25 AI engineers you should be following in 2026

Your X/LinkedIn feed is a product. What you put in is what you get out, and if most of what you’re ingesting is founder announcements and AI hype roundups, you’re essentially running your professional development on vibes. The engineers below are the people who actually shape how this technology works: they publish the research, ship the tooling, break down the internals, and post about it in ways that reward careful reading rather than a quick dopamine hit.💡This list is deliberately weighted toward practitioners over thought leaders. The thought leaders will survive the snub. They have plenty of LinkedIn comments to keep them warm.The researchers whose posts read like early-access papersThese are the people whose threads on X or posts on Substack regularly end up cited in other people’s papers, conference talks, and internal company wikis.25. Andrej Karpathy (@karpathy): Now at Anthropic working under Nick Joseph on pre-training, Karpathy joined in May 2026 after a stint building Eureka Labs. His nanochat project has driven the cost to replicate GPT-2’s CORE benchmark score down to approximately $73 on a single 8xH100 node, a 600x reduction over seven years.24. Lilian Weng (@lilianweng): Her long-form posts on agents, reasoning, and safety are effective reference documentation for the field. “LLM Powered Autonomous Agents” remains one of the most-cited posts in AI engineering, years after publication. Her writing on test-time compute is required reading if you work with reasoning models.23. François Chollet (@fchollet): Creator of Keras, co-founder of Ndea (with Mike Knoop), and architect of the ARC Challenge, which carries a $1 million prize for genuine progress on abstract reasoning. Chollet is unusually willing to point out when scaling is producing diminishing returns rather than the next breakthrough. In a field full of people telling you the ceiling is infinite, this is genuinely useful for calibration.22. Nathan Lambert (@natolambert): His Interconnects newsletter is one of the clearest sources on post-training, RLHF, and reasoning models. When a new paper drops on reinforcement learning from verifiable rewards, Lambert’s breakdown is usually the one engineers share.21. Noam Brown (@polynoamial): Led the development of o1 and OpenAI’s reasoning model line, through to the gold medal at the International Mathematical Olympiad in 2025. His threads on self-play and reinforcement learning applied to LLMs are precise, dense, and worth reading slowly.20. Oriol Vinyals (@oriolvinyalsml): VP of Research and Deep Learning Lead at Google DeepMind, Gemini co-lead, and the person behind AlphaStar, AlphaFold, and WaveNet. If you want to understand where frontier multimodal architectures are heading, this is the feed.AIAI New York, June 2026Catch up on every session from AIAI New York with sessions from Meta, Perplexity, Walmart, Pika AI, RingCentral & more.19. Soumith Chintala (@soumithchintala): The original force behind PyTorch, now CTO of Thinking Machines Lab after leaving Meta in November 2025. Still codes. Still engages directly with open-source contributors. Still the kind of person who makes you feel slightly bad about how many tabs you have open versus how much you’ve actually shipped.18. Shreya Shankar (@sh_reya): Research and evaluation specialist whose work on data pipelines and LLM judges is the closest thing the field has to a rigorous framework for knowing whether your system actually works. Her paper “Who Validates the Validators?” is required reading before you ship an LLM-as-judge setup.The builders whose posts come with actual codeThis group posts less about theory and more about what they built, what failed, and what the system looked like when they were done.17. Simon Willison (@simonw): A developer who has been cataloging LLM behavior since before most companies had an AI strategy. His 2026 PyCon lightning talk on six months of LLM developments was widely shared as a tight synthesis of what actually changed. His prompt injection work is the canonical reference on instruction/data separation, and his documentation habits are frankly embarrassing for the rest of us, in the best possible way.16. Swyx / Shawn Wang (@swyx): Co-host of the Latent Space podcast and newsletter, founder of the AI Engineer World’s Fair (6,000+ engineers in 2026). He coined the phrase “AI engineer” as a distinct role, and the field largely accepted it. His essays on what AI engineering actually involves are more useful than most job descriptions.15. Hamel Husain (@hamelhusain): His essay “Your AI Product Needs Evals” is the canonical starting point for evaluation infrastructure, and his follow-up “A Field Guide to Rapidly Improving AI Products” covers the full loop from evals to error analysis to data flywheels. Practical, opinionated, and almost entirely correct.Generative AI Summit | Los AngelesJoin Generative AI Summit Los Angeles alongside hundreds of pioneering engineers, developers & executives that are facilitating the latest tech revolution.14. Jason Liu (@jxnlco): Developer Experience Engineer on the Codex team at OpenAI. Posts on RAG architecture as a systems problem, tool routing, and what structured outputs actually enable beyond simple JSON extraction. His Instructor library has been adopted by thousands of production teams.13. Abhishek Thakur (@abhi1thakur): Four-time Kaggle Grandmaster and the person who built AutoTrain at Hugging Face. His YouTube tutorials combine implementation rigor with genuine accessibility. If you need to understand fine-tuning in practice, start here.12. Sebastian Raschka (@rasbt): His newsletter covers LLM implementation and research with a level of technical precision that rewards re-reading. Combines explanations with working PyTorch code consistently, which makes it genuinely different from most ML writing.11. Lee Robinson (@leerob): VP of Developer Experience at Cursor, formerly at Vercel. His tutorials on integrating coding agents into real development workflows are among the most cited in the AI engineering community. Relevant if your team is deciding how to work with Claude Code, Codex, or Cursor at scale.10. Mira Murati (@miramurati): Former OpenAI CTO and founder of Thinking Machines Lab (2025). One of the few people who has led teams shipping GPT-4, ChatGPT, and DALL-E, then started again from scratch. Her perspective on safety-aligned industrial roadmaps is grounded in operational reality.9. Yohei Nakajima (@yoheinakajima): Creator of BabyAGI, one of the first public demonstrations of task-driven autonomous agents. Now a VC at Untapped Capital. His posts on agent loop design remain practically useful even as the tooling around them has evolved significantly.8. Eugene Yan (@eugeneyan): Applied science at Amazon and a meticulous blogger on production ML. His survey of task-specific LLM evaluation techniques is one of the best practical inventories of what actually works per use case.7. Cassie Kozyrkov (@quaesita): Founded the field of Decision Intelligence at Google. Her writing cuts through model performance discussions to the actual question: what decision are you trying to improve, and does this metric capture it? Required following when your team is about to pick the wrong metric again. And at some point, every team does.6. Timnit Gebru (@timnitgebru): Computer scientist and leading voice on algorithmic fairness, bias, and the structural conditions that produce unreliable AI systems. Her work is less about building better models and more about interrogating what “better” means. Worth following for the questions as much as the answers.5. Kate Crawford (@katecrawford): Author of *Atlas of AI* and co-director of the AI Now Institute at NYU. Her research on compute geography, labor conditions, and data provenance is the infrastructure context that most engineering discussions leave out.The SRE toil problemStop absorbing toil. Start engineering reliability.4. Jeremy Howard (@jeremyphoward): Co-founder of fast.ai and answer.ai, with a sustained commitment to making frontier techniques accessible to practitioners who lack the resources of a major lab. His recent work on small-team frontier engineering is increasingly relevant as capable open-source models become widely available.3. Ethan Mollick (@emollick): Associate Professor at Wharton whose experiments on how LLMs affect productivity, creativity, and decision quality are the most rigorous publicly available work in that space. If your organization is trying to measure AI’s actual impact on knowledge work, his papers are the benchmark.2. Omar Khattab (@lateinteraction): Assistant Professor at MIT and creator of DSPy, the framework for programming LLM pipelines rather than prompting them. His AI Engineer World’s Fair talk on building AI systems that survive the bitter lesson is one of the better architectural arguments for moving past brittle prompt engineering.1. Dex Horthy (@dexhorthy): Founder of HumanLayer and creator of the 12-Factor Agents reference, which is one of the clearest design frameworks for production agent systems. If your team is shipping agents and hitting the usual reliability problems, his posts will feel like direct answers.A practical note on followingA list like this can turn into a consumption problem if you treat it like a shopping cart. The signal-to-noise ratio on X degrades fast once your feed fills up, so a better approach is to pick a smaller cluster and read everything they post for 30 days. Figure out whose thinking model matches yours, and whose gaps complement yours. Then expand. The goal is triangulation: when three people from different parts of this list reference the same paper or shift in the same week, that is the signal worth acting on. One person saying something is interesting. Three saying it independently means to clear your afternoon…Meet AI builders in personIf you want to take the feed offline, the Agentic AI Summit Berlin on September 15, 2026, at The Ritz Carlton is built for exactly the practitioners this list represents: engineers and technical leaders shipping agentic systems, with no expo-hall filler. 300+ attendees, focused sessions, and the kind of hallway conversations that actually move things forward.Hands-on workshops and hackathons: Build and iterate on real agent systems alongside engineers facing the same production challenges you are.Practitioner-only sessions: Speakers from Hugging Face, NVIDIA, Siemens Energy, Lovable, and GetYourGuide covering what is working in production right now, across agent architecture, MLOps, and applied AI at scale.A network worth keeping: The summit draws senior technical leaders from across European AI, the kind of room where the conversations continue well past the closing keynote.Early bird passes save €100. Secure your seat here