are coding agents changing the engineer's job?

In 2026, a scale-up's head of engineering described a team where reviewing agent-generated code has nearly disappeared. It now happens upstream, on specs and design. The big undertaking becomes the harness: structuring the code, linting and processes so agents are effective, and so several engineers each steering their own agents don't get in each other's way.

The profile in demand changes accordingly: a product engineer with real architectural judgment, able to assess what an agent produces. A weak engineer equipped with agents produces much more bad code, and faster. Before choosing a training program, check whether it has you write specs, review generated code and structure a repository, not just train models.

Source: OPIIEC, "Les besoins en compétences, emplois et formations en matière d'intelligence artificielle en France," April 2025 summary, BVA 2024 survey.

What level should you aim for, based on your profile?

The right entry point depends on what you can already code, not on your job title. Serious training programs set explicit prerequisites: working Python, basics in linear algebra and probability, technical English. The RNCP title "Architecte en intelligence artificielle" (RNCP41993), for example, requires level 6 (3 years of post-secondary study) and an admission test in SQL, Python and IT infrastructure.

You are a computer science student

You have time for a degree-granting track: a master's or an AI specialization at an engineering school, ideally tied to one of the IA Clusters. It is also the easiest time to land a lab or R&D team internship. A deployed, documented project on GitHub will be worth more than one extra elective.

You are a developer or engineer already in a job

You already have half of the AI engineer role: code, tests, shipping to production. What you lack is the model layer: machine learning, deep learning, RAG, agents, output evaluation. A certifying intensive program of a few hundred hours, part-time or funded by your employer, covers that gap without restarting a 2-year degree.

You come from data or a scientific discipline

A data scientist or a PhD in applied mathematics already masters the models. A data engineer already masters the pipelines. The path to ML engineer or MLOps engineer runs through engineering: containers, CI/CD, orchestration, monitoring. A short, targeted MLOps specialization is often more useful than a new general-purpose degree.

You manage a team you want to upskill

Start from your product's use cases, not a provider's catalog. A team folding LLMs into a SaaS product doesn't have the same needs as a team training its own models. Your company's skills development plan and your OPCO (sector training fund) can fund this kind of path (see the funding section).

AI engineer training: which schools and master's programs to choose

For a degree-granting AI engineer training, the most reliable benchmark is the map of IA Clusters. In May 2024, the French state selected 9 AI research and training hubs under France 2030, funded with €360M in total. Their mandate includes sharply increasing the number of AI graduates, at every level.

What are the 9 state-labeled IA Clusters?

Source: ANR, press release of 28 May 2024 on the winners of the "IA Cluster" call for projects. Alphabetical order.

These hubs bring together universities, schools and research bodies. The ANR's call for projects also asks them to attract international students and build industry partnerships, which brings their programs closer to company needs.

Which master's and engineering degrees should you target?

Two examples illustrate what the top-tier offering looks like:

• The IASD master's (Intelligence artificielle, systèmes, données, i.e. artificial intelligence, systems, data) at Université PSL, run jointly by Dauphine, the ENS and Mines Paris and tied to PR[AI]RIE. According to PSL, it enrolls around sixty students a year and is also available as a work-study program.
• The M2 MVA (Mathématiques, vision, apprentissage, i.e. mathematics, vision, learning), created in 1996 by the mathematics department of ENS Paris-Saclay. Université Paris-Saclay states an intake capacity of 220 places.

Both tracks target very mathematical profiles, often headed toward R&D or a PhD. For a developer aiming at AI engineering, AI specializations at engineering schools and work-study master's programs are often closer to the ground. To compare the university offering, the national Mon Master platform (France's national master's application platform) lists every national master's degree and centralizes first-year applications.

Do you need a PhD to work in AI?

No for an AI engineer or MLOps engineer role, often yes for a research role. We recruit PhD holders for research scientist, research engineer or R&D lead roles, mainly at lab-born start-ups and corporate R&D teams. For these roles, we weigh the scientific track record as much as the ability to move from research to product. A CIFRE thesis, carried out inside a company, often checks both boxes.

What AI bootcamps and intensive programs are recognized?

An AI bootcamp only carries title value if it prepares for a certification registered with the RNCP, in full or through skill blocks. That is also what makes it fundable through the CPF with no specific cap since 2026. The 3 providers below meet that criterion, with different positioning.

Source: provider program pages and France Compétences records, checked in November 2026. Prices, durations and sessions as announced by each provider, to verify before enrolling.

AI engineer training: which one to choose as a developer?

An AI engineer training that's useful to a developer goes beyond model training. Check the program for 3 elements: current generative AI (RAG, agents, output evaluation), deployment and production monitoring, and a complete end-to-end project. Jedha's AI Engineer track, for instance, devotes a module to AI engineering (Docker, deployment, monitoring, continuous evaluation) and ends with a final project.

MLOps and LLMOps training: for whom?

An MLOps training targets profiles who already master machine learning. Jedha requires 2 years of experience in data science or AI for its MLOps & LLMOps program, which covers governance (GDPR, the AI Act), GPU computing, infrastructure as code and securing generative systems. Liora has you choose between an MLOps specialization and an LLMOps specialization starting at sprint 3, with a project of around 80 hours mentored by a coach. LLMOps adds to classic MLOps the topics specific to language models: response quality, token cost, prompt injection.

Machine learning and data science training: the foundation before specializing

For a profile starting from zero in modeling, a general machine learning training remains the entry point: statistics, supervised and unsupervised learning, deep learning. Le Wagon's Data Science & IA bootcamp plays that role for profiles already familiar with programming and mathematics, going as far as machine learning engineering and deployment. LLM or MLOps specialization comes next.

What LLM and machine learning courses online are worth the detour?

Several reference online courses are free and written by the people building the tools, but they award no formal title and are not CPF-fundable. They serve to consolidate a foundation, test a topic before investing in a long program, or follow a technology that moves faster than certifying curricula.

Which courses from recognized institutions should you take?

Inria's MOOC "Machine learning in Python with scikit-learn," hosted on FUN MOOC, is entirely free. It is designed by lead scikit-learn developers, including one of its creators, and requires Python basics. It is a good test before a paid machine learning training: if cross-validation and hyperparameter tuning concepts resist you, better to know it before signing up.

What LLM training online suits a developer?

For a free LLM training, Hugging Face's Learn platform offers an LLM course, an Agents course and a course on the Model Context Protocol (MCP). The Agents course covers the smolagents, LangGraph and LlamaIndex frameworks, an agentic RAG use case and a bonus module on agent observability and evaluation. According to Hugging Face, it leads to a certification after completing a use case and an evaluation.

These courses have an advantage few certifying programs offer: they track the tools actually used by teams shipping agents to production. Their limit is the lack of mentoring: without a personal project attached, they are rarely finished.

What AI certifications make a difference to recruiters?

A cloud vendor certification attests mastery of a stack, not the ability to ship an AI system. It carries weight when the target role runs on that stack, and almost none when it replaces an actual project. Before preparing for an exam, also check that it still exists: 3 reference certifications disappeared in 2026.

Which cloud AI certifications are valid in 2026?

Sources: official AWS, Google Cloud and Microsoft Learn pages, including the list of retired Microsoft exams, checked in November 2026.

Why did AI certifications disappear in 2026?

Vendors rebuilt their tracks around generative AI and agents. AWS now points toward its ML Engineer Associate certification, centered on shipping models to production. Microsoft replaced the AI-102 with the AI-103, built around Microsoft Foundry, generative applications and agents. Google folded generative AI (Model Garden, Vertex AI Agent Builder) into the Professional ML Engineer exam.

For a candidate, the consequence is simple: a retired certification stays on your resume, but it dates your profile. Better to present the current version, or a recent project on the same stack.

What our consultants actually look at

When we assess an AI profile, we first look at what has been built, published and shipped to production. A thesis, publications or a production model say more than a training program's name. For an AI engineer role, 4 signals keep coming up in our interviews:

• a deployed system, even a modest one, with accessible code and explained architecture choices;
• an evaluation practice: how the candidate measures response quality, detects a regression, trades off cost against latency;
• the ability to write good specs and judge code produced by an agent: we test it during qualification, on the same footing as the stack;
• the ability to defend their choices in front of a technical team.

Training mostly counts as proof of a foundation: a level-7 RNCP title or a master's reassures on fundamentals, a cloud certification on a stack. To put these elements forward, see the 5 things tech recruiters look at on a developer's resume.

How do you choose your AI training?

Compare training programs on what they make you produce, not on what they promise. 6 criteria are enough to sort out what matters:

1. Prerequisites. A program that accepts everyone into an AI engineer curriculum will have to spend time on basics. If you already code, look for a program that assumes Python is acquired.
2. Recognition. National degree, full RNCP title, RNCP skill blocks, Répertoire spécifique certification, or a plain certificate: this choice determines the value of the outcome and how it can be funded.
3. The production share. Deployment, monitoring, output evaluation, costs: if the program stops at training models, it prepares you for data science more than for AI engineering.
4. The final project. A complete system, deployed and presentable in an interview, is worth more than a series of exercises.
5. Working with coding agents. Writing specs, reviewing and judging generated code, structuring a repository so agents can work in it cleanly: tech teams now expect this. If the program doesn't cover it, work on it through a personal project.
6. The format. Full-time or part-time, in-person or remote, work-study: the right format is the one you'll see through to the end.

How do you check that a training program is genuinely certifying?

Every RNCP title and every RS certification has a public record on the France Compétences website. Before enrolling, check 4 pieces of information:

• the record's status: "active" or not;
• the registration's expiry date: the RNCP41993 record "Architecte en intelligence artificielle," for example, is registered until 27 February 2029;
• the certifying body: the organization that actually awards the title, not necessarily the one delivering the training;
• placement statistics published on the record, with the number of certified graduates per year and the placement rate in the targeted role.

Also check whether the training prepares you for the full title or only certain blocks. A track that validates 2 blocks out of 4 is certifying and fundable, but it does not award the full title. The full title can require additional conditions: for RNCP38587 prepared by Liora, for example, a work placement of at least 130 days.

How do you fund your AI training with the CPF, an OPCO or France Travail?

Funding depends first on the type of certification targeted, then on your status. The rules changed in 2026: an RNCP title and a Répertoire spécifique certification are no longer funded the same way through the CPF.

Is an AI training eligible for CPF funding in 2026?

Yes, if it prepares for a certification registered with the RNCP or the Répertoire spécifique and it is listed on Mon Compte Formation. Three rules have applied since 2026:

Sources: Légifrance (France's 2026 Finance Act), Service Public (news item of 2 April 2026), Centre Inffo (20 and 26 February 2026).

In practice, an MLOps training that prepares for blocks of an RNCP title stays fundable within the limit of your balance. Preparing for an RS-registered certification is capped at €1,500 in funding rights drawn from the account's annual top-up, employer contributions not included. Jobseekers do not pay the flat co-payment, and an employer contribution also exempts you from it.

How do you get your training funded by your employer or your OPCO?

If you're an employee, the most direct route is your company's skills development plan, partly funded by its OPCO (sector training fund). France Compétences offers an online tool to identify which OPCO your company belongs to. An employer can also top up your CPF: that contribution falls outside the 2026 caps and exempts you from the flat co-payment.

On the manager's side, training a developer who already knows your product and your stack is an alternative worth weighing before launching a search for a scarce AI profile. The two often combine: a senior profile hired to set the architecture, developers trained to keep it running. That's the logic we hear from the CTOs and heads of engineering we work with: hire better rather than hire more.

What support exists for a career change or a jobseeker?

• The professional transition project (PTP), funded by your region's Transitions Pro association (the regional body funding professional transition projects), lets an employee take a certifying training to change careers. According to the Ministry of Labour, pay is fully maintained up to 2 times the Smic (France's minimum wage) during the training.
• France Travail's individual training grant (AIF) can top up funding for a jobseeker.

These schemes can complement the CPF. Check each condition with Transitions Pro, France Travail or on Service Public before committing: they evolve from one year to the next.

Bluecoders' take

"On an AI engineer search, we ask the candidate to show us a system they've shipped, then to explain how they know it works. Their training tells us what they learned. How they evaluate their own outputs, and judge the code an agent produces for them, tells us whether they're ready for production. The teams we work with review less and less code and more and more specs: that judgment is what they're hiring for."

Cécilia Fille, COO & Partner at Bluecoders

Are you hiring, or targeting an AI role?

Are you an engineer or developer targeting an AI engineer, ML engineer or MLOps engineer role? Show us what you've built: a recruiter specialized in your field reads every application and replies, even when it's a no. Submit my application

Are you building an AI team and torn between training and hiring? Our AI, Data & Sciences practice, led by Géraud Mesnard, recruits the ML engineers, AI engineers and researchers of AI start-ups and R&D teams. Let's talk about your hiring

How to cite this resource

Reusing the information on this page is allowed, with a link back to the page. Citation format:

Bluecoders, AI Training 2026 for Developers and Engineers in France: Schools, Bootcamps, Certifications and Funding, 2026 edition, sources consulted in November 2026. https://www.bluecoders.com/en/ressources/formations-ia-developpeurs

Journalists: our validated figures and press contacts are gathered in our press kit.

FAQ

What training should I take to become an AI engineer as a developer?

A developer already has the software foundation of the AI engineer role. What's missing is the model layer: machine learning, deep learning, RAG, agents and output evaluation. A certifying intensive program of a few hundred hours, tied to an RNCP title, covers that gap. Choose a program that goes all the way to deployment and monitoring, with a complete project to present in an interview.

What LLM training should a developer in France take?

To start at no cost, Hugging Face's LLM and Agents courses cover language models, agent frameworks (smolagents, LangGraph, LlamaIndex) and evaluation. For a certifying, fundable path, AI engineer or LLMOps training programs that prepare for blocks of an RNCP title build LLMs, RAG and agents into their curriculum.

Is an MLOps training eligible for CPF funding?

Yes, if it prepares for an RNCP title or some of its blocks, and it is listed on Mon Compte Formation. Since 26 February 2026, RNCP titles are not subject to the €1,500 cap that applies to Répertoire spécifique certifications. Since 2 April 2026, every enrollment costs a €150 co-payment, except for jobseekers and when an employer contributes.

Do you need a master's degree to become an AI engineer?

No for an AI engineer or MLOps engineer role: recruiters look first at what the candidate has built and shipped to production. A master's or a level-7 RNCP title reassures on fundamentals. For a research scientist or research engineer role, a research master's or a PhD is often expected.

Which AI certifications add value to a developer's resume?

The cloud certifications active in November 2026 are the AWS Machine Learning Engineer Associate, the Google Cloud Professional Machine Learning Engineer and the Microsoft Azure AI Apps and Agents Developer Associate (AI-103). They carry weight when the role runs on the same stack. The AWS Machine Learning Specialty, the Azure Data Scientist Associate (DP-100) and the Azure AI Engineer Associate (AI-102) were retired in 2026.

What skills should a developer work on to code with AI agents?

Five skills keep coming up in the hiring we do: writing precise specs, reasoning in architecture, judging code produced by an agent, structuring a repository for agents (code organization, linting, conventions) and evaluating the outputs of an AI system. In teams working this way, review shifts from code toward specs and design. A developer who can frame and judge an agent's work becomes more productive; one who can't produces more bad code.

How long does it take to train in AI when you already code?

Count on a few weeks to a few months depending on the goal. Inria's scikit-learn MOOC represents about 36 hours according to FUN. The intensive programs cited in this resource range from 150 hours (MLOps specialization) to 450 hours (AI engineer track), i.e. 1 to 3 months full-time or 10 weeks to 7 months part-time. A specialized master's degree takes 1 to 2 years.

Can an AWS or Azure certification be funded through the CPF?

Only if the training that prepares for it is registered on the Répertoire spécifique or built into an RNCP track, and listed on Mon Compte Formation. In the first case, the CPF covers a maximum of €1,500 in funding rights since 26 February 2026. Taking the exam itself is paid directly to the vendor.

How much does an AI engineer training cost?

Inria's and Hugging Face's online courses are free. The certifying intensive programs checked in November 2026 range from €4,000 to €7,500 for those that disclose their price. A public master's degree carries national enrollment fees, a private school its own rates. The CPF, the employer's OPCO, France Travail's AIF, or a professional transition project can fund all or part of these costs.

Sources

• ANR, Intelligence artificielle : 9 nouveaux « IA Clusters » et 2 nouveaux lauréats « Compétences et métiers d'avenir » soutenus par l'État au titre de France 2030, press release of 28 May 2024. Link
• Légifrance, Loi n° 2026-103 du 19 février 2026 de finances pour 2026, Article 203. Link
• Centre Inffo, Limitation de l'utilisation des droits CPF par la loi de finances pour 2026, 20 February 2026. Link
• Centre Inffo, CPF : conditions de financement des bilans de compétences, du permis de conduire et des formations menant à une certification du répertoire spécifique (Decrees No. 2026-126 and No. 2026-127 of 24 February 2026), 26 February 2026. Link
• Service Public, CPF : la participation forfaitaire obligatoire augmente (Decree No. 2026-234 of 30 March 2026), updated 2 April 2026. Link
• Ministère du Travail, Le projet de transition professionnelle, fact sheet. Link
• Ministère de l'Enseignement supérieur et de la Recherche, Mon Master, plateforme nationale de candidature. Link
• AWS, AWS Certified Machine Learning Specialty, official page. Link
• Microsoft Learn, Microsoft Certified: Azure AI Apps and Agents Developer Associate (exam AI-103). Link
• Microsoft Learn, Retired certification exams (AI-102, DP-100). Link
• Google Cloud, Professional Machine Learning Engineer, official page. Link
• OPIIEC, Les besoins en compétences, emplois et formations en matière d'intelligence artificielle en France, "À retenir" summary of 28 April 2025, BVA 2024 survey. Link
• France Compétences, record RNCP41993, Architecte en intelligence artificielle. Link
• France Compétences, record RNCP38616, Concepteur développeur en intelligence artificielle et analyse big data. Link
• Université Paris Dauphine - PSL, Master 2 Intelligence artificielle, systèmes, données (IASD). Link
• Université Paris-Saclay, M2 Mathématiques, vision, apprentissage. Link
• Jedha, Formation AI Engineer and Formation MLOps & LLMOps, program pages. Link · Link
• Le Wagon, Formation Data Science & IA, program page. Link
• Liora (formerly DataScientest), Formation MLOps / LLMOps, program page. Link
• Inria, Machine learning in Python with scikit-learn, FUN MOOC. Link
• Hugging Face, Hugging Face Learn (LLM, Agents and MCP courses). Link