The Agentic Web holds transformative promise for the democratization of AI, serve as antidote to AGI and unlock trillions of dollars of economic benefits. But it faces threats of fragmentation and centralization as the Internet of AI Agents evolves. Universal interoperability, permissionless innovation, and user sovereignty over data and agents will require transparent protocols and rapid advances to reconcile the needs of a diverse field.
Networked AI Agents in Decentralized Architecture (NANDA) at MIT offers a three-phase roadmap for this emerging landscape. Phase 1 establishes foundation elements—secure agent identity, discovery indices, and interoperability—via open protocols designed to support trust and accountable governance. Phase 2 introduces economic structures such as knowledge pricing, decentralized marketplaces, and reputation-based transactions, enabling agents to coordinate and exchange value at scale. Phase 3 aims for the emergence of agent societies, fostering large-scale co-learning, adaptive population models, and collaborative networks to tackle complex real-world tasks.
This framework is informed by academic work from the Raskar Lab, including algorithmic advances in Automated Machine Learning (AutoML) tailored for distributed health data, split learning for privacy-preserving model training, formal methods for dynamic knowledge valuation, and techniques for co-learning and collaborator selection in decentralized settings. Ensuring the agentic web remains open, safe, and transparent, NANDA’s development builds upon open standards, participatory governance, and research-driven safeguards.
Growing satellite constellations in Low Earth Orbit are taking advantage of the lower cost of launch and commercial electronics and components. They leverage intersatellite connectivity and increased onboard compute capability to improve communications and Earth observations. We discuss overcoming the challenges of the space environment and enabling technologies for the future, such as laser communications, dynamic tasking algorithms, direct to cellular, and in-space robotic assembly.
Methane is a potent greenhouse gas emitted from a wide range of human activities including oil and gas production, coal mining, waste management, and agriculture. Satellites have unique capabilities to quantify and attribute methane emissions worldwide. In this talk, I will discuss recent advances in satellite remote sensing of methane emissions, including targeted observation of individual point sources, global mapping of large emitters with land-imaging satellites, real-time tracking of extreme releases from geostationary orbit, and continuous monitoring of total regional emissions from oil and gas fields.
The new space economy is currently experiencing a rapid expansion, with a compound annual growth rate estimated between 7% and 11%. This significant growth encompasses an increasing number of launches, with projections indicating daily launches to space by 2027, as well as a substantial rise in the number of operational satellites. This presentation will provide an overview of the new space economy and elaborate on its co-existence with the traditional government-driven space enterprise. One of the direct consequences of this growth is an increase in the resident space object (RSO) population, underscoring the critical need for enhanced and improved Space Situational Awareness. We will demonstrate how the integration of ground-based radar and optical observations with on-orbit optical sensing can lead to more effective decision-making for collision avoidance maneuvers and other crucial operational considerations.
In this dynamic talk, Prof. Dennis Whyte, MIT Plasma Science and Fusion Center, presents a compelling vision for fusion energy as the transformative solution to global energy and climate challenges. He explains how fusion—mimicking the power of stars—offers a carbon-free, virtually limitless, and safe energy source that can scale globally. Prof. Whyte highlights recent breakthroughs at MIT, including the development of high-temperature superconducting magnets that drastically reduce the size and cost of fusion reactors. These innovations have led to the creation of SPARC, a compact fusion experiment, and the spinout of Commonwealth Fusion Systems, aimed at commercializing fusion by the early 2030s. Emphasizing fusion's potential to decarbonize not just electricity, but also heavy industry and fuel production, Whyte outlines a clear, science-driven pathway to realizing practical, scalable fusion power within this decade.
The next wave of innovation is being shaped by AI systems that don’t just respond; they act. From agentic AI that collaborates and makes decisions autonomously to decentralized architectures that push intelligence to the edge, MIT researchers are leading the charge. They are reimagining how organizations secure, interpret, and operationalize data. This track brings together thought leaders from across MIT to explore the strategic, organizational, and human implications of AI at scale. Topics will include quantum-safe infrastructure, explainable AI, cyber-physical resilience, agent-based platforms, and the role of trust, transparency, and ethics in intelligent systems. For enterprises navigating an era defined by autonomy, agility, and risk, this track connects frontier research with real-world impact.
As AI becomes increasingly autonomous and decentralized, global enterprises must rethink how intelligence is built, governed, and scaled. This closing discussion brings together corporate, academic, research, and entrepreneurial perspectives to explore practical strategies for integrating agentic and distributed AI systems across complex organizations. Attendees will gain insight into how leading thinkers are addressing questions of trust, transparency, and control, balancing innovation with accountability to create resilient, adaptive, and competitive enterprises in a future defined by intelligent autonomy.
Generative programming is a paradigm that takes decades of theoretical research and practical experience in algorithms and software engineering and applies it to the way we interact with LLMs. Instead of developing prompts by trial and error, which usually results in long and complex prompts, generative programs combine known and well-tested control flows and design patterns, such as divide and conquer, with short, consumable prompts. I will describe Mellea, an open-source library for writing generative programs, replacing brittle prompts with structured, maintainable, robust, and efficient AI workflows. I will then discuss some challenges and proposed solutions for efficiently handling multiple models and LLMs’ internal memory management within a generative program.
Life sciences are no longer confined to the realm of biology—they have evolved into a multidisciplinary frontier. This session examines the dynamic intersection of biology, engineering, and computational science, where bold ideas give rise to transformative innovation. By integrating AI, advanced technologies, and foundational biological research, the session will highlight how cross-disciplinary collaboration accelerates the path from scientific discovery to real-world application at MIT. Emphasizing the translation of visionary research into impactful solutions, this track invites participants to reimagine what becomes possible when disciplines converge to shape the future.
Antibody–drug conjugates (ADCs) are the gold standard for targeted drug delivery systems, but their chemical design imposes constraints that, if addressed, could enable a new generation of cancer therapeutics and imaging modalities. For example, due to bioconjugation limitations, the payload scope of ADCs is restricted to highly potent payloads with inherently unselective mechanisms of action, leading to narrow therapeutic windows and resistance. This seminar will introduce a new platform called Antibody–Bottlebrush prodrug Conjugates (ABCs) that can potentially address these challenges. ABCs feature a modular design that allows drug-to-antibody ratios (DARs) from ~1–135 while maintaining strong target binding, efficient cellular uptake, and favorable pharmacokinetics and biodistribution. Leveraging their capability to access very high DARs, ABCs can carry payloads (e.g., 10-fold less potent than existing ADC payloads) that are insufficiently potent to be used in traditional ADCs, thereby enabling new mechanisms-of-action. Moreover, ABCs are readily amenable to using various payload combinations, release mechanisms, and non-drug (e.g., imaging) agents. ABCs display efficacies on par with or superior to clinical ADCs in preclinical tumor models at clinically relevant payload doses, motivating their further clinical translation.