Stephan Zheng
San Francisco, California, United States
1K followers
500+ connections
About
Building AI agents that help us co-invent the world of tomorrow.
Before this…
Activity
1K followers
-
Stephan Zheng shared thisSometimes, vibe coding isn't enoughStephan Zheng shared thisBuilding software for space is one of the hardest engineering problems there is. As AI becomes more capable of writing and reasoning about complex code, we wanted to explore the key question: can AI help us invent at the frontier? Here's what we found →
-
Stephan Zheng shared thisSearch and you will find (with AI agents)!MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)
MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)
Stephan Zheng shared thisAI agents can be very effective when they use LLMs, but coding agents to work backwards to fix mistakes is time-consuming. MIT method executes AI agent programs by backtracking & making multiple attempts, helping coders work w/these systems efficiently: https://bit.ly/45QNta0 -
Stephan Zheng shared thisExcited to share a big milestone from the Asari AI team: we’ve built agents that migrated multiple critical C libraries to Safe Rust — thoroughly tested, no shortcuts, no slop. One example: libmcs, software that literally runs in space. Our agents met and exceeded ESA (European Space Agency - ESA) testing standards, verified in partnership with GTD, a space-industry testing provider. This requires thousands of reasoning steps about systems and efficiently collaborating with humans on key design decisions. The Safe Rust code is open-source (links below) — check yourself and give us feedback! Blog: https://lnkd.in/gNpbTJh2 + more deep dives soon. -- Why did we challenge our agents on this? Making software infrastructure safe is a national security imperative. Our agents worked on libraries that are part of Defense Advanced Research Projects Agency (DARPA) Tractor program: - Real production C → Safe Rust - Compression: gzip, zlib - Parsing: libyaml - Space-grade math: libmcs - No unsafe escapes - Thorough testing and consistency checks -- Why does this work? Our agents are rigorous but also treat system design as a craft. That involves: - Reasoning about architecture and requirements - Efficiently surfacing key design decisions for human review - Coherently do full redesigns or more faithful translations - Writing and passing comprehensive tests - Not sacrificing performance - Not shipping slop 👉 We’re open-sourcing the Safe Rust code so anyone can audit, verify, and contribute: https://lnkd.in/gai6RTuC https://lnkd.in/gyxCQRH3 https://lnkd.in/gkr8TZuS https://lnkd.in/gFCagiBB Please let us know what you think — GitHub Issues and comments are very welcome! I’m super proud of the team that pushed our agents to this level and are moving towards engineering superintelligence. This project offers a glimpse of how mission-critical systems will be built and invented together with AI. We’re continuously scaling and generalizing our agents & testing with partners. If you care about building AI that helps invent the systems of the future, come build with us — we’re hiring: https://lnkd.in/gJ4CjNeM
-
Stephan Zheng reposted thisStephan Zheng reposted thisMigrating critical infrastructure from C to memory-safe Rust at scale is hard. We challenged our AI agents on four production libraries consisting of 40,000+ lines of code that power essential systems worldwide: • libmcs: high-precision math library that meets strict European Space Agency (ESA) standards • gzip: one of the most widely-used command-line tools for file compression • libyaml: the foundational parsing library used for YAML configuration files • zlib: used by virtually every operating system and application for data compression ➡️ Result: All four successfully translated to Safe Rust with comprehensive test validation. Our agents maintained correctness across thousands of interdependent decisions, demonstrating they can handle the rigorous system design and verification required for production engineering. We've open-sourced all four Rust translations. Development combined human-AI co-design for specifications with fully autonomous implementation by our agents. Read the full blog post here: https://lnkd.in/gWyMhxuH Open-source repos: libmcs: https://lnkd.in/gQuR25sP gzip: https://lnkd.in/ge7ArKiq libyaml: https://lnkd.in/gDhBEc72 zlib: https://lnkd.in/gygpvw4k All feedback welcome!
-
Stephan Zheng reposted thisStephan Zheng reposted thisWe're excited to share our first blog post, introducing EnCompass: a framework that fundamentally rethinks how AI agents search and reason through complex, multi-step tasks. Presented at NeurIPS 2025. The Challenge: As AI agents tackle complex workflows like code translation or system debugging, possible paths multiply exponentially and small errors compound, yet traditional approaches hard-code search strategies directly into workflows, making them brittle and difficult to improve. Our Approach: EnCompass separates what agents do from how they explore solutions. Developers annotate workflows with branchpoints, then apply various search strategies such as beam search, best-of-N, and tree search without rewriting lines and lines of code. The Results: Across our three case studies (code translation, hypothesis generation, iterative code refinement): • 3-6× fewer lines of code to implement search strategies • Cleaner, more maintainable agent workflow code • Practical implementation of techniques to unlock better scaling laws Read the full blog post: https://lnkd.in/gu8cH72y Paper: https://lnkd.in/g-bCXCtz
-
Stephan Zheng shared thisWhether you were affected by the Meta layoffs or just ready for your next big challenge—we’re hiring. Come build AI infrastructure and supercompute that help people invent the systems of tomorrow. I’m always happy to chat about what you could be building with us — hello@asari.ai / https://www.asari.ai.
-
Stephan Zheng posted thisCome chat about Asari AI agents and the future of AI at #NeurIPS2024 with Yisong Yue and me -- DMs are open :)
-
Stephan Zheng shared thisHow are AI agents going to help us model the world and reason about complexity? Looking forward to giving a talk about this at Caltech 👇Stephan Zheng shared thisAre you at Caltech and interested in modeling and optimizing how AI agents learn and make decisions in the economy, science, and other complex systems? Come to our talk and Q&A at Caltech next Monday, 12 pm PT! You'll hear how AI agents can be used to simulate our world and about recent advances in reasoning with AI 🌟 Sign up here: https://lnkd.in/gtpUWsV4 We're also hosting coffee chats: https://lnkd.in/gPSgtdem
-
Stephan Zheng shared thisOur angel JP Millon is on LinkedIn!Stephan Zheng shared thisSmall correction: our angel investor -- JP Millon -- *does* have a LinkedIn profile!
-
Stephan Zheng liked thisStephan Zheng liked thisBuilding software for space is one of the hardest engineering problems there is. As AI becomes more capable of writing and reasoning about complex code, we wanted to explore the key question: can AI help us invent at the frontier? Here's what we found →
-
Stephan Zheng liked thisStephan Zheng liked this"Oliver Binkhorst klopte opnieuw bij Quote aan met de cijfers van DutchTechSF, de community voor Nederlandse ondernemers en investeerders in Silicon Valley." Geschreven door Vita Van Lennep voor Quote DutchTechX — spun out of DutchTechSF, the Dutch-American entrepreneurial community in the San Francisco Bay Area — evangelizes Dutch entrepreneurial successes in the US, organizes inclusive social events in AMS and SF, and runs Trek—the flagship program for Netherlands-based founders and entrepreneurial talent. The next edition will take place in May. Message me for details or follow the breadcrumbs. #WeAreDutchTech #WeAreGiant Lees hier het volledige verhaal: https://lnkd.in/gSU7h3yw
-
Stephan Zheng liked thisStephan Zheng liked thisGreat piece for those not in the SF bubble and who want to know what the Claude Code hype is all about - https://lnkd.in/g8pzRwEe TLDR; it's quick dopamine for workaholics :)
Experience
Education
-
-
-
-
-
-
-
-
-
-
-
-
-
Publications
-
Screening of heterogeneous surfaces: charge renormalization of Janus particles
Journal of Physics: Condensed Matter
Nonlinear ionic screening theory for heterogeneously charged spheres is developed in terms of a mode-decomposition of the surface charge. A far-field analysis of the resulting electrostatic potential leads to a natural generalization of charge renormalization from purely monopolar to dipolar, quadropolar, etc., including mode-couplings. Our novel scheme is generally applicable to large classes of surface heterogeneities, and is explicitly applied here to Janus spheres with differently charged…
Nonlinear ionic screening theory for heterogeneously charged spheres is developed in terms of a mode-decomposition of the surface charge. A far-field analysis of the resulting electrostatic potential leads to a natural generalization of charge renormalization from purely monopolar to dipolar, quadropolar, etc., including mode-couplings. Our novel scheme is generally applicable to large classes of surface heterogeneities, and is explicitly applied here to Janus spheres with differently charged upper and lower hemispheres, revealing strong renormalization effects for all multipoles
Other authors -
Honors & Awards
-
Lorentz Prize for Theoretical Physics
Royal Holland Society of Sciences and Humanities
The Lorentz Prize was awarded for the work in my master's thesis in Theoretical Physics, in which I elaborated on a new mathematical technique that connects knot theory (pure math) and string theory (physics) in a new way.
-
Huygens Scholarship Programme
Dutch government
Twice (2009, 2011) awarded the Huygens scholarship for international study, for excellence in physics and mathematics.
Other similar profiles
Explore more posts
-
Christophe Jurczak
Although the title might seem cryptic at best 😉 Pasqal's quantum simulation team has just published a paper on arXiv that I like a lot: https://lnkd.in/eU6BAhv3 In this work, theorists remotely programmed Pasqal’s neutral atom QPU to realize and probe a complex many-body system with excellent precision. Through this direct manipulation, they uncovered new emerging symmetries distinct from previously known theoretical models. This, and other developments of similar nature with superconducting or ion-based QPUs, represent a fundamental shift in the way experimental and theoretical physicists can interact: Quantum computing platforms are now becoming programmable laboratories accessible to scientists worldwide, enabling real-time iterations between theory and experiment. Looking forward to exciting science and, ultimately, applications of quantum materials ! #QuantumComputing #CondensedMatterPhysics #QuantumMaterials #QuantumSimulation #Pasqal Alexandre Dauphin Loïc Henriet Joseph Vovrosh Julius de Hond Sergi Julià-Farré, Ph. D.
3 Comments -
Omanshu Thapliyal
As a control systems researcher, I'm fascinated by the potential of Transformers beyond traditional NLP. This blog post contains my notes as a controls researcher finding entry points into transformers, and how sequence modeling is now finding applications beyond NLP into controls problems! https://lnkd.in/gkE4Mk-j
-
Aniekan Afangideh
So, there is a new paper by Sabine Hossenfelder that attempts to explain the collapse of the wavefunction experienced in quantum mechanics as an offshoot of a "yet-to-be-discovered" fundamental quantum theory where particles and geometries are the same. The work offers insight into why quantum states appear to collapse into eigenstates upon wavefunction collapse. It suggests that local collapse naturally arises in a superdeterministic framework derived from general assumptions about quantum gravity. I believe this paves a way for more research on the field of hidden variable theories and quantum information theory. Pre-print: https://lnkd.in/dDDS9Np5
13 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Stephan Zheng
2 others named Stephan Zheng are on LinkedIn
See others named Stephan Zheng