Interns Gone Wasm: WebAssembly Research & Innovation | Ep21 | WebAssembly Unleashed

Key Concepts

• WebAssembly (WASM): A portable, safe, and efficient bytecode format designed for executing code in various environments.
• WebAssembly Component Model: A system for composing WASM modules into larger applications, emphasizing modularity and security.
• AI Agents: Software entities designed to perform tasks autonomously, often requiring access to various tools and systems.
• Model Context Protocol (MCP): A standardized protocol for AI agents to interact with different tools and accomplish tasks.
• Fuzzing: A software testing technique that involves providing random, invalid, or unexpected inputs to a system to uncover bugs and vulnerabilities.
• Garbage Collection (GC): An automatic memory management technique that reclaims memory occupied by objects that are no longer in use.
• Time Travel Debugging: A debugging technique that allows developers to step through code execution both forwards and backwards in time.
• Determinism: The property of a system where the same inputs always produce the same outputs, making it easier to reproduce and debug issues.
• Wit: WebAssembly interface type system

1. Microsoft's WASI Integration for AI Agents

• Microsoft is developing a project that integrates WASM with AI agents using the Model Context Protocol (MCP).
• AI agents need secure systems to access tools and complete tasks.
• WASM components provide a secure sandbox environment for AI agents to execute tasks.
• WASI's capability-based security ensures that AI agents only have access to the resources they need, preventing unintended actions.
• The project allows agents to discover WASM components in OCI registries, download them, and request user consent before execution.
• This integration promotes the publication and use of WASM components.

2. Introduction to Khan Keramov and Arjun Ramesh

• Khan Keramov is a fourth-year PhD student at the University of Utah, researching fuzzing WASM implementations.
• Arjun Ramesh is a fifth-year PhD student at Carnegie Mellon, focusing on WASM in edge computing and embedded systems, particularly debuggability.
• Both Khan and Arjun interned with the hosts and contributed to WASM security and debugging.

3. Khan Keramov's Research on Fuzzing WASM

• Khan chose WASM because it claims to be safe, portable, and simple, challenging the notion of "silver bullets" in software engineering.
• He focuses on finding bugs in WASM implementations to ensure their security.
• Fuzzing involves providing random inputs to a system to find bugs and crashes.
• Khan is working on fuzzing the garbage collection (GC) proposal in WASM, implemented in Wasmtime.
• Fuzzing GC in WASM is different from general GC fuzzing due to the presence of types and the need to correctly translate GC from other languages to WASM.
• Khan and his advisor are collaborating with F5 on fuzzing GC.
• "Don't try it until you fuzz it" - Khan Keramov

4. Arjun Ramesh's Research on Time Travel Debugging

• Arjun focuses on debuggability in embedded systems and edge computing, where traditional debugging methods are limited.
• WASM provides a portable platform for building debug infrastructure across different computing environments.
• Time travel debugging involves recording program execution traces and allowing developers to step through the execution both forwards and backwards.
• Arjun is collaborating with F5 on this project.
• The goal is to enable developers to easily trace back and debug the root cause of bugs that cause crashes.
• WASM and the component model offer deterministic semantics at both the compute and interface layers, making time travel debugging possible.
• Initial experiments show overheads of below 10% for full recording, excluding IO.
• Time travel debugging allows developers to step back in reverse direction or tell me what was the last time this memory address was written to.

5. Future Plans

• Khan plans to work in the industry after completing his PhD, potentially returning to academia to teach WASM.
• Arjun wants to continue building virtual machines and is leaning towards working in the industry, possibly returning to academia for teaching.

6. Conclusion

• The episode highlights the importance of security and debuggability in WASM.
• Microsoft's WASI integration for AI agents demonstrates the potential of WASM in secure computing.
• Khan's work on fuzzing and Arjun's work on time travel debugging contribute to making WASM a more robust and developer-friendly platform.
• The guests express enthusiasm for the future of WASM and its potential impact on various industries.
• "Determinism is sometimes just a fancy way of saying I can crash the same way every single time."