Understanding artificial intelligence takes more than following the CEOs of its largest companies. Researchers develop new capabilities, engineers turn them into useful systems, educators make them accessible, and independent investigators examine their limitations and consequences.
This list brings those perspectives together. Each person is included for a specific contribution: what they build, what they study, or what they help us understand. Inclusion does not imply agreement with their predictions or endorsement of their companies.
The list is grouped by contribution and alphabetized by surname within each section. It is a reading map, not a ranking.
1. Frontier laboratories and research leadership
Sam Altman — OpenAI
As OpenAI’s co-founder and CEO, Altman helps shape how advanced AI is financed, deployed and negotiated with governments. His importance extends beyond individual models to infrastructure, distribution and institutional influence. Follow him to understand the commercial and political direction of frontier AI, while reading his forecasts in the context of OpenAI’s interests. Source
Daniela Amodei — Anthropic
Anthropic’s co-founder and president brings an essential operational perspective to a field often described through research breakthroughs alone. Her work concerns building the organization that delivers and supports AI at scale. She belongs here because hiring, organizational incentives and commercial execution help determine whether a laboratory’s stated principles survive its growth. Source
Dario Amodei — Anthropic
Anthropic’s co-founder and CEO combines a research background with leadership of the company behind Claude. He is a prominent voice on increasing model capabilities, economic transformation and the risks of advanced AI. His contribution is the connection between technical development and strategic choices about deployment, safeguards and governance. Source
Demis Hassabis — Google DeepMind, Alphabet and Isomorphic Labs
Hassabis connects frontier AI with scientific discovery. Google’s updated leadership announcement places him as Chair of Google DeepMind and Chief Scientist of Alphabet, while continuing to lead Isomorphic Labs. He is particularly relevant to the question of how increasingly general AI can contribute to science and drug discovery. Source
Koray Kavukcuoglu — Google DeepMind
Kavukcuoglu deserves a place alongside the more publicly recognizable Hassabis. As Senior Vice President of Google DeepMind, he oversees Gemini model development, frontier research, and the Gemini application and developer teams. His relevance lies in the organizational and technical work that connects research advances with widely used AI systems. Source
Arthur Mensch — Mistral AI
Mistral’s co-founder and CEO represents an important European approach to advanced AI: competitive models, open-weight releases and greater control over deployment. He contributes to discussions about technological sovereignty, enterprise adoption and the economics of maintaining alternatives to American and Chinese providers. His perspective is especially useful when discussing who controls the AI supply chain. Source
Mira Murati — Thinking Machines Lab
Murati founded Thinking Machines Lab after serving as OpenAI’s CTO. The company focuses on making capable AI more adaptable and useful in collaboration with people. She is worth following for the intersection of research, product design and customization: how general-purpose models become systems that fit particular users and workflows. Sources, Thinking Machines Lab
Ilya Sutskever — Safe Superintelligence
An OpenAI co-founder and former chief scientist, Sutskever now pursues advanced AI through Safe Superintelligence. His relevance comes from both his contribution to modern deep learning and his laboratory’s explicit focus on developing safe superintelligence. His comparatively limited public output makes substantive interviews and research disclosures especially worth examining. Source
2. China’s model builders
Chinese laboratories belong in the main discussion of AI. Their work influences model availability, deployment costs and the options available to developers worldwide. “Open-weight” is the more precise description when model weights are available but the complete training process is not.
Liang Wenfeng — DeepSeek
DeepSeek’s founder is essential to understanding the competitive pressure Chinese laboratories place on established AI providers. His company’s research and model releases make efficiency and access central questions. Follow his work to examine how much progress depends on resources, how much on research choices, and how openly released models change competition. Sources, DeepSeek research
Junyang Lin — Pragmatik Labs; formerly Qwen
Also known as Justin Lin or Lin Junyang, he helped build Alibaba’s Qwen model family before leaving its leadership in March 2026. He subsequently announced Pragmatik Labs, focused on agents spanning digital and physical environments. He belongs here for his experience developing widely used models and his transition toward systems that act. Source
Jie Tang — Tsinghua University and Zhipu AI / Z.ai
Tang combines academic research with the development of the GLM model ecosystem. His work spans knowledge representation, pretrained models and increasingly capable agents. He adds a useful perspective on the relationship between university research and commercial laboratories, and on model development beyond the best-known American companies. Sources, Zhipu’s research direction
Yang Zhilin — Moonshot AI / Kimi
Yang, also listed in Western name order as Zhilin Yang, founded the company behind Kimi. His work connects language-model research with reasoning, multimodal capabilities and agentic applications. He is the person to include in a personalities list; @Kimi_Moonshot is an organizational account rather than a personal identity. Source
3. Foundations and alternative research directions
Yoshua Bengio — LawZero
Bengio helped establish modern deep learning and now serves as Co-President and Scientific Director of LawZero. His current work explores AI designed to be safe from the outset, including approaches that separate understanding from autonomous goal pursuit. He contributes both foundational expertise and a concrete research agenda for reducing loss-of-control risks. Sources, Research direction
Geoffrey Hinton — Deep learning and AI risk
Hinton’s work helped establish the neural-network methods underlying modern AI. His public contribution now includes sustained discussion of how more capable systems might affect employment, power and human control. He belongs in the list for that combination of foundational achievement and engagement with the consequences of the technology he helped develop. Source
Yann LeCun — AMI Labs and New York University
LeCun is a deep-learning pioneer and Executive Chairman of AMI Labs. His research interests include machine perception, robotics and architectures that learn representations of the world. He is particularly valuable for understanding alternatives to the assumption that scaling today’s language models is sufficient for broadly capable intelligence. Source
Fei-Fei Li — World Labs and spatial intelligence
Li’s work connects computer vision with AI that can perceive and reason about physical and virtual environments. Through World Labs, she is developing spatial intelligence and world models. She belongs here because understanding space, objects and interactions is a major research challenge that cannot be adequately represented by a list focused only on conversational AI. Source
Richard Sutton — Reinforcement learning and continually learning agents
Sutton is a foundational figure in reinforcement learning, with research affiliations including the University of Alberta, Amii and Keen Technologies. His work examines how agents learn through interaction and experience. He contributes a long-term perspective on learning, prediction and action that is directly relevant to today’s interest in autonomous agents. Source
4. Evaluation, interpretability and the limits of current systems
Beth Barnes — METR
Barnes founded and leads METR, which evaluates the capabilities and risks of advanced AI systems. Her work addresses a question that product demonstrations often leave unresolved: what can a system accomplish reliably under meaningful test conditions? She is an important source for distinguishing impressive examples from evidence of dependable autonomous performance. Source
François Chollet — ARC Prize and Ndea
The creator of Keras also developed ARC, a benchmark family concerned with generalization and adaptation to unfamiliar problems. Through ARC Prize and Ndea, Chollet investigates both how intelligence should be measured and how different architectures might achieve it. He is valuable for challenging the equation of benchmark success with general intelligence. Sources, Ndea
Arvind Narayanan — Princeton University and AI Snake Oil
Narayanan studies the societal impact of digital technologies and co-authored AI Snake Oil with Sayash Kapoor. His contribution is methodological clarity: separating different kinds of AI, scrutinizing evidence and examining whether systems work in their intended settings. He is especially useful when a technical result becomes an expansive commercial or political claim. Source
Chris Olah — Anthropic and mechanistic interpretability
Olah co-founded Anthropic and works on understanding the internal mechanisms of neural networks. His research asks how learned computations can be translated into explanations people can inspect. He belongs here because understanding what models do internally could improve diagnosis, auditing and safety beyond simply observing their outputs. Sources, Interpretability research
5. Engineering, agents and technical education
Boris Cherny — Claude Code
Cherny created Claude Code at Anthropic. His work concerns the practical design of AI systems that operate inside real software-development workflows. He contributes firsthand experience with agent interfaces, developer habits and the evolving division of work between people and models. He is particularly relevant to anyone building or evaluating coding agents. Source
Chip Huyen — AI engineering and production systems
Huyen is a computer scientist and author focused on bringing AI into production. Her work connects model capabilities with evaluation, data, infrastructure and operational constraints. She belongs on the list because building a useful AI application involves many decisions that are invisible in model announcements but decisive in actual use. Source
Andrej Karpathy — AI research and education
Karpathy joined Anthropic’s pretraining team in May 2026, adding another chapter to a career spanning OpenAI, Tesla and education. His technical explanations remain unusually useful for understanding neural networks and language models from the inside. He contributes both research experience and an ability to make difficult implementation details accessible. Sources, Educational work
Andrew Ng — DeepLearning.AI, AI Fund and applied AI
Ng’s relevance extends well beyond co-founding Coursera. Through education, investment and applied-AI ventures, he works on expanding the number of people and organizations able to use AI effectively. He contributes a practical perspective on skills, implementation and identifying useful applications, alongside a long history in machine-learning research. Source
Sebastian Raschka — LLM research engineering and education
Raschka explains language-model architectures and training methods through papers, code and small-scale implementations. His current work also covers inference and evaluation. He is worth following because he helps readers move from recognizing technical terminology to understanding what a method actually does and how to reproduce its essential behavior. Source
Thibault “Tibo” Sottiaux — OpenAI products and platform
Sottiaux’s remit has expanded beyond Codex to OpenAI’s broader product and platform organization. His work connects frontier capabilities with applications, developer tools and infrastructure. He belongs here for insight into the engineering and product decisions required to turn increasingly capable models into systems people can use consistently. Source
Peter Steinberger — OpenClaw and OpenAI
Steinberger created OpenClaw and subsequently joined OpenAI’s Codex team. His background spans developer tools, open-source software and AI-powered applications. He adds a hands-on perspective on agents that interact with tools and complete tasks, including the product-design questions that emerge when software becomes more proactive. Source
Lilian Weng — Research synthesis and AI systems
Weng’s Lil’Log provides substantial explanations of topics including agents, reasoning, training and model safety. Following her time at Thinking Machines Lab, her return to OpenAI was reported in 2026. Her lasting contribution to this list is the ability to organize a fragmented research literature into explanations that technically engaged readers can use. Sources, Role update
Simon Willison — Open-source tools and practical AI analysis
Willison builds open-source software and documents his experiments with language models and development tools. His value is the connection between working examples, technical inspection and clearly stated limitations. He is especially useful for readers who want to understand new capabilities through reproducible practice rather than relying entirely on vendor presentations. Source
6. Open models and shared research infrastructure
Clément Delangue — Hugging Face
Hugging Face’s co-founder and CEO works on infrastructure through which developers share and use models, datasets and tools. He belongs here because access and distribution shape who can participate in AI development. His perspective is useful for discussions about openness, independent development and the relationship between research communities and commercial platforms. Sources, Public statement
Nathan Lambert — Trillium Labs and Interconnects
Lambert co-founded Trillium Labs in October 2026 after previously working on post-training at Ai2. The nonprofit aims to publish training recipes, data, evaluations and intermediate checkpoints. Alongside his Interconnects writing, this makes him important to the discussion of whether researchers can reproduce and investigate the methods behind increasingly closed commercial systems. Source
7. Society, language, work and environmental costs
Emily M. Bender — Computational linguistics and accountability
Bender is a University of Washington professor whose work includes language technology and its societal effects. She contributes linguistic expertise to debates about meaning, communication and what model outputs actually demonstrate. She is particularly valuable when human-like language encourages claims about understanding, intention or intelligence that require closer examination. Source
Timnit Gebru — Distributed AI Research Institute
Gebru founded DAIR to pursue independent, community-rooted research into AI and its consequences. Her work examines power, discrimination and the people affected by technological systems, including those whose labor supports them. She belongs here because evaluating AI requires examining who benefits, who bears its costs and whose priorities guide development. Sources, Profile
Sasha Luccioni — Sustainable AI Group
Luccioni studies the environmental impacts of AI and now identifies her role as Co-Founder and Chief Scientific Officer of Sustainable AI Group. Her work focuses on measuring and comparing those impacts so organizations can act on them. She contributes evidence to a discussion too often reduced to unsupported claims about AI being either inherently sustainable or uniformly wasteful. Source
Ethan Mollick — Wharton and One Useful Thing
Mollick studies and explains how AI changes work, education and human collaboration. His writing connects research with practical experiments in using the technology. He is worth following for the question that comes after a capability announcement: how does this change what people can do, and under which conditions does it actually help? Sources, One Useful Thing
8. Infrastructure, capital and strategic influence
Leopold Aschenbrenner — Situational Awareness
Aschenbrenner is an investor and the author of Situational Awareness, an influential argument about rapid AI progress and its industrial and geopolitical consequences. He belongs in a category about scenarios and capital allocation. His contribution is a strongly articulated thesis worth examining; its predictions should be assessed as predictions rather than treated as a scientific consensus. Sources, Investment activity
Jensen Huang — NVIDIA
NVIDIA’s founder and CEO shapes the computing platforms on which much of AI development and deployment depends. His importance includes hardware, networking, software and the economics of large-scale infrastructure. Follow him to understand the material constraints behind model progress, while recognizing that his account of demand also represents a major supplier’s commercial perspective. Sources, NVIDIA’s platform direction
Elon Musk — xAI, Grok and the SpaceX AI business
Musk belongs in a broad AI list because of his role in funding, infrastructure, distribution and public debate. Following SpaceX’s acquisition of xAI, those activities sit within a larger industrial organization. His relevance is primarily strategic: the deployment of resources and products, and the influence of ownership on how AI reaches users. Sources, Compute infrastructure
Mark Zuckerberg — Meta
Meta’s founder and CEO influences how AI reaches consumers and businesses through an exceptionally large distribution network. His decisions affect model strategy, infrastructure investment, agents and enterprise products. He belongs here because understanding AI’s future requires understanding the platforms through which it becomes part of everyday communication and economic activity. Sources, AI strategy
The selection remains deliberately incomplete. Its purpose is to provide a range of people whose work can be examined: through papers, code, evaluations, products and arguments—not simply through their visibility.
No comments yet