Audrey Woods, MIT CSAIL Alliances | August 31, 2026
MIT CSAIL is where big ideas get started, and they don’t just stay in the lab. CSAIL innovation becomes products, companies, and even entire markets. CSAIL Alliances is how industry gets a front-row seat.
We're launching Startups at CSAIL, a periodic look at the newest ventures spinning out of our research labs. This is just a sampler; for the full menu, join CSAIL Alliances, subscribe to our newsletter, or contact your Client Relations Coordinator.
In this edition: robots, AI agents, secure AI, and AI on the stage!
Walden Robotics
For those who missed the announcement, this July, MIT CSAIL Professor Russ Tedrake launched a full-stack Physical AI startup centered on building and deploying general-purpose robots in real work environments. Collaborating with Toyota, Walden Robotics has already been deploying robots in production environments like Toyota factories since February 2026. The company’s product builds on research done in the Toyota Research Institute, including Diffusion Policy and Large Behavior Models (LBMs) which enable machines to quickly learn new tasks and improve through on-the-ground practice. Tedrake, who serves as the company’s CEO, says, “Core advances in Physical AI, and all of the excitement and attention surrounding it, has made disruptive change possible.”
Eka Robotics
Also in the robotics space, Eka Robotics spun out of the lab of MIT CSAIL Associate Professor Pulkit Agrawal, who co-founded the company with Tuomas Haarnoja, previously at Google DeepMind. The startup develops advanced robotic dexterity using a pioneering Vision-Force-Action (VFA) model, bypassing traditional data bottlenecks by training robots via simulation to process physical forces. Their model unites generality, performance, and safety, allowing robots to generalize across objects, tasks, and environments. Wired says, “Eka’s robotic claw feels like we're approaching a ChatGPT moment for the physical world.”
Gekit
Gekit builds tools for Agentic Data Access. By improving the quality of the enterprise data fed to agents, Gekit can reduce latency and cost, and increasingly improve answer quality. The Gekit system focuses exclusively on data management services, and can thereby take advantage of database-wide information that is not available to conventional standalone agents. Gekit works across data providers, can be adopted in minutes by either humans or coding agents, and can operate with as little or as much human supervision as the customer prefers.
Refractal
Announced soon after OpenAI’s model escaped its sandbox and hacked into Hugging Face, CSAIL Alliances Startup Connect Member Refractal was created "to secure AI where it matters most." The company offers a software platform for AI security, with runtime monitoring, policy enforcement, and continual learning to protect and capture how AI systems behave in practice. Their system can flag possible vectors of attack as well as highlight what agents are able to do, which tools and skills they use, what actions they take, what data they access, and what real-world side effects they create. Co-founder and CEO Strahinja Janjusevic, who graduated from MIT in 2026, says, “the MIT commencement speaker told graduates to ‘run toward the hardest problems.’ I am doing exactly that.”
Autolab.ai
Co-founded by CSAIL PhD student Artem Lukoianov, Autolab.ai helps companies train models on their proprietary data using custom-built evaluations. Autolab developed an agentic platform that helps applied AI/ML teams automate the manual workflow of training, evaluating, and optimizing their models. Lukoianov says, “Our early pilots have helped customers achieve significant model quality improvements and reduce their development time by as much as 60%, without requiring additional compute or headcount.” Potential use cases for Autolab’s system are: task-specific model training, self-improving agents, classical ML model development, and hardware-specific model optimization.
Stage Intelligence
Shifting to the entertainment world, Stage Intelligence is building AI tools for live event venues. As the company explains, “Every show starts with the same problem: getting complete, accurate information from clients in time.” Designed for venue managers or technical directors, their platform offers an AI-powered interface that sits between a venue's clients and its production office, asking the expert follow-up questions a seasoned production manager would ask, reconciling the answers against what a venue actually owns and allows, and producing one clean advance. Z. Chen—who recently finished their master’s at MIT in music technology and studied in the new Human-AI Resonance Lab in CSAIL under Associate Professor Anna Huang—hopes their technology allows “venue teams to spend less time on paperwork and more time on the show.”
CSAIL’s a busy place and there’s always more in the pipeline. Subscribe to the CSAIL Alliances newsletter so you never miss an edition of Startups at CSAIL, and follow us on LinkedIn for real-time updates as these companies (and the next ones) take shape.
