Building AI Agents on the Frontend with Sam Bhagwat and Abhi Aiyer

What It Covers

Sam Bhagwat and Abhi Aiyer discuss Maastra, an open-source TypeScript framework for building AI agents with primitives like workflows, tools, and RAG, addressing the gap in frontend-focused AI development tooling.

Key Questions Answered

• •TypeScript-first architecture: Maastra targets JavaScript developers by providing AI agent primitives in TypeScript rather than Python, eliminating language switching overhead and enabling full-stack development with consistent tooling across frontend and backend components.
• •Agent complexity spectrum: Agentic behavior exists on a spectrum from single LLM calls with tools to multi-agent networks. Agents struggle with more than 8-10 tools simultaneously, requiring workflow orchestration to group tools into manageable categories for reliable execution.
• •Workflow determinism pattern: Engineers often know the exact steps needed for complex tasks. Maastra workflows let developers write deterministic code paths while using LLMs only at specific execution points, avoiding unreliable model-driven orchestration for predictable processes.
• •MCP integration strategy: Model Context Protocol servers function as decentralized integration hubs, allowing Maastra agents to access third-party tools written in any language. This eliminates manual integration work while expanding agent capabilities through community-built MCP servers.