From: aidotengineer
Sierra is a conversational AI platform designed for businesses [00:00:30]. While often associated with chat experiences and customer service, Sierra is expanding its reach, with most interactions expected to be over the phone by the end of the year [00:00:53]. The platform is used by customers for various aspects of the customer experience, including sales, subscription management, and product recommendations [00:01:02].
The Evolution of AI in Business
The journey of AI development has seen significant progress, even in recent years [00:01:45]. Reflecting on “ancient history” in AI often means looking back only a few years within the current decade [00:01:45]. However, the speaker recounts working at Google in 2016, a period he refers to as the “AI caves” [00:01:52]. At that time, efforts were focused on helping computers differentiate between objects, like Chihuahuas and blueberry muffins, which led to the first version of Google Lens [00:02:15].
Google Lens has evolved significantly from its infancy, where it was primarily good at identifying plants [00:03:06]. Today, it allows users to search, shop, translate non-Latin character sets, and even assist with math homework [00:04:07]. This progress is attributed to consistent, step-by-step iteration over a decade [00:04:43].
Challenges of Building with AI
Building with AI presents unique challenges compared to traditional software development [00:03:50]. While traditional software is deterministic, fast, cheap, rigid, and governed by logical if-statements, large language models (LLMs) are often non-deterministic, slow, expensive, yet flexible, creative, and capable of reasoning [00:11:42]. This non-determinism, or “building on top of a foundation of jello,” means that established software development practices cannot be directly applied [00:11:34].
The speaker notes that LLMs, by reminding us of ourselves in their unpredictability and occasional struggles with math, allow developers to be better designers through empathy with the system [00:16:48].
The Agent Development Life Cycle (ADLC)
To address the unique nature of LLMs and facilitate continuous improvement for AI agents, Sierra developed the Agent Development Life Cycle (ADLC) [00:12:12]. This process builds upon concepts from the traditional software development life cycle, while inventing new ones where necessary [00:11:26]. The ADLC emphasizes iterative refinement with customers in production to ensure productivity and robustness [00:13:35].
Key aspects of the ADLC include:
- Quality Assurance (QA): Customers have access to Sierra’s Experience Manager to review conversations, view performance reports in real-time, and provide feedback [00:12:45].
- Issue Resolution: If an agent makes an error, such as incorrect inventory information, the issue can be reported [00:13:05]. This leads to the filing of an issue, the creation of a test, and subsequent new releases once the test passes [00:13:15]. Over time, agents accumulate hundreds to thousands of tests, leading to continuous improvement [00:13:27].
- Opportunity Identification: Beyond fixing mistakes, the ADLC also identifies opportunities for the agent to go “above and beyond” for customers [00:13:37].
- AI-driven Improvement: While initially manual, Sierra is now able to integrate AI into each part of this life cycle to speed up improvements [00:14:10]. Reasoning models, for instance, act as a force multiplier, making AI more effective in development, testing, and QA [00:15:05].
Case Study: Chubbies and Duncan Smothers
In 2024, Chubbies, an apparel company known for its customer experience, partnered with Sierra to integrate AI into their business operations [00:07:02]. Recognizing that an AI agent is becoming as essential as a website or mobile app, Chubbies introduced “Duncan Smothers,” an AI agent on their website [00:07:25].
Duncan Smothers is designed to be highly capable and engaging [00:07:48]. Its capabilities include:
- Sizing and Fit Assistance: Empathetically helps customers with sizing questions and offers product recommendations [00:08:11].
- Inventory Tracking: Informs customers about product availability and helps them choose new items [00:08:27].
- Package Tracking and Refunds: Provides multiple tracking numbers for orders and can issue refunds [00:08:36].
These are examples of autonomous agents taking action, not just answering questions [00:08:49]. The results for Chubbies have shown improved customer assistance, faster resolution, and higher satisfaction [00:08:56].
Sierra treats every agent as a product, providing a fully featured developer platform and customer experience operations platform [00:09:04]. This partnership approach involves dedicated agent engineering and product management teams working closely with customers [00:09:32].
Integrating Voice AI in Business Operations
Sierra launched its general availability for voice capabilities in October, enabling large customers like SiriusXM to instantly answer phone calls from tens of millions of customers [00:15:31]. Sierra’s approach to voice AI is similar to responsive web design, where the same underlying platform and agent code can adapt to different channels and modalities, whether chat or voice [00:16:13]. This allows for channel-specific customization like phrasing and parallelized requests for lower latency, while generally working “out of the box” [00:16:27].
This highlights the concept of integrating AI agents into existing infrastructure, allowing businesses to maintain a unified AI system across various customer touchpoints.