From: aidotengineer
Sierra is a conversational AI platform for businesses [00:00:29]. While historically associated with chat experiences and customer service, Sierra is expanding its reach [00:00:47]. By the end of the current year, most of their interactions are expected to occur over the phone [00:00:53]. The platform is also being used for a wider range of customer experience aspects, including sales, subscription management, and product recommendations [00:01:02].
Evolution of AI and Customer Experience
The development of AI has seen rapid advancements, with significant breakthroughs occurring in the last decade [00:01:48]. Early AI work, like the first version of Google Lens in 2016, focused on tasks such as differentiating between similar-looking objects like Chihuahuas and blueberry muffins [00:02:13]. This early stage involved high non-determinism, akin to a “slot machine” experience, where consistency was a challenge [00:03:48]. Today, Google Lens offers advanced capabilities like visual shopping, translation, and even solving math homework [00:04:07]. This progress is attributed to consistent, step-by-step iteration over a decade, emphasizing the need for a continuous improvement process [00:04:43].
AI Agents for Customer Engagement
Leading businesses recognize the necessity of AI agents to represent their brand and assist customers [00:07:31]. Chubbies, an apparel company, partnered with Sierra to deploy an AI agent named Duncan Smothers [00:07:41]. Duncan is designed to be highly capable and engaging, handling various customer inquiries on the Chubbies website [00:07:55].
Examples of Duncan Smothers’ capabilities:
- Sizing and Fit: Empathetically assists customers with sizing questions and offers product recommendations [00:08:11].
- Inventory Tracking: Informs customers about product stock and helps them choose new items [00:08:27].
- Package Tracking and Refunds: Provides multiple tracking numbers for orders and can issue refunds, demonstrating an agent’s ability to take action beyond just answering questions [00:08:36].
These capabilities lead to improved customer satisfaction and efficiency, enabling businesses to help more customers more quickly [00:08:56].
The Agent Development Life Cycle (ADLC)
Sierra’s philosophy is that “every agent is a product” [00:09:06]. This means that building reliable AI-based customer support systems requires a fully featured developer platform and customer experience operations platform [00:09:11]. Sierra provides dedicated agent engineering and product management teams that work closely with customers like Chubbies [00:09:32].
The Agent Development Life Cycle (ADLC) is Sierra’s process for building and improving AI agents [00:12:12]. It borrows concepts from traditional software development but is adapted for the unique characteristics of large language models (LLMs) [00:11:26]. LLMs are non-deterministic, can be slow and expensive, but are flexible, creative, and capable of reasoning [00:11:51]. The ADLC aims to leverage LLM strengths while integrating traditional software where beneficial [00:12:02].
Key aspects of the ADLC include:
- Quality Assurance: Customers have access to Sierra’s Experience Manager to review conversations, monitor agent performance, and provide feedback on issues like incorrect information or missed opportunities [00:12:44].
- Continuous Improvement: Reported issues lead to test creation, which, once passing, allows for new releases. Agents continuously improve, growing from a handful of tests at launch to hundreds and thousands over time [00:13:15].
- AI-driven Acceleration: AI is increasingly integrated into each stage of the ADLC to speed up improvements, making the process more effective, especially for larger customers handling millions of requests [00:14:10]. Reasoning models, in particular, act as a “force multiplier” for each step, enhancing development, testing, and QA [00:15:02].
Voice AI Capabilities
In October, Sierra launched its voice capabilities, allowing customers like Sirius XM to pick up the phone immediately and answer customer calls [00:15:31]. The approach to voice AI at Sierra is similar to responsive web design [00:15:53]. The core AI agent platform and code remain the same, but they are responsive to different channels (e.g., chat, phone) and modalities [00:16:13]. While customization for phrasing and latency optimization is possible, the system largely “just works” out of the box [00:16:27].
Challenges and Potential of AI Voice Agents
Designing with AI requires empathy, putting oneself in the “shoes of the robot” [00:17:03]. Large language models, while powerful, remind us of human unpredictability, slowness, and occasional inaccuracies in tasks like math [00:16:51].
For voice-first AI and multimodal agents, a significant challenge is replicating the human experience of real-time conversation [00:17:28]. This includes handling latency, tone, and phrasing [00:10:19]. Sierra aims to build robust voice agents that provide rich experiences by giving LLMs inputs and experiences akin to human interaction [00:18:01]. The goal is not just to “wire everything together and hope it works” but to create sophisticated systems that can respond effectively on the spot, even with delays in transcribed input [00:17:30].