Tubegraph

Conversational AI applications in business

5 min read

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 to include a broader range of customer interactions and touch points [00:00:47]. By the end of the year, most of its interactions are expected to be over the phone [00:00:53].

Current applications for Sierra’s platform include:

The Evolution of AI and Iterative Improvement

Reflecting on the history of AI, significant advancements have been made in a relatively short period, often within the last decade [00:01:45]. As far back as 2016, AI was in its “caves,” with researchers at Google working on problems like distinguishing between Chihuahuas and blueberry muffins for what would become Google Lens [00:02:01].

Google Lens, initially capable of identifying plants, has evolved significantly [00:03:08]. Today, it allows users to:

This evolution highlights the importance of consistent, step-by-step iteration over time [00:04:43]. Just as software development relies on a life cycle for continuous improvement, so too must AI development [00:04:57].

Case Study: Chubbies and Duncan Smothers

Chubbies, a clothing company, partnered with Sierra to implement an AI agent named Duncan Smothers [00:07:37]. Chubbies recognized that in 2025, just as businesses needed websites and mobile apps previously, they now require an AI agent to represent their business and assist customers [00:07:25].

Duncan Smothers is designed to be both highly capable and engaging [00:07:50]. Examples of Duncan’s capabilities include:

  • Sizing and Fit: Providing empathetic assistance and product recommendations based on customer measurements [00:08:11].
  • Inventory Tracking: Informing customers about product availability and suggesting alternative items [00:08:27].
  • Package Tracking and Refunds: Providing multiple tracking numbers for orders and initiating refunds autonomously [00:08:37].

The results for Chubbies include helping more customers more quickly and with higher satisfaction [00:08:58]. This demonstrates the potential of AI in enterprise to take autonomous actions beyond merely answering questions [00:08:52].

The Agent Development Life Cycle (ADLC)

Sierra treats every AI agent as a product, requiring a fully featured developer and customer experience operations platform [00:09:06]. This approach involves dedicated agent engineering and product management teams working closely with customers [00:09:32].

The challenge with large language models (LLMs) is that they are non-deterministic, slow, expensive, and creative, unlike traditional software which is deterministic, fast, cheap, and rigid [00:11:34]. The Agent Development Life Cycle (ADLC) at Sierra is a methodology designed to leverage the strengths of LLMs while integrating traditional software where beneficial [00:12:02].

Quality Assurance in ADLC

A key component of the ADLC is Quality Assurance (QA) [00:12:43]. Customers have access to Sierra’s Experience Manager, allowing them to:

When an issue is reported (e.g., incorrect inventory information), it leads to an issue being filed, a test being created, and ultimately a new release once the test passes [00:13:15]. Over time, Sierra agents grow from having a handful of tests at launch to hundreds and thousands as they improve [00:13:27]. This feedback loop is essential for continuously improving AI conversational systems.

The ADLC also identifies opportunities for agents to go “above and beyond,” such as an agent having a budget to delight customers by arranging direct delivery of items not available online [00:13:42]. Initially, this process was manual, but with advancements in AI, AI is now being integrated into each stage of the life cycle to speed up improvements [00:14:00].

Voice AI and Responsive Agent Design

Sierra launched general availability for its Voice AI and its applications in enhancing customer experience capabilities in October [00:15:31]. For large customers like SiriusXM, this allows them to answer customer calls immediately [00:15:42].

Sierra’s approach to voice is similar to responsive web design:

  • Under the hood, it’s the same platform and agent code [00:16:16].
  • The agent is responsive to the channel (e.g., chat, phone) and modality (e.g., text, voice) through which a customer reaches out [00:16:20].
  • Customization is possible for phrasing or parallelizing requests to achieve lower latency [00:16:28].

Empathy in AI Design

A fascinating aspect of building with AI is that large language models can remind us of ourselves: they can be unpredictable, slow, and not great at math [00:16:46]. However, this also allows designers to approach AI development with a new form of empathy, putting themselves in the “shoes of the robot” or the “primordial soup of the Jell-O” to create better experiences [00:17:02].

When considering multimodal agents (like those combining text and voice), a key design question is how an agent would perform if it only received transcribed text with a few hundred milliseconds of delay and had to respond instantly [00:17:40]. Sierra aims to build robust systems that allow LLMs to have the same inputs and experiences as humans, leading to richer and more effective AI applications [00:18:04].

Graph View

Backlinks