Why the Best AI Agents Should Be Simple: An Interview with Sierra's Zack Reneau-Wedeen

Sierra's Product Head Zack Reneau-Wedeen shared his strategy for designing AI agents to maximize customer experience. Rather than being tied to a single model, he adopts an approach of running the optimal model—Claude, Gemini, GPT, and others—in parallel depending on the nature of the task.

The key strategies are as follows:

• Model optimization: If a particular model has strengths in a particular task (e.g., speech recognition for a certain accent), that strength is leveraged, while multiple models are run simultaneously to compensate for each model's weaknesses (e.g., hallucination during silence).
• Single-agent architecture: Rather than a complex Multi-agent system that mirrors an organization's structure as-is, Sierra aims for a single agent per brand (One agent per brand) approach that fully represents the brand's voice. This is to maintain full context without any loss of information.
• Outcome-based pricing: Rather than simple usage, costs are set based on actual business outcomes such as completed sales, sharing value with customers.

Sierra is preparing for an era of Agentic Commerce that goes beyond simple chatbots to cover the entire customer lifecycle—from browsing to booking, sales, and loyalty management.