Presentation: From Copy-Paste to Composition: Building Agents Like Real Software
Quick Answer
Jake Mannix discusses the evolution of building LLM-based agents at Walmart, emphasizing the shift from traditional programming to using English as a programming language, while highlighting the challenges and advancements in tool design through the Model Context Protocol (MCP).
Key Points
- Walmart's agentic middleware supports over 20,000 developers and data scientists.
- facilitates interoperability and shared libraries for agent tools.
- Tool design requires careful granularity to avoid confusion for .
- MCP servers allow agents to dynamically connect and execute backend services.
- Current implementation relies on copy-pasting tool schemas into system prompts.
DeepSignal Analysis
What happened
Jake Mannix from Walmart discussed the development of LLM-based agents, highlighting the transition to using English as a programming language. He emphasized the challenges faced in tool design, particularly with the Model Context Protocol (MCP), which aims to streamline agent functionality but has limitations in implementation and flexibility.
Key evidence
- Mannix noted that Walmart has over 20,000 software developers and data scientists actively building agents, indicating a significant investment in this technology.
- He described the Model Context Protocol (MCP) as a necessary step to avoid redundant tool integrations across different agents, which would otherwise lead to increased complexity.
- Mannix criticized the current implementation of MCP, stating that it relies on copy-pasting tool descriptions directly into the agent's main program, which can lead to confusion and errors.
Why it matters
The discussion highlights a pivotal moment in software engineering where traditional programming paradigms are being challenged by new methods of agent development. The reliance on English as a programming language and the introduction of MCP reflect a significant shift in how software is created and integrated. However, the limitations of these approaches raise questions about the future of agent design and the need for more robust frameworks.
Source Excerpt
Jake Mannix explains how to mature AI agents using "virtual tools" for encapsulation, interface abstraction, and deterministic taint tracking to prevent lethal data exfiltration risks.
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from InfoQ AI, ML & Data Engineering
See more →Google Cloud Workbench Notebooks Extension Connects VS Code to Google Cloud's Jupyter Notebooks
The Google Cloud Workbench Notebooks extension for VS Code allows developers to seamlessly connect their local IDE to managed Jupyter notebook environments on Google Cloud, enhancing ML workflow efficiency. This integration eliminates context switching, enabling smooth transitions from local experimentation to high-performance cloud computing.
