The landscape of artificial intelligence experts has been reconfigured since the explosion of foundation models. The profiles that matter in 2026 are no longer just researchers publishing in academic conferences: we observe a hybrid core where startup founders, open source infrastructure maintainers, and algorithmic ethics specialists intersect. Identifying whom to follow requires understanding these fault lines.
Foundation Models and LLM Engineering: Technical Profiles to Watch
Mastery of language model architectures today distinguishes experts who advance the field from those who merely comment on it. In Europe, three names consistently appear in structured analyses for 2026: Arthur Mensch, Guillaume Lample, and Timothée Lacroix, co-founders of Mistral AI.
Their work focuses on LLM engineering, from pre-training to production deployment. This trio embodies a rare expertise that combines fundamental research (publications on attention mechanisms, mixture-of-experts) with industrial scaling.
Following these profiles allows for an understanding of the real technical trade-offs: choices of tokenization, quantization strategies, and the trade-off between model size and inference latency. These are topics that most mainstream lists do not address, but which determine the operational value of an AI system in business. To situate these figures in a broader landscape, a resource lists the best experts in artificial intelligence according to various selection criteria.
Open Source and AI Platforms: Influence through Infrastructure
Open source infrastructure shapes the sector more than closed research. Thomas Wolf and Clément Delangue, respectively CSO and CEO of Hugging Face, have become global references not through their academic publications, but through the platform they maintain.
Hugging Face hosts tens of thousands of models, datasets, and demonstration spaces. When an AI practitioner looks for a pre-trained model or a fine-tuning pipeline, they almost systematically go through this infrastructure. Wolf and Delangue de facto guide the development practices of the entire community.
We recommend following their positions on the governance of open models, a topic that pits advocates of open access against European regulators. Their insights complement those of the founders of Mistral AI: the former build the distribution layer, while the latter build the model layer.
Criteria for Evaluating an Open Source Expert
- Direct contributions to the code or documentation of widely adopted projects (verifiable commits, not just conference appearances)
- Clear positioning on model licenses (Apache 2.0, RAIL, copyleft) and their implications for businesses
- Ability to explain the technical limitations of the tools they promote, not just their strengths
AI for Science and Applied Machine Learning: An Underestimated Angle
Expert lists focus on mainstream generative AI. We observe that the most significant advancements often come from AI applied to the sciences. Max Welling (CuspAI) works on AI for discovering new materials. Christopher Bishop leads the AI4Science program at Microsoft.
These profiles are directly relevant to professionals deploying data systems in industry, pharmaceuticals, or energy. Their expertise in machine learning extends beyond natural language processing to cover generative models applied to molecular simulation or optimization of industrial processes.
Jürgen Schmidhuber, often cited for his historical work on recurrent neural networks (LSTM), remains a key technical voice. His contribution is not measured by followers on social media, but by academic citations and patents that still shape current architectures.
French-speaking AI Experts: Training and Professional Outreach
The French market has its own figures focused on AI skills training and corporate integration. Alexandre Kantjas (9x) offers advanced expertise on the strategic deployment of models, with an operational angle aimed at technical decision-makers.
Julien Simon produces structured educational content on machine learning and language models, accessible on YouTube. His uniqueness: he breaks down deep learning concepts without sacrificing technical rigor, making it a useful resource for data teams upskilling.
What Distinguishes an Actionable Expert from a Commentator
- Regular production of verifiable technical content (tutorials, notebooks, reproducible benchmarks)
- Documented experience in deploying AI projects in production, not just prototyping
- Ability to contextualize tools relative to business constraints (inference cost, GDPR compliance, model maintenance)
- Transparency about failures and limitations encountered during real projects
This framework allows filtering profiles that provide real expertise from those who relay product announcements without critical perspective.
AI Ethics and Regulation: Voices Not to Be Ignored
The UN launched a scientific panel dedicated to AI in 2026, a sign that regulation now structures the debate as much as technology. Experts in algorithmic ethics are no longer confined to the margins of the sector.
Following the work of this panel and the researchers contributing to it provides access to a complementary understanding of the issues: biases in training data, environmental impact of computing infrastructures, sovereignty issues over foundation models. For companies deploying AI systems at scale, these dimensions directly condition the regulatory viability of their projects.
The most relevant experts to follow in 2026 share a common trait: they produce verifiable artifacts (code, models, technical reports) rather than opinions. Building your monitoring around this distinction remains the most reliable filter for developing a solid and operational AI culture.