Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn to build reliable agentic AI systems through a comprehensive examination of current limitations and innovative solutions in this conference talk. Explore the fundamental flaws of existing approaches including Static Chaining and Prompt & Pray methodologies, understanding how static systems create rigid, narrow solutions requiring intensive development and maintenance, while prompt-based approaches often yield unreliable results particularly in tool calling scenarios. Discover the accuracy, consistency, and robustness challenges plaguing today's agentic systems, especially regarding complex context retrieval and tool maturity issues. Examine AI21's Maestro framework as a paradigmatic shift toward reliable agentic AI development, featuring dynamic task planning and multi-path execution capabilities that allow users to define behavioral guardrails and constraints in natural language that automatically translate into verifiable code. Master innovative features including budget limiting, flexible model and tool selection, customizable output schemas, and comprehensive execution traceability with structured validation reporting. Understand how customer examples enable offline simulation for execution strategy improvement and cost-performance optimization of runtime planners. Gain insights into Maestro's model-agnostic architecture and its compatibility with other agentic frameworks, positioning you to implement more reliable and robust AI agent systems in your own projects.
